Short-wave infrared hyperspectral imaging for organic media classification and mapping in works of art: the case of Willi Baumeister

Short-wave infrared hyperspectral imaging for organic media classification and mapping in works of art: the case of Willi Baumeister | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Short-wave infrared hyperspectral imaging for organic media classification and mapping in works of art: the case of Willi Baumeister Simon Mindermann, Katja Lorenz, Eva Mariasole Angelin, Marcello Picollo, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7347319/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Hyperspectral imaging (HSI) has become an increasingly valuable non-invasive technique in heritage science, particularly for the documentation and identification of pigments. However, the classification of binding media—especially organic, natural, semi-synthetic, and synthetic —remains less explored. This study presents a workflow for the mapping of binding media using short-wave infrared (SWIR) HSI (1000–2500 nm), supported by a reference spectral library of pure binding media films. To assess the reliability of the HSI system, spectral data were compared with measurements obtained using a benchtop spectrophotometer. A tailored data pre-processing and classification pipeline was developed and applied to three paintings by the German modernist Willi Baumeister. The results demonstrate successful spatial mapping of binding media, including alkyd resin/linseed oil, polyvinyl acetate and cellulose nitrate. This research underscores both the capabilities and limitations of SWIR-HSI as a non-invasive screening method to inform targeted sampling strategies. At the same time, it complements existing knowledge and contributes to a deeper understanding of the artist’s working methods. Figures Figure 1 Figure 2 Figure 3 Figure 4 1. Introduction The identification—as well as the spatial distribution—of binding media is essential for understanding the specific artistic techniques employed in the creation of a painting, and thus for gaining insight into an artist’s creative process. Micro-invasive analytical methods, such as Fourier-transform infrared spectroscopy (FT-IR) in transmission mode and gas chromatography-mass spectrometry (GC-MS), can yield detailed information on organic binders. However, due to the cultural and documentary significance of artworks, sampling is often severely limited or even prohibited. Additionally, the typically heterogeneous distribution of painting materials—both pigments and binding media—complicates the selection of representative sampling locations. Choosing appropriate sites for analysis is, therefore, both challenging and critical to ensuring meaningful interpretation of the analytical data. Spot analysis techniques, such as fiber-optics reflectance spectroscopy, FT-IR in reflection mode, and in-situ Raman spectroscopy, are now widely available for the non-invasive analysis of painting surfaces. However, the examination of large areas and the correlation of data across different regions remain complex and time-consuming. Hyperspectral imaging (HSI) has the potential to overcome these limitations. HSI is an advanced imaging technique that combines spatial and spectral information with high spectral resolution, yielding hundreds to thousands of spectral bands and providing near-continuous reflectance spectra. The most frequently used HSI systems in Heritage Science cover a spectral range between approximately 400–1000 nm, covering the visible and adjacent near-infrared range (VNIR). The use of systems operating in the short-wave infrared (SWIR) range (approximately 1000–2500 nm) is less common due to the requirement for more advanced—and therefore more costly—detector technology. HSI has been successfully applied to various analytical tasks in cultural heritage studies in the last decades [ 1 , 2 ]. The classification and mapping of different compounds across the surface, mainly inorganic and organic colorants using the VNIR spectral range [ 3 ], the visualization of hidden or faded features [ 4 – 6 ], or to aid the assessment of the state of conservation by, e.g., mapping the distribution of soluble salts on walls [ 7 ]. Nevertheless, the analysis and mapping of natural or synthetic organic binders by HSI is less established. Indeed, for non-invasive binding media analysis, the mid-infrared (MIR) range, between approximately 2500 and 25000 nm (2.5–25 µm or 4000 − 400 cm − 1 ), is generally regarded as the most suitable spectral region. Here, many relevant organic molecules exhibit highly characteristic absorptions [ 8 , 9 ]. However, imaging in the MIR suffers from several drawbacks, e.g., hardware requirements, limited commercial availability, and high equipment costs [ 10 ]. As a drawback, reflectance spectra recorded in this range exhibit several distortions, including Reststrahlen bands (due to high reflectivity at phonon resonances) and first derivative (caused by dispersion and surface reflection effects), which complicate direct interpretation [ 11 , 12 ]. Many artists' pigments absorb prominently in the MIR, potentially overlaying the binding media signals [ 13 – 16 ]. Previous studies have proven that apart from the MIR, SWIR-HSI has the potential to map organic compounds efficiently and reliably [ 17 – 20 ]. In the SWIR range, binding media are characterized by overtone and combination bands of fundamental vibrations typically observed in the mid-infrared (MIR) region. These features generally appear as broader bands. Absorptions related to C–H, N–H, and O–H functional groups can be used to classify many commonly employed organic binders, including natural drying oils, proteinaceous materials, and synthetic resins. Compared to the MIR region, relatively few inorganic materials used in artists’ paints—primarily certain sulfates, carbonates, and compounds containing structural water or hydroxyl groups—exhibit strong absorptions in the SWIR. This results in reduced spectral interference, thereby facilitating the identification of organic binding media. It is important to emphasize that most paintings are complex, multi-component, and multi-layered systems, which can produce equally complex reflectance spectra. Furthermore, variations in surface texture and measurement depth can result in shifts in band intensity and position, making spectral interpretation particularly challenging. As is common practice in the field, mock-ups are employed to simulate paint systems and generate reference datasets, thereby supporting the interpretation of reflectance spectra. These can vary from simple spreads of paint [ 21 ] to accurate reconstructions of artistic techniques, including multiple paint and preparatory layers as well as different supports [ 22 ]. Variations in admixture composition, paint film thickness, and surface texture can significantly affect the reference data collected in reflectance, thereby complicating the direct comparison of analytical results from real objects with reference databases. In the present study, the use of reference spectra from pure binders is tested, exploring the applicability of HSI spectral libraries across a broader range of case studies. A few libraries of SWIR-reflectance data of such pure materials exist, e.g., a database focused on the material palette of historic manuscripts published online by the University of Granada [ 23 ] or a database on modern pictorial materials in the 200–2500 nm range published online by the IFAC-CNR [ 24 ]. However, to the authors' best knowledge, comprehensive SWIR reference data for organic binding media, natural, semi-synthetic, and synthetic, used in easel paintings are not yet available. This study presents work toward the use of spectral libraries for binder classification and mapping in paintings using SWIR-HSI. To guarantee transferability, a generalizable approach using pure, naturally dried films of binding media as references is followed. An appropriate mock-up design is evaluated to enable optimal data acquisition using the HSI scanning system, taking into account the transparent, viscous, and film-forming properties of the materials under investigation. To employ the reference spectra for the study of artworks, a suitable pre-processing routine—deviating from typical workflows [ 25 ] —was developed to account for the specific challenges posed by painting surfaces. An opportunity for a case study application was provided by the ongoing research project “Innovation or Replacement? Art Technological Investigations of Willi Baumeister’s Paintings between 1930 and 1955” (2022–2027), which explores the material palette, painting techniques, and artistic process of the German modernist artist Willi Baumeister (1889–1955) ( https://www.abk-stuttgart.de/forschung/konservierung-und-restaurierung-von-kulturgut/innovation-oder-ersatz ), hereafter referred to as Baumeister Research Project . Archival studies confirm that the artist routinely experimented with traditional, as well as at that time, newly available, synthetic materials [ 26 , 27 ]. Several paintings by the artist are currently under analytical investigation. Among these is a collection of so-called “fragments” - paintings that were discarded by the artist, from the collection of the artist’s estate (Willi Baumeister Stiftung, Stuttgart). Following the investigation of the distribution of binding media by means of HSI used in three fragments by Baumeister (inventory numbers V_004, V_006, and V_013) is presented. 2. Materials and Methods 2.1 Reference Materials for Spectral Library The materials used as references are listed in Table 1 . The selection was based on archival research into Baumeister's technique and material palette [ 26 , 27 ]. Natural materials, as well as semi-synthetic and synthetic resins, have been selected as representatives of possible binding media used by the artist. Table 1 Reference materials and sources of supply. Product Type Original Product Name Supplier Acrylic Copolymer Paraloid™ B-72 Kremer Pigmente GmbH & Co. KG, (Aichstetten, Germany) Alkyd Resin Alkyd resin AM Boiled Linseed Oil Kremer Linseed Oil Varnish Cellulose Nitrate Zapon lacquer Gum Arabic Gum arabic Mastic Resin Mastix (Chios, Greece) Rabbit-Skin Glue Rabbit skin glue Ketone Resin Laropal®K 80 BASF SE (Ludwigshafen, Germany) Buttermilk Milbona pure buttermilk (max. 1% fat) Hohenloher Molkerei eG (Schwäbisch Hall, Germany) Polyvinyl Acetate Dispersion Mowilith D50 Celanese Emulsions GmbH (Frankfurt (Main), Germany) Vinyl-Acetat Homopolymer Raviflex BL 5S -A VINAVIL S.p.A. (Milan, Italy) 2.2 Mockup Preparation As in related studies, various support materials were tested for the application of binding media, including commonly used microscopy glass slides and polytetrafluoroethylene (PTFE) plates. [ 17 , 18 , 28 – 30 ]. Finally, matted aluminum sheet metal plates were selected for the study as the optimal option following initial testing based on its high values of diffuse, spectrally flat reflectance across the whole SWIR range. The pure binding media were applied using pipettes and left to dry for 14 days before scanning. Some binders, namely cellulose nitrate, gum arabic, and animal skin glue, did not adhere well to the support material, leading to delamination and flaking surfaces. However, sufficiently large areas with well-distributed binder films could be produced for all materials [ 31 ]. While the precise thickness was not measured in this study, the absence of sine-wave-like features and the high apparent signal-to-noise ratio in the reflectance spectra point towards a thickness above 15 µm [ 32 ]. 2.3 Analytical Methods For HSI, the push-broom scanner HySpex SWIR-384 was used. The scanner is built around a thermally stabilized MCT sensor that records in the SWIR with a spectral range between 930 and 2500 nm, split into 288 spectral channels. The resulting spectral resolution is approximately 5.45 nm. The sensor captures 384 spatial pixels per slit. A lens with an optimal working distance of 890 mm was used, resulting in an approximate field of view of 320 mm and a spatial resolution (pixel size) of 1.2 mm. The scanner is mounted on a computer-controlled, biaxial motorized rotation stage. Two "flicker-free" direct current, broadband, linear light sources are attached to the camera mounted on the tripod. According to the manufacturer, these halogen lamps emit continuously in the 400–2500 nm spectral range, guaranteeing optimal illumination during each scan and high reproducibility between different scans with the same system. For calibration, we used four Spectralon® calibration targets with certified reflectance values of 5%, 20%, 50%, and 90%, which were scanned alongside the mockups and case study paintings. To validate the HSI spectra, the reference materials were also analyzed in the same spectral range using a Perkin-Elmer Lambda 1050 spectrophotometer in reflectance mode (60 mm integrating sphere) with a 1 nm acquisition interval. The spectra were recorded without a specular component using a 99% Spectralon® standard as reflecting background behind microscopy glass slides on which the pure reference media films were applied. 2.4 HSI Data Acquisition and Processing The scans were acquired and calibrated using the applications HySpex Ground and HySpex Rad (HySpex, Oslo, Norway) and the Envi 6.0 software suite (NV5 Geospatial, Broomfield, CO, USA). Spectral data was pre-processed and processed using Python (3.11.5) and the Python packages Spectral Python ( https://www.spectralpython.net ), NumPy [ 33 ], and SciPy [ 34 ]. 2.4.1 Radiometric Calibration Before analysis, the data were calibrated to convert the raw pixel brightness values into quantitative reflectance values. This compensates for variations in signal strength due to the wavelength dependence of the illumination and the detector's sensitivity [ 25 ]. Raw data were first corrected for dark current noise using the HySpex Rad software. Empirical line correction (ELC) was used as implemented in Envi to remove the spectral component of the illumination system and ensure comparability and transferability of the reference spectra. The ELC method uses regions of interest (ROI) in the scan with known reflectance values to compute correction equations and calibrate radiance to reflectance values in the scan [ 35 ]. If used with reflectance targets, this can result in a more robust calibration than using, for example, flat field correction, since the targets often deviate from a fully flat reflectance, especially in the SWIR [ 25 ]. The Spectralon® calibration targets and corresponding reference spectra provided by the supplier were paired with the respective ROIs in the scan to compute the correction equations. 2.4.2 Data Processing Because of the increasing noise levels towards the edges of the sensor range, the spectral data were cut, resulting in a dataset of 266 bands, spanning the spectral range of approximately 1001–2445 nm. For the comparison of reference data and case study data, additional pre-processing was necessary because of differing noise and signal intensity levels. The Savitzky-Golay algorithm was applied for smoothing [ 36 ] with a moving window of 21 and a polynomial degree of 4 in the SciPy Signal Processing implementation. Scatter correction was used for improving the comparability between spectra from reference samples and paintings using standard normal variate (SNV): $$\:{X}_{SNV}=\frac{X-\stackrel{-}{X}}{{s}_{X}}$$ Where \(\:\stackrel{-}{X}\) represents the mean spectra and \(\:{s}_{X}\) the standard deviation of the spectra [ 37 ]. To create the reference dataset, the saturated pixels featuring reflectance values above 100 were masked; then, approximately 100 pixels showing intense signals were selected for each binder reference using the Envi ROI tool. From these selections, the mean spectra were computed and used as reference spectra. In this way, reflectance differences within the mockups could be mitigated and noise levels reduced. For data classification, spectral angle mapping (SAM) method was used. SAM is a widely applied, efficient supervised classification algorithm that determines class membership by calculating the angle between target and reference spectra, both regarded as n-dimensional vectors with n equaling the number of bands [ 1 , 2 , 38 ]. The classification threshold is defined by the spectral angle, with a small angle resulting in a more selective classification. SAM is robust to illumination differences, which was deemed essential for the comparison of reference spectra and case study data [ 39 ]. 3. Results and Discussion 3.1 Spectral Library In the following, a spectral library of 10 reference materials is presented (Figure 1). The plots show the designated main binders (Figure 1a-d), followed by additional natural (Figure 1e-h) and synthetic materials (Figure 1i-j). As validation, the average spectra obtained from HSI measurements are plotted with their corresponding control spectra recorded using the PE 1050 spectrophotometer (dashed lines). Considering absorbance band positions and signal intensities, no relevant differences between the HSI and the spectrophotometer spectra are apparent. The spectral resolution of the SWIR-HSI system appears well suited for classification purposes. The dominant spectral features of the organic reference materials are, in most cases, attributable to overtones and combination bands of the fundamental vibrations of C–H and O–H and, to a lesser extent, to the C=O and N–H groups. N–H absorptions are characteristic of proteinaceous materials, while O–H signals can originate from adsorbed water in hygroscopic organic binders or from hydroxyl groups chemically bound within their molecular structure. C–H signals are mainly caused by aliphatic or aromatic hydrocarbons, such as lipidic components, which often produce relatively sharp and distinctive spectral features. Linseed oil is one of the most important binding media in art history. Chemically, linseed oil is a triglyceride with a high amount of unsaturated fatty acids, enabling a drying process through polymerization. Alkyd resins are a group of materials that were developed in the early 20 th century. As polyester-based compounds modified with fatty acids, they share a similar structure to natural drying oils. Alkyd resin and linseed oil both show very prominent lipidic bands. The strongest signals are the doublets at approximately 1730-1760 nm (2ν (CH 2 )) and 2300-2350 nm (ν(CH 2 ) + δ(CH 2 )), followed by 3ν(C=O) at approximately 1930 nm [17]. The discrimination of linseed oil and alkyd resin in the SWIR was studied by [16]. They showed that the binders can be differentiated by small but sharp signals of aromatic components in the alkyd resin between 2100-2200 nm. Linseed oil shows a broader absorption attributed to ν(C=O) + ν(CH 2 ) in this area [29]. Polyvinyl acetate (PVAc) is a polymer synthesized by free-radical polymerization of vinyl acetate monomers. It is mainly used as an adhesive. In the SWIR, PVAc shows a recognizable triplet of CH 2 combinations and overtones at 2250, 2300, and 2360 nm and further sharp signals at approx. 2140 nm (ν(C=O) + ν(CH), 1910 nm (C=O), and 1720 nm (νCH ) [40, 41] . Solved PVAc and a PVAc dispersion material were tested and no distinct differences were found. Cellulose nitrate is a modified natural polymer derived from cellulose, partially esterified with nitric acid. Due to its widespread use as an industrial coating, it was readily available in Baumeister's time. Its spectral features in the SWIR primarily originate from its cellulose backbone. Bands at approximately 2250, 2300, and 2350 nm are assigned to (3δ(CH) + 3δ(CH 2 ) and (ν(CH) + δ(CH 2 ) / ν(CH 2 ) + δ(CH 2 ), the signal at 1710 nm as an overtone of C-H or CH 2 stretching modes. The sharp feature at 1900 nm is assigned to combinations of hydroxyl bands, as well as several not clearly resolved bands between ca. 2000 and 2100 nm [42]. Mastic gum is a natural triterpenoid resin, famously used as a varnish and transparent finishing layer. Bands in the SWIR are caused mainly by CH 2 modes, with the most intense signal being the combination of ν(CH 2 ) + δ(CH 2 ) at around 2300 nm and 2ν(CH 2 ) at 1700 nm, with 2ν(CH 2 ) as a shoulder at 1740 nm [28]. Animal glue is an adhesive material gained from refining animal connective tissue. In the SWIR, it is recognizable by its protein-based absorptions. These are overtones and combination bands of carbonyl, amino, and methylene groups, which include the 2ν(NH) at around 1500 nm, the νs(NH) + δ(NH) around 2040 nm, and 2(ν(CO) amide I + amide II) at approximately 2175 nm. Further bands are assigned to overtones and combinations of O-H (1400-1500 nm, 1920-1940 nm) and CH 2 (ca. 1730, 2280, and 2350 nm), the latter stemming from fatty components of the animal skin glue [28]. Gum arabic is a water-soluble tree exudate ( Acacia senegal and Acacia seyal) primarily composed of polysaccharides and is mainly applied in water and gouache paints. The spectra of gum arabic are characterized by three bands caused by overtones and combination bands of the various fundamental vibrational modes of its polysaccharide main components: δ(OH) + ν(CO) at 2100 nm, ν(CH) + δ(CH) at 2320 nm and ν(CH) + ν(CC) at 2490 nm. The other prominent bands around 1450 nm and 1940 nm are overtones and combinations of OH absorptions [20]. Historically, buttermilk was a byproduct of butter churning. Modern products are usually produced by controlled fermentation of milk through selected bacterial cultures to achieve a more reproducible product. The product, which Baumeister used as a matting agent, was likely of the historical variety [26]. While the main chemical components (water, proteins, lactose, and fat) contributing to the SWIR spectrum are assumed to be identical, modern products are expected to have a higher fat content. The most prominent bands are due to overtones and combination bands of O-H from water, with a broad band between 1400 and 1500 nm (2ν(OH) + δ(OH)) and a sharp feature at 1936 nm (ν(OH) + δ(OH)). [43] states that proteinaceouscomponents contribute the strongest at 2060 nm (amide A + amide I) and 2162 nm (amide B + amide II). These were also found to be the most reliable bands for the detection of buttermilk. The relatively sharp bands at 2303 and 2346 nm are assigned to symmetric ν(CH 2 ) + symmetric δ(CH 2 ) of the milk fat content [44]. Acrylic resins are synthetic polymers composed of polymerized esters of acrylic acid or methacrylic acid (e.g. polymethylmethacrylate). The main bands of acrylic resin are based on the combination bands of ν(CH 2 ) + δ(CH 2 ) at around 2260 nm and 2300 nm and the combination band of ν(CH) + ν(C=O) at 2130 nm [41]. Ketone resins are synthetic resins, typically derived from cyclohexanone or acetophenone. The spectrum of ketone resin shows its main absorptions between approximately 1700 and 1850 nm, assigned to combinations and 2 nd overtones of symmetric and asymmetric CH 2 and CH 3 stretching modes. The signal at 1930 nm is believed to be based on O-H and/or C=O bonds. Between 2000 and 2130 nm, the combinations of bending modes of CH 2 and O-H absorptions are visible. The most intense signals lie at approximately 2300 nm and 2345 nm. These are attributed to combinations of stretching and bending modes of CH 2 and CH 3 -based absorptions [45]. 3.2 Classification Case Study As a case study, three fragments of Baumeister (V_004, V_006, and V_013) were selected for the application of the spectral library for binder classification in paintings. Previous analyses using GC-MS have provided information on the binding media present, based on a limited number of samples taken from accessible and less aesthetically critical areas of the fragments. (unpublished analyses, done within the framework of Baumeister Research Project). According to these results, the main binders in these fragments are drying oil (all fragments), CN (V_006), and PVAc (V_013). Based on these previous analyses, reference spectra of linseed oil, alkyd resin, CN, and PVAc were selected as classification endmembers for the case study. Linseed oil and alkyd resin were treated as one material class, due to their high spectral similarity. As a first and more general approach, a unified classification threshold was used for all classes to minimize bias and the risk of overfitting. The thresholds were selected empirically for each case to result in relatively complete classifications of the paintings' surfaces, except for areas with too low signal intensities. As this approach proved insufficient to resolve multi-component systems, the individual thresholds were adjusted, as discussed in detail in the following sections.For evaluation of the classifications, ROIs of the paints in question were manually selected, and their mean spectra were computed. The spectral features are discussed and compared to the reference library. 3.2.1 Willi Baumeister Fragment “V_004 ” Based on optical observation, Fragment V_004 (35.5 cm x 46 cm, dated to ca. 1948, Willi Baumeister Stiftung, Stuttgart) is executed on fiberboard, followed by a red ground layer (or possibly remains of an older composition), partially visible through some lacunae (loss of material). The visible paint layers are composed of strongly structured, off-white-colored, and black areas. The white areas can be visually separated into a greyish-white layer, which is partially covered with a slightly yellow-toned layer. Figure 2 shows the results of the classification and the investigation of spectral features. For classification with SAM, a maximum spectral angle of 0.25 radians as a uniform threshold for all classes was selected. This resulted in the classification of the white-painted areas as mainly alkyd resin/linseed oil, as shown in Figure 2a. Some pixels, mainly bordering the black shapes, were classified as CN. While this could point towards a different paint system, the classification could also be affected by differing signal intensities due to surface features or varying degrees of opacity. The black paint remains unclassified due to low reflectance. Figure 2b shows the mean spectra of both white paints and the linseed oil reference. The spectra of the white areas show well-resolved features, most prominently between 1600-1800 nm and 2200-2400 nm. These bands are attributed to 2ν(CH₂) and to ν(CH₂) + δ(CH₂) combination bands. This interpretation is consistent with the reference spectra of linseed oil and alkyd resin. The ν(C=O) + ν(CH 2 ) centered at approx. 2140 nm is not visible in the spectra of V_004. While the general shape of the spectra appears very similar, the position of the absorption band at approximately 1900-1920 nm appears to be shifted to longer wavelengths, possibly due to humidity in the systems. The overall flatter appearance of the case study spectra is attributed to overlaying features of multiple secondary components in the paints. The spectrum of the greyish-white shows slightly higher overall absorption. The SWIR spectra do not allow a clear differentiation of the two paints. The result is in accordance with previous analyses of the white paints, which determined a drying oil as the main component of both paints (unpublished analyses, done within the framework of the Baumeister Research Project). 3.2.2 Willi Baumeister Fragment “V_006 ” The painting V_006 (25 x 34.9 cm, dated ca. 1945, Willi Baumeister Stiftung, Stuttgart) was executed on cardboard that was covered with a thin, white ground layer. Subsequently, Baumeister applied a light, pink-orange paint layer as background, followed by a composition of white, grey, and brown lines, framed by a light-grey, translucent paint along the borders of the painting. The composition was crossed out by the artist with a few quick brushstrokes with a thin, dark paint to mark it as discarded. Along the brushstrokes, this paint is very thin and almost transparent. More opaque areas are visible only at the centers and ends of the brushstrokes, where more of the paint accumulated. For classification, SAM was used initially with a maximum angle of 0.4 radians for all classes. As expected, the darker pixels in the center of the brushstrokes remained unclassified, as well as the darker borders. Using the initial unified threshold, almost all classified pixels were labeled as CN. This conflicts with the results of the previous analysis, in which CN was detected only in the paint used for crossing out, and a drying oil as main component of the underlying black, white, and pink-orange paints (unpublished analyses, done within the framework of the Baumeister Research Project). Due to analytical information available, the maximum angles were gradually modified, arriving at a threshold of 0.325 radians for CN and 0.475 radians for linseed oil/alkyd resin. This resulted in a classification of the pixels from thinner areas of the black paint as CN (red in Figure 3b). Pixels of the surrounding areas were classified as oil/alkyd resin (green in Figure 3b). This distribution was confirmed by additional non-invasive spot analysis using extended range FT-IR in reflectance mode [Angelin et al. 2025, manuscript submitted to the journal Heritage Science npj]. All spectra in Figure 3a show a pronounced contribution from the underlying cardboard support, with cellulose absorptions dominating the primary spectral components [46]. The differences between the black overpaint and the underlying pink-orange paint are subtle. For both paint layers, the most intense absorption features are the C–H combination bands between 2200 and 2400 nm, and an intense absorption band centered near 1925 nm, which is likely associated with O–H vibrations, originating from cellulose (cardboard support). For the pink-orange colored paint, the C–H overtone region around 1700–1800 nm is slightly more pronounced than for the black overpaint, while the latter shows more features between 2000 and 2200 nm. SAM classification successfully distinguishes the paint layers, despite the strong spectral contribution from the cellulose of the cardboard support. 3.2.3 Willi Baumeister Fragment “V_013” Fragment V_013 (54.7 cm x 33 cm, dated to 1955, Willi Baumeister Stiftung, Stuttgart, figure 4a) was executed on a hardboard support, covered with a thin white ground layer. This was followed by translucent black glaze. In the middle, a composition with red, green, and dark blue paints was executed. This was later overpainted with an opaque, black, oval shape. The underlying colors remain visible in lacunae. Due to the low reflectance, no valid spectra could be measured in the black central region. Figure 4 shows the classification result and evaluation spectra of the grey areas. The spectra show a pronounced slope towards shorter wavelengths, starting at approximately 1550 nm, probably due to noise and low signal intensities. For classification, the spectral region was therefore limited to wavelengths between 1600 and 2445 nm. A uniform maximum angle for SAM classification of 1.0 radians was selected, resulting in a relatively low number of unclassified pixels in the grey paint (Figure 4a). While the black oval shape in the center remains unclassified, the grey painted regions are predominantly labeled as PVAc, with a few pixels labeled as CN. A mean spectrum of 100 points of the grey paint in the corners closely matches the reference of PVAc, with the prominent absorption triplets centered on absorption bands at 2300 nm and 1720 nm (Figure 4b). This matches the previous analysis of the grey paint (unpublished analyses, done within the framework of the Baumeister Research Project). 4. Conclusions In this work, the potential of SWIR-HSI as a tool for the classification and mapping of major components of organic binding media in paintings was studied. A spectral library of binding media was presented, based on pure, naturally dried materials on matted aluminum sheet metal. This mockup design proved to be critical in the production of reference spectra with a high apparent signal-to-noise ratio. The comparison with spectra measured with a PE 1050 spectrophotometer demonstrated that the spectral resolution of the HSI system was sufficient for this task. A tailored pre-processing workflow was applied to use the references as endmembers in a classification case study. Using SAM, the main binding media component of three painting fragments by Willi Baumeister were classified. The results were evaluated via the discussion of paint spectra and compared with previous micro-invasive research. This proved the strong potential of a transferable spectral library for binder classification and mapping in a non-invasive analytical workflow. As is apparent from the case study application, HSI could be employed, e.g., as a screening method to guide possible sampling strategies even without further knowledge. The mappings can provide spatial information, complementing and extending spot-based analytical information. The study also stresses the evident limitations of HSI. Without additional information, setting individual classification thresholds for each material can be challenging, considering the complex material mixtures and distributions in paintings. In paintings with multiple binders in mixture or superposition, or with strong spectral contribution of other materials (cardboard support), this can lead to misinterpretations. For validation and optimization of the classification settings, additional analysis using reflectance FT-IR effectively improves results while keeping to a non-invasive approach. However, for a concise identification of multi-component paints, chromatographic methods like GC/MS remain necessary. Declarations Acknowledgements The present work had the financial support of the InsiTUMlab - Analytical infrastructure for non-destructive in-situ studies of Cultural Heritage, founded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Project No. 492717225, including a PhD position (Simon Mindermann). The Hyperspectral system has been purchased with funding from the Major Instrumentation Program of the DFG – Project n. 450461305. We thank Hadwig Goez of the Willi Baumeister Stiftung GmbH (Stuttgart, Germany) for the opportunity to study the Baumeister fragments. Author Contributions All the authors have contributed to the current work. S.M.: Formal Analysis, Investigation, Writing – Original Draft and Review & Editing, Visualization of supervised HSI data, data acquisition, planned and conducted all data processing and interpretation, and wrote the manuscript. K.L.: Methodology, Investigation, Writing – Review & Editing. E.A.: Methodology, Validation, Writing – Review & Editing. M.P.: Investigation and Review. C.C.: Validation, Writing – Original Draft and Review & Editing, Supervision. C.K.: Validation, Writing – Review & Editing, W.N.: Resources, Writing – Review & Editing. C.S.: Conceptualization, Methodology, Writing – Review & Editing, Supervision, Project Administration, Funding Acquisition. Corresponding author: Correspondence to [email protected] . Competing Interests All authors declare no competing financial or non-financial interests. Data Availability Statement The datasets generated and analysed during the current study will be provided by the corresponding author upon request. Data will be made available via the mediaTUM repository (https://mediatum.ub.tum.de). References Cucci, C., Delaney, J. K. & Picollo, M. Reflectance Hyperspectral Imaging for Investigation of Works of Art: Old Master Paintings and Illuminated Manuscripts. Accounts of chemical research 49 , 2070-9; 10.1021/acs.accounts.6b00048 (2016). Striova, J., Dal Fovo, A. & Fontana, R. Reflectance imaging spectroscopy in heritage science. La Rivista del Nuovo Cimento 43 , 515-66; 10.1007/s40766-020-00011-6 (2020). Delaney, J. K., Dooley, K. A., van Loon, A. & Vandivere, A. Mapping the pigment distribution of Vermeer’s Girl with a Pearl Earring. Heritage Science 202010.1186/s40494-019-0348-9. van der Snickt, G., et al. Exploring a Hidden Painting Below the Surface of René Magritte’s Le Portrait. Applied spectroscopy 70 , 57-67; 10.1177/0003702815617123 (2016). Cucci, C., Picollo, M., Stefani, L. & Jiménez, R. The man who became a Blue Glass. Reflectance hyperspectral imaging discloses a hidden Picasso painting of the Blue period. Journal of Cultural Heritage 62 , 484-92; 10.1016/j.culher.2023.07.003 (2023). Cucci, C., et al. Short-wave infrared reflectance hyperspectral imaging for painting investigations: A methodological study. Journal of the American Institute for Conservation 58 , 16-36; 10.1080/01971360.2018.1543102 (2019). Cucci, C., et al. Remote-sensing hyperspectral imaging for applications in archaeological areas: Non-invasive investigations on wall paintings and on mural inscriptions in the Pompeii site. Microchemical Journal 158 , 105082; 10.1016/j.microc.2020.105082 (2020). Nodari, L. & Ricciardi, P. Non-invasive identification of paint binders in illuminated manuscripts by ER-FTIR spectroscopy: a systematic study of the influence of different pigments on the binders’ characteristic spectral features. Herit Sci 201910.1186/s40494-019-0249-y. Rosi, F., et al. Noninvasive Analysis of Paintings by Mid‐infrared Hyperspectral Imaging. Angewandte Chemie 125 , 5366-9; 10.1002/ange.201209929 (2013). Rosi, F., et al. Broad Range Mid-IR Reflection Spectroscopy for Macroscale Standoff Hyperspectral Imaging of Paintings. ACS sensors 202510.1021/acssensors.5c00865. Angelin, E. M., et al. Application of Infrared Reflectance Spectroscopy on Plastics in Cultural Heritage Collections: A Comparative Assessment of Two Portable Mid-Fourier Transform Infrared Reflection Devices. Appl Spectrosc 75 , 818-33; 10.1177/0003702821998777 (2021). Fabbri, M., Picollo, M., Porcinai, S. & Bacci, M. Mid-Infrared Fiber-Optics Reflectance Spectroscopy: A Noninvasive Technique for Remote Analysis of Painted Layers. Part I: Technical Setup. Appl Spectrosc 55 , 420-7; 10.1366/0003702011952172 (2001). Sessa, C., Bagán, H. & García, J. F. Influence of composition and roughness on the pigment mapping of paintings using mid-infrared fiberoptics reflectance spectroscopy (mid-IR FORS) and multivariate calibration. Analytical and bioanalytical chemistry 406 , 6735-47; 10.1007/s00216-014-8091-2 (2014). Miliani, C., Rosi, F., Daveri, A. & Brunetti, B. G. Reflection infrared spectroscopy for the non-invasive in situ study of artists’ pigments. Appl. Phys. A 106 , 295-307; 10.1007/s00339-011-6708-2 (2012). Sessa, C., Jiménez de Garnica, R., Rosi, F., Fontana, R. & Garcia, J. F. A Study of Picasso’s Painting Materials and Techniques in Six of His Early Portraits. Journal of the American Institute for Conservation 55 , 198-216; 10.1080/01971360.2016.1235438 (2016). Gabrieli, F., Dooley, K. A., Zeibel, J. G., Howe, J. D. & Delaney, J. K. Standoff Mid-Infrared Emissive Imaging Spectroscopy for Identification and Mapping of Materials in Polychrome Objects. Angewandte Chemie (International ed. in English) 57 , 7341-5; 10.1002/anie.201710192 (2018). Dooley, K. A., et al. Standoff chemical imaging finds evidence for Jackson Pollock’s selective use of alkyd and oil binding media in a famous ‘drip’ painting. Analytical Methods 9 , 28-37; 10.1039/C6AY01795A (2017). Dooley, K. A., et al. Mapping of egg yolk and animal skin glue paint binders in Early Renaissance paintings using near infrared reflectance imaging spectroscopy. The Analyst 138 , 4838-48; 10.1039/c3an00926b (2013). Viguerie, L. de, et al. Mapping pigments and binders in 15 th century Gothic works of art using a combination of visible and near infrared hyperspectral imaging. Microchemical Journal 155 , 104674; 10.1016/j.microc.2020.104674 (2020). Ricciardi, P., et al. Near infrared reflectance imaging spectroscopy to map paint binders in situ on illuminated manuscripts. Angewandte Chemie (International ed. in English) 51 , 5607-10; 10.1002/anie.201200840 (2012). Capobianco, G., et al. Methodological approach for the automatic discrimination of pictorial materials using fused hyperspectral imaging data from the visible to mid-infrared range coupled with machine learning methods. Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy 304 , 123412; 10.1016/j.saa.2023.123412 (2024). Laureti, S., et al. Looking Through Paintings by Combining Hyper-Spectral Imaging and Pulse-Compression Thermography. Sensors (Basel, Switzerland) 201910.3390/s19194335. López-Baldomero, A. B., et al. Hyperspectral Database of Synthetic Historical Inks. archiving 21 , 11-6; 10.2352/issn.2168-3204.2024.21.1.3 (2024). Picollo M, Basilissi G, Cucci C, Stefani L, Tsukada M. Database of Reflectance Spectra (DRS). https://spectradb.ifac.cnr.it/drs/. Accessed 6 Aug 2025. Pillay, R., Hardeberg, J. Y. & George, S. Hyperspectral imaging of art: Acquisition and calibration workflows. Journal of the American Institute for Conservation 58 , 3-15; 10.1080/01971360.2018.1549919 (2019). Beerhorst & Ursula. Studien zur Maltechnik von Willi Baumeister. Zeitschrift für Kunsttechnologie und Konservierung 6 , 55-80 (1992). Palm, U. & Neugebauer, W. Innovation or replacement? Investigating Willi Baumeister’s painting materials and technique using art technological research. In: ICOM-CC, editor; 24-25 November; Paris; 2022. p. 117–125. Vagnini, M., et al. FT-NIR spectroscopy for non-invasive identification of natural polymers and resins in easel paintings. Analytical and bioanalytical chemistry 395 , 2107-18; 10.1007/s00216-009-3145-6 (2009). Wang, C. & Moretti, S. Fast and non-invasive identification of binding media in easel paintings by a portable hyperspectral sensor. Estudos de Conservação e Restauro , 36-46 (2018). Longoni, M., et al. FT-NIR Spectroscopy for the Non-Invasive Study of Binders and Multi-Layered Structures in Ancient Paintings: Artworks of the Lombard Renaissance as Case Studies. Sensors (Basel, Switzerland) 202210.3390/s22052052. Lorenz, K. Hyperspectral Imaging for Separation and Identification of Binders on Paintings. Stuttgart: Staatliche Akademie der Bildenden Künste Stuttgart; 2023. Koinig, G., et al. Influence of reflective materials, emitter intensity and foil thickness on the variability of near-infrared spectra of 2D plastic packaging materials. Waste Management 144 , 543-51; 10.1016/j.wasman.2021.12.019 (2022). Harris, C. R., et al. Array programming with NumPy. Nature 585 , 357-62; 10.1038/s41586-020-2649-2 (2020). Virtanen, P., et al. SciPy 1.0: fundamental algorithms for scientific computing in Python. Nature methods 17 , 261-72; 10.1038/s41592-019-0686-2 (2020). Smith, G. M. & Milton, E. J. The use of the empirical line method to calibrate remotely sensed data to reflectance. International Journal of Remote Sensing 20 , 2653-62; 10.1080/014311699211994 (1999). Savitzky, A. & Golay, M. J. E. Smoothing and Differentiation of Data by Simplified Least Squares Procedures. Analytical Chemistry 36 , 1627-39; 10.1021/ac60214a047 (1964). Rinnan, Å., van Berg, F. den & Engelsen, S. B. Review of the most common pre-processing techniques for near-infrared spectra. TrAC Trends in Analytical Chemistry 28 , 1201-22; 10.1016/j.trac.2009.07.007 (2009). Delaney, J. K., Dooley, K. A., Radpour, R. & Kakoulli, I. Macroscale multimodal imaging reveals ancient painting production technology and the vogue in Greco-Roman Egypt. Scientific reports 7 , 15509; 10.1038/s41598-017-15743-5 (2017). Kruse, F., et al. The Spectral Image Processing System (SIPS) ‐ Interactive Visualization and Analysis of Imaging Spectrometer Data. Remote Sensing of Environment , 145-63 (1993). Wheeler, O. H. Near Infrared Spectra Of Organic Compounds. Chemical Reviews 59 , 629-66; 10.1021/cr50028a004 (1959). Rosi, F., Daveri, A., Moretti, P., Brunetti, B. G. & Miliani, C. Interpretation of mid and near-infrared reflection properties of synthetic polymer paints for the non-invasive assessment of binding media in twentieth-century pictorial artworks. Microchemical Journal 124 , 898-908; 10.1016/j.microc.2015.08.019 (2016). Zapata, F., Ferreiro-González, M. & García-Ruiz, C. Interpreting the near infrared region of explosives. Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy 204 , 81-7 (2018). Manley, M. Near-infrared spectroscopy and hyperspectral imaging: non-destructive analysis of biological materials. Chemical Society reviews 43 , 8200-14; 10.1039/c4cs00062e (2014). Slobodan Šašić & Yukihiro Ozaki. Band Assignment of Near-Infrared Spectra of Milk by Use of Partial Least-Squares Regression. Appl. Spectrosc. 54 , 1327-38 (2000). Beć, K. B., Grabska, J. & Czarnecki, M. A. Spectra-structure correlations in NIR region: Spectroscopic and anharmonic DFT study of n-hexanol, cyclohexanol and phenol. Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy 197 , 176-84 (2018). Cichosz, S. & Masek, A. IR Study on Cellulose with the Varied Moisture Contents: Insight into the Supramolecular Structure. Materials (Basel, Switzerland) 202010.3390/ma13204573. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7347319","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":511027201,"identity":"cb9fe881-3ab7-4775-823d-d77aead53fdd","order_by":0,"name":"Simon Mindermann","email":"","orcid":"","institution":"Technical University of Munich (TUM)","correspondingAuthor":false,"prefix":"","firstName":"Simon","middleName":"","lastName":"Mindermann","suffix":""},{"id":511027202,"identity":"130ad7e2-b357-412c-8542-694d0f34819f","order_by":1,"name":"Katja Lorenz","email":"","orcid":"","institution":"Akademie der Bildenden Künste Stuttgart (ABK Stuttgart)","correspondingAuthor":false,"prefix":"","firstName":"Katja","middleName":"","lastName":"Lorenz","suffix":""},{"id":511027203,"identity":"f8bf7b74-af44-4a88-8d76-f310ea66bca8","order_by":2,"name":"Eva Mariasole Angelin","email":"","orcid":"","institution":"Technical University of Munich (TUM)","correspondingAuthor":false,"prefix":"","firstName":"Eva","middleName":"Mariasole","lastName":"Angelin","suffix":""},{"id":511027204,"identity":"ef4cea3d-5614-490f-a487-710166a071d3","order_by":3,"name":"Marcello Picollo","email":"","orcid":"","institution":"Istituto di Fisica Applicata “Nello Carrara” of the Consiglio Nazionale delle Ricerche (CNR- IFAC)","correspondingAuthor":false,"prefix":"","firstName":"Marcello","middleName":"","lastName":"Picollo","suffix":""},{"id":511027205,"identity":"50b164a1-3f67-4ed8-abc1-14b6b0ab95e7","order_by":4,"name":"Costanza Cucci","email":"","orcid":"","institution":"Istituto di Fisica Applicata “Nello Carrara” of the Consiglio Nazionale delle Ricerche (CNR- IFAC)","correspondingAuthor":false,"prefix":"","firstName":"Costanza","middleName":"","lastName":"Cucci","suffix":""},{"id":511027206,"identity":"784dd96f-3a85-47c5-b856-bb603c3342bd","order_by":5,"name":"Christoph Krekel","email":"","orcid":"","institution":"Akademie der Bildenden Künste Stuttgart (ABK Stuttgart)","correspondingAuthor":false,"prefix":"","firstName":"Christoph","middleName":"","lastName":"Krekel","suffix":""},{"id":511027207,"identity":"d04606e1-0490-45b9-b7c3-08e3c777dd91","order_by":6,"name":"Wibke Neugebauer","email":"","orcid":"","institution":"Akademie der Bildenden Künste Stuttgart (ABK Stuttgart)","correspondingAuthor":false,"prefix":"","firstName":"Wibke","middleName":"","lastName":"Neugebauer","suffix":""},{"id":511027208,"identity":"199200e1-719a-44c6-a914-7668c980d1bf","order_by":7,"name":"Clarimma Sessa","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA30lEQVRIiWNgGAWjYDACdiBOQHBtwKQEXi3MqFrSJIjTggQOE9bC38z87MPDnG3yDPyHD3/8UXO+zuDaAcYbH/BokTjMZjwjcdttwwaJtDRpnmO3JQxuJzBbzsCjxYCZwZgBqIWxQYLHjJmBDayFTZoHrxb2zyAt9g385z9//PHvHETLH7xaeMC2JDYw5DBI8LYdgGjB532JwzzFIC3JbRJpZtK8fcmSM28nNlv24NHC396+mfHnttu2/fyHH3/88c2On+928sEbP/BZAwNsCCZjAzEaRsEoGAWjYBTgAQAKrUmPAO1ytAAAAABJRU5ErkJggg==","orcid":"","institution":"Technical University of Munich (TUM)","correspondingAuthor":true,"prefix":"","firstName":"Clarimma","middleName":"","lastName":"Sessa","suffix":""}],"badges":[],"createdAt":"2025-08-11 14:23:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7347319/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7347319/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":91072366,"identity":"72e7578f-7f36-4d9c-9bab-5aa8995c0f9d","added_by":"auto","created_at":"2025-09-11 10:54:23","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":251972,"visible":true,"origin":"","legend":"\u003cp\u003eReference spectra acquired using the HSI system (black line) and the validation spectra acquired using the PE 1050 spectrophotometer (dashed line)\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7347319/v1/2972b0e637b137da9f9e829a.png"},{"id":91070803,"identity":"ddb7fbb8-bed8-44b1-af87-a0211aa9aedc","added_by":"auto","created_at":"2025-09-11 10:46:23","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":361073,"visible":true,"origin":"","legend":"\u003cp\u003ea: Willi Baumeister, Fragment V_004. Classification results on the SWIR-scan of V_004. The white areas are predominantly classified as oil/alkyd resin. Some regions are classified as CN. These are probably false classifications due to declining signal quality in the border regions between the black and white painted areas.\u003c/p\u003e\n\u003cp\u003eb: Reference spectrum of linseed oil (black line) and mean spectra of 100 pixels each of two off-white regions of V_004 (dashed and dotted lines)\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7347319/v1/14af097bd7b8d456c1d91a7c.png"},{"id":91070806,"identity":"c4defa38-1d68-43a7-b319-ceaf6ce83bff","added_by":"auto","created_at":"2025-09-11 10:46:23","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":308723,"visible":true,"origin":"","legend":"\u003cp\u003ea: Classification results on the SWIR-scan of case study painting V_006. While the black and darker areas of the painting remain unclassified due to low signal intensity, the regions around the black, crossed-out areas are predominantly classified as CN, the pink-orange colored areas are classified as oil/alkyd resin.\u003c/p\u003e\n\u003cp\u003eb: Mean spectra of 10 pixels each of the black paint used for crossing out (black line) and the pink-orange colored paint (dashed line), showing very similar spectral features, dominated by cellulose signals of the cardboard support.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7347319/v1/497c3e9e4fcf0201c3ffa1b8.png"},{"id":91072367,"identity":"7e110c49-01f9-4ed7-88a1-3c654f9663cd","added_by":"auto","created_at":"2025-09-11 10:54:23","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":353736,"visible":true,"origin":"","legend":"\u003cp\u003ea: Classification results on the SWIR-scan of case study painting V_013. The black and darker areas of the grey paint remain unclassified due to low intensities. The remaining grey paint is predominantly classified as PVAc.\u003c/p\u003e\n\u003cp\u003eb: PVAc reference spectrum and mean spectrum of 100 pixels of the gray paint of V_013 (dashed line)\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7347319/v1/6d3eae4d2dc967443c66caa6.png"},{"id":92365517,"identity":"52200cdc-a7d9-4a4d-b25e-9abb547cc147","added_by":"auto","created_at":"2025-09-28 21:23:25","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2028221,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7347319/v1/2719efdb-fcef-4b8a-b35a-89f28e0fae24.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Short-wave infrared hyperspectral imaging for organic media classification and mapping in works of art: the case of Willi Baumeister","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe identification\u0026mdash;as well as the spatial distribution\u0026mdash;of binding media is essential for understanding the specific artistic techniques employed in the creation of a painting, and thus for gaining insight into an artist\u0026rsquo;s creative process. Micro-invasive analytical methods, such as Fourier-transform infrared spectroscopy (FT-IR) in transmission mode and gas chromatography-mass spectrometry (GC-MS), can yield detailed information on organic binders. However, due to the cultural and documentary significance of artworks, sampling is often severely limited or even prohibited. Additionally, the typically heterogeneous distribution of painting materials\u0026mdash;both pigments and binding media\u0026mdash;complicates the selection of representative sampling locations. Choosing appropriate sites for analysis is, therefore, both challenging and critical to ensuring meaningful interpretation of the analytical data.\u003c/p\u003e\u003cp\u003eSpot analysis techniques, such as fiber-optics reflectance spectroscopy, FT-IR in reflection mode, and in-situ Raman spectroscopy, are now widely available for the non-invasive analysis of painting surfaces. However, the examination of large areas and the correlation of data across different regions remain complex and time-consuming.\u003c/p\u003e\u003cp\u003eHyperspectral imaging (HSI) has the potential to overcome these limitations. HSI is an advanced imaging technique that combines spatial and spectral information with high spectral resolution, yielding hundreds to thousands of spectral bands and providing near-continuous reflectance spectra. The most frequently used HSI systems in Heritage Science cover a spectral range between approximately 400\u0026ndash;1000 nm, covering the visible and adjacent near-infrared range (VNIR). The use of systems operating in the short-wave infrared (SWIR) range (approximately 1000\u0026ndash;2500 nm) is less common due to the requirement for more advanced\u0026mdash;and therefore more costly\u0026mdash;detector technology. HSI has been successfully applied to various analytical tasks in cultural heritage studies in the last decades [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The classification and mapping of different compounds across the surface, mainly inorganic and organic colorants using the VNIR spectral range [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], the visualization of hidden or faded features [\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], or to aid the assessment of the state of conservation by, e.g., mapping the distribution of soluble salts on walls [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Nevertheless, the analysis and mapping of natural or synthetic organic binders by HSI is less established. Indeed, for non-invasive binding media analysis, the mid-infrared (MIR) range, between approximately 2500 and 25000 nm (2.5\u0026ndash;25 \u0026micro;m or 4000\u0026thinsp;\u0026minus;\u0026thinsp;400 cm\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e), is generally regarded as the most suitable spectral region. Here, many relevant organic molecules exhibit highly characteristic absorptions [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. However, imaging in the MIR suffers from several drawbacks, e.g., hardware requirements, limited commercial availability, and high equipment costs [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. As a drawback, reflectance spectra recorded in this range exhibit several distortions, including Reststrahlen bands (due to high reflectivity at phonon resonances) and first derivative (caused by dispersion and surface reflection effects), which complicate direct interpretation [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. Many artists' pigments absorb prominently in the MIR, potentially overlaying the binding media signals [\u003cspan additionalcitationids=\"CR14 CR15\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e\u003cp\u003ePrevious studies have proven that apart from the MIR, SWIR-HSI has the potential to map organic compounds efficiently and reliably [\u003cspan additionalcitationids=\"CR18 CR19\" citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. In the SWIR range, binding media are characterized by overtone and combination bands of fundamental vibrations typically observed in the mid-infrared (MIR) region. These features generally appear as broader bands. Absorptions related to C\u0026ndash;H, N\u0026ndash;H, and O\u0026ndash;H functional groups can be used to classify many commonly employed organic binders, including natural drying oils, proteinaceous materials, and synthetic resins. Compared to the MIR region, relatively few inorganic materials used in artists\u0026rsquo; paints\u0026mdash;primarily certain sulfates, carbonates, and compounds containing structural water or hydroxyl groups\u0026mdash;exhibit strong absorptions in the SWIR. This results in reduced spectral interference, thereby facilitating the identification of organic binding media.\u003c/p\u003e\u003cp\u003eIt is important to emphasize that most paintings are complex, multi-component, and multi-layered systems, which can produce equally complex reflectance spectra. Furthermore, variations in surface texture and measurement depth can result in shifts in band intensity and position, making spectral interpretation particularly challenging. As is common practice in the field, mock-ups are employed to simulate paint systems and generate reference datasets, thereby supporting the interpretation of reflectance spectra. These can vary from simple spreads of paint [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] to accurate reconstructions of artistic techniques, including multiple paint and preparatory layers as well as different supports [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Variations in admixture composition, paint film thickness, and surface texture can significantly affect the reference data collected in reflectance, thereby complicating the direct comparison of analytical results from real objects with reference databases.\u003c/p\u003e\u003cp\u003eIn the present study, the use of reference spectra from pure binders is tested, exploring the applicability of HSI spectral libraries across a broader range of case studies. A few libraries of SWIR-reflectance data of such pure materials exist, e.g., a database focused on the material palette of historic manuscripts published online by the University of Granada [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e] or a database on modern pictorial materials in the 200\u0026ndash;2500 nm range published online by the IFAC-CNR [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. However, to the authors' best knowledge, comprehensive SWIR reference data for organic binding media, natural, semi-synthetic, and synthetic, used in easel paintings are not yet available.\u003c/p\u003e\u003cp\u003eThis study presents work toward the use of spectral libraries for binder classification and mapping in paintings using SWIR-HSI. To guarantee transferability, a generalizable approach using pure, naturally dried films of binding media as references is followed. An appropriate mock-up design is evaluated to enable optimal data acquisition using the HSI scanning system, taking into account the transparent, viscous, and film-forming properties of the materials under investigation. To employ the reference spectra for the study of artworks, a suitable pre-processing routine\u0026mdash;deviating from typical workflows [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] \u0026mdash;was developed to account for the specific challenges posed by painting surfaces. An opportunity for a case study application was provided by the ongoing research project \u003cem\u003e\u0026ldquo;Innovation or Replacement? Art Technological Investigations of Willi Baumeister\u0026rsquo;s Paintings between 1930 and 1955\u0026rdquo;\u003c/em\u003e (2022\u0026ndash;2027), which explores the material palette, painting techniques, and artistic process of the German modernist artist Willi Baumeister (1889\u0026ndash;1955) (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.abk-stuttgart.de/forschung/konservierung-und-restaurierung-von-kulturgut/innovation-oder-ersatz\u003c/span\u003e\u003cspan address=\"https://www.abk-stuttgart.de/forschung/konservierung-und-restaurierung-von-kulturgut/innovation-oder-ersatz\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), hereafter referred to as \u003cem\u003eBaumeister Research Project\u003c/em\u003e. Archival studies confirm that the artist routinely experimented with traditional, as well as at that time, newly available, synthetic materials [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Several paintings by the artist are currently under analytical investigation. Among these is a collection of so-called \u0026ldquo;fragments\u0026rdquo; - paintings that were discarded by the artist, from the collection of the artist\u0026rsquo;s estate (Willi Baumeister Stiftung, Stuttgart). Following the investigation of the distribution of binding media by means of HSI used in three fragments by Baumeister (inventory numbers V_004, V_006, and V_013) is presented.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cp\u003e\u003cstrong\u003e2.1 Reference Materials for Spectral Library\u003c/strong\u003eThe materials used as references are listed in Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e. The selection was based on archival research into Baumeister\u0026apos;s technique and material palette [\u003cspan class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e27\u003c/span\u003e]. Natural materials, as well as semi-synthetic and synthetic resins, have been selected as representatives of possible binding media used by the artist.\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eReference materials and sources of supply.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eProduct Type\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eOriginal Product Name\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSupplier\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAcrylic Copolymer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eParaloid\u0026trade; B-72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" rowspan=\"7\"\u003e\n \u003cp\u003eKremer Pigmente GmbH \u0026amp; Co. KG, (Aichstetten, Germany)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAlkyd Resin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAlkyd resin AM\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBoiled Linseed Oil\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eKremer Linseed Oil Varnish\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCellulose Nitrate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eZapon lacquer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGum Arabic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eGum arabic\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMastic Resin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMastix (Chios, Greece)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRabbit-Skin Glue\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRabbit skin glue\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eKetone Resin\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eLaropal\u0026reg;K 80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBASF SE (Ludwigshafen, Germany)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eButtermilk\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMilbona pure buttermilk (max. 1% fat)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHohenloher Molkerei eG (Schw\u0026auml;bisch Hall, Germany)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003ePolyvinyl Acetate Dispersion\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMowilith D50\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eCelanese Emulsions GmbH (Frankfurt (Main), Germany)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVinyl-Acetat Homopolymer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eRaviflex BL 5S -A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eVINAVIL S.p.A. (Milan, Italy)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2 Mockup Preparation\u003c/strong\u003eAs in related studies, various support materials were tested for the application of binding media, including commonly used microscopy glass slides and polytetrafluoroethylene (PTFE) plates. [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e28\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e30\u003c/span\u003e]. Finally, matted aluminum sheet metal plates were selected for the study as the optimal option following initial testing based on its high values of diffuse, spectrally flat reflectance across the whole SWIR range. The pure binding media were applied using pipettes and left to dry for 14 days before scanning. Some binders, namely cellulose nitrate, gum arabic, and animal skin glue, did not adhere well to the support material, leading to delamination and flaking surfaces. However, sufficiently large areas with well-distributed binder films could be produced for all materials [\u003cspan class=\"CitationRef\"\u003e31\u003c/span\u003e]. While the precise thickness was not measured in this study, the absence of sine-wave-like features and the high apparent signal-to-noise ratio in the reflectance spectra point towards a thickness above 15 \u0026micro;m [\u003cspan class=\"CitationRef\"\u003e32\u003c/span\u003e].\u003c/p\u003e\u003cspan\u003e\n \u003cp\u003e\u003cstrong\u003e2.3 Analytical Methods\u003c/strong\u003eFor HSI, the push-broom scanner HySpex SWIR-384 was used. The scanner is built around a thermally stabilized MCT sensor that records in the SWIR with a spectral range between 930 and 2500 nm, split into 288 spectral channels. The resulting spectral resolution is approximately 5.45 nm. The sensor captures 384 spatial pixels per slit. A lens with an optimal working distance of 890 mm was used, resulting in an approximate field of view of 320 mm and a spatial resolution (pixel size) of 1.2 mm. The scanner is mounted on a computer-controlled, biaxial motorized rotation stage. Two \u0026quot;flicker-free\u0026quot; direct current, broadband, linear light sources are attached to the camera mounted on the tripod. According to the manufacturer, these halogen lamps emit continuously in the 400\u0026ndash;2500 nm spectral range, guaranteeing optimal illumination during each scan and high reproducibility between different scans with the same system. For calibration, we used four Spectralon\u0026reg; calibration targets with certified reflectance values of 5%, 20%, 50%, and 90%, which were scanned alongside the mockups and case study paintings. To validate the HSI spectra, the reference materials were also analyzed in the same spectral range using a Perkin-Elmer Lambda 1050 spectrophotometer in reflectance mode (60 mm integrating sphere) with a 1 nm acquisition interval. The spectra were recorded without a specular component using a 99% Spectralon\u0026reg; standard as reflecting background behind microscopy glass slides on which the pure reference media films were applied.\u003c/p\u003e\n\u003c/span\u003e\u003cspan\u003e\n \u003cp\u003e\u003cstrong\u003e2.4 HSI Data Acquisition and Processing\u003c/strong\u003eThe scans were acquired and calibrated using the applications HySpex Ground and HySpex Rad (HySpex, Oslo, Norway) and the Envi 6.0 software suite (NV5 Geospatial, Broomfield, CO, USA). Spectral data was pre-processed and processed using Python (3.11.5) and the Python packages Spectral Python (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.spectralpython.net\u003c/span\u003e\u003c/span\u003e), NumPy [\u003cspan class=\"CitationRef\"\u003e33\u003c/span\u003e], and SciPy [\u003cspan class=\"CitationRef\"\u003e34\u003c/span\u003e].\u003c/p\u003e\n\u003c/span\u003e\u003cspan\u003e\n \u003cp\u003e\u003cstrong\u003e2.4.1 Radiometric Calibration\u003c/strong\u003eBefore analysis, the data were calibrated to convert the raw pixel brightness values into quantitative reflectance values. This compensates for variations in signal strength due to the wavelength dependence of the illumination and the detector\u0026apos;s sensitivity [\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e]. Raw data were first corrected for dark current noise using the HySpex Rad software. Empirical line correction (ELC) was used as implemented in Envi to remove the spectral component of the illumination system and ensure comparability and transferability of the reference spectra. The ELC method uses regions of interest (ROI) in the scan with known reflectance values to compute correction equations and calibrate radiance to reflectance values in the scan [\u003cspan class=\"CitationRef\"\u003e35\u003c/span\u003e]. If used with reflectance targets, this can result in a more robust calibration than using, for example, flat field correction, since the targets often deviate from a fully flat reflectance, especially in the SWIR [\u003cspan class=\"CitationRef\"\u003e25\u003c/span\u003e]. The Spectralon\u0026reg; calibration targets and corresponding reference spectra provided by the supplier were paired with the respective ROIs in the scan to compute the correction equations.\u003c/p\u003e\n\u003c/span\u003e\u003cspan\u003e\n \u003cp\u003e\u003cstrong\u003e2.4.2 Data Processing\u003c/strong\u003eBecause of the increasing noise levels towards the edges of the sensor range, the spectral data were cut, resulting in a dataset of 266 bands, spanning the spectral range of approximately 1001\u0026ndash;2445 nm. For the comparison of reference data and case study data, additional pre-processing was necessary because of differing noise and signal intensity levels. The Savitzky-Golay algorithm was applied for smoothing [\u003cspan class=\"CitationRef\"\u003e36\u003c/span\u003e] with a moving window of 21 and a polynomial degree of 4 in the SciPy Signal Processing implementation. Scatter correction was used for improving the comparability between spectra from reference samples and paintings using standard normal variate (SNV):\u003c/p\u003e\n\u003c/span\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e$$\\:{X}_{SNV}=\\frac{X-\\stackrel{-}{X}}{{s}_{X}}$$\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003eWhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\stackrel{-}{X}\\)\u003c/span\u003e\u003c/span\u003e represents the mean spectra and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{s}_{X}\\)\u003c/span\u003e\u003c/span\u003e the standard deviation of the spectra [\u003cspan class=\"CitationRef\"\u003e37\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003eTo create the reference dataset, the saturated pixels featuring reflectance values above 100 were masked; then, approximately 100 pixels showing intense signals were selected for each binder reference using the Envi ROI tool. From these selections, the mean spectra were computed and used as reference spectra. In this way, reflectance differences within the mockups could be mitigated and noise levels reduced. For data classification, spectral angle mapping (SAM) method was used. SAM is a widely applied, efficient supervised classification algorithm that determines class membership by calculating the angle between target and reference spectra, both regarded as n-dimensional vectors with n equaling the number of bands [\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e38\u003c/span\u003e]. The classification threshold is defined by the spectral angle, with a small angle resulting in a more selective classification. SAM is robust to illumination differences, which was deemed essential for the comparison of reference spectra and case study data [\u003cspan class=\"CitationRef\"\u003e39\u003c/span\u003e].\u003c/p\u003e"},{"header":"3. Results and Discussion","content":"\u003cp\u003e\u003cstrong\u003e3.1 Spectral Library\u003c/strong\u003e\u003cstrong\u003e\u0026nbsp;\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn the following, a spectral library of 10 reference materials is presented (Figure 1). The plots show the designated main binders (Figure 1a-d), followed by additional natural (Figure 1e-h) and synthetic materials (Figure 1i-j). As validation, the average spectra obtained from HSI measurements are plotted with their corresponding control spectra recorded using the PE 1050 spectrophotometer (dashed lines). Considering absorbance band positions and signal intensities, no relevant differences between the HSI and the spectrophotometer spectra are apparent. The spectral resolution of the SWIR-HSI system appears well suited for classification purposes.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe dominant spectral features of the organic reference materials are, in most cases, attributable to overtones and combination bands of the fundamental vibrations of C\u0026ndash;H and O\u0026ndash;H and, to a lesser extent, to the C=O and N\u0026ndash;H groups. N\u0026ndash;H absorptions are characteristic of proteinaceous materials, while O\u0026ndash;H signals can originate from adsorbed water in hygroscopic organic binders or from hydroxyl groups chemically bound within their molecular structure. C\u0026ndash;H signals are mainly caused by aliphatic or aromatic hydrocarbons, such as lipidic components, which often produce relatively sharp and distinctive spectral features.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLinseed oil\u003c/strong\u003e is one of the most important binding media in art history. Chemically, linseed oil is a triglyceride with a high amount of unsaturated fatty acids, enabling a drying process through polymerization. \u003cstrong\u003eAlkyd resins\u003c/strong\u003e are a group of materials that were developed in the early 20\u003csup\u003eth\u003c/sup\u003e century. As polyester-based compounds modified with fatty acids, they share a similar structure to natural drying oils. Alkyd resin and linseed oil both show very prominent lipidic bands. The strongest signals are the doublets at approximately 1730-1760 nm (2\u0026nu; (CH\u003csub\u003e2\u003c/sub\u003e)) and 2300-2350 nm (\u0026nu;(CH\u003csub\u003e2\u003c/sub\u003e) + \u0026delta;(CH\u003csub\u003e2\u003c/sub\u003e)), followed by 3\u0026nu;(C=O) at approximately 1930 nm [17]. The discrimination of linseed oil and alkyd resin in the SWIR was studied by [16]. They showed that the binders can be differentiated by small but sharp signals of aromatic components in the alkyd resin between 2100-2200 nm. Linseed oil shows a broader absorption attributed to \u0026nu;(C=O)\u0026nbsp;+\u0026nbsp;\u0026nu;(CH\u003csub\u003e2\u003c/sub\u003e) in this area \u0026nbsp;[29].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003ePolyvinyl acetate (PVAc)\u0026nbsp;\u003c/strong\u003eis a polymer synthesized by\u0026nbsp;free-radical polymerization of vinyl acetate monomers. It is mainly used as an adhesive. In the SWIR, PVAc shows a recognizable triplet of CH\u003csub\u003e2\u003c/sub\u003e combinations and overtones at 2250, 2300, and 2360 nm and further sharp signals at approx. 2140 nm (\u0026nu;(C=O) + \u0026nu;(CH), 1910 nm (C=O), and 1720 nm (\u0026nu;CH\u003cem\u003e)\u0026nbsp;\u003c/em\u003e[40, 41]\u003cem\u003e.\u003c/em\u003e Solved PVAc and a PVAc dispersion material were tested and no distinct differences were found.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCellulose nitrate\u003c/strong\u003e is a modified natural polymer derived from cellulose, partially esterified with nitric acid. Due to its widespread use as an industrial coating, it was readily available in Baumeister\u0026apos;s time. Its spectral features in the SWIR primarily originate from its cellulose backbone. Bands at approximately 2250, 2300, and 2350 nm are assigned to (3\u0026delta;(CH) + 3\u0026delta;(CH\u003csub\u003e2\u003c/sub\u003e) and (\u0026nu;(CH) + \u0026delta;(CH\u003csub\u003e2\u003c/sub\u003e) / \u0026nu;(CH\u003csub\u003e2\u003c/sub\u003e) + \u0026delta;(CH\u003csub\u003e2\u003c/sub\u003e), the signal at 1710 nm as an overtone of C-H or CH\u003csub\u003e2\u0026nbsp;\u003c/sub\u003estretching modes. The sharp feature at 1900\u0026nbsp;nm is assigned to combinations of hydroxyl bands, as well as several not clearly resolved bands between ca. 2000 and 2100 nm [42].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMastic gum\u003c/strong\u003e is a natural triterpenoid resin, famously used as a varnish and transparent finishing layer. Bands in the SWIR are caused mainly by CH\u003csub\u003e2\u003c/sub\u003e modes, with the most intense signal being the combination of \u0026nu;(CH\u003csub\u003e2\u003c/sub\u003e) + \u0026delta;(CH\u003csub\u003e2\u003c/sub\u003e) at around 2300 nm and 2\u0026nu;(CH\u003csub\u003e2\u003c/sub\u003e) at 1700 nm, with 2\u0026nu;(CH\u003csub\u003e2\u003c/sub\u003e) as a shoulder at 1740 nm [28].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAnimal glue\u003c/strong\u003e is an adhesive material gained from refining animal connective tissue. In the SWIR, it is recognizable by its protein-based absorptions. These are overtones and combination bands of carbonyl, amino, and methylene groups, which include the 2\u0026nu;(NH) at around 1500 nm, the \u0026nu;s(NH) + \u0026delta;(NH) around 2040 nm, and 2(\u0026nu;(CO) amide I + amide II) at approximately 2175 nm. Further bands are assigned to overtones and combinations of O-H (1400-1500 nm, 1920-1940 nm) and CH\u003csub\u003e2\u003c/sub\u003e (ca. 1730, 2280, and 2350 nm), the latter stemming from fatty components of the animal skin glue [28].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eGum arabic\u003c/strong\u003e is a water-soluble tree exudate (\u003cem\u003eAcacia senegal\u003c/em\u003e and \u003cem\u003eAcacia seyal)\u0026nbsp;\u003c/em\u003eprimarily composed of polysaccharides and is mainly applied in water and gouache paints. The spectra of gum arabic are characterized by three bands caused by overtones and combination bands of the various fundamental vibrational modes of its polysaccharide main components: \u0026delta;(OH) + \u0026nu;(CO) at 2100 nm, \u0026nu;(CH) + \u0026delta;(CH) at 2320 nm and \u0026nu;(CH) + \u0026nu;(CC) at 2490 nm. The other prominent bands around 1450 nm and 1940 nm are overtones and combinations of OH absorptions [20].\u003c/p\u003e\n\u003cp\u003eHistorically, \u003cstrong\u003ebuttermilk\u003c/strong\u003e was a byproduct of butter churning. Modern products are usually produced by controlled fermentation of milk through selected bacterial cultures to achieve a more reproducible product. The product, which Baumeister used as a matting agent, was likely of the historical variety [26]. While the main chemical components (water, proteins, lactose, and fat) contributing to the SWIR spectrum are assumed to be identical, modern products are expected to have a higher fat content. The most prominent bands are due to overtones and combination bands of O-H from water, with a broad band between 1400 and 1500 nm (2\u0026nu;(OH) + \u0026delta;(OH)) and a sharp feature at 1936 nm (\u0026nu;(OH) + \u0026delta;(OH)). [43] states that proteinaceouscomponents contribute the strongest at 2060 nm (amide A + amide I) and 2162 nm (amide B + amide II). These were also found to be the most reliable bands for the detection of buttermilk.\u0026nbsp;The relatively sharp bands at 2303 and 2346 nm are assigned to symmetric \u0026nu;(CH\u003csub\u003e2\u003c/sub\u003e) + symmetric \u0026delta;(CH\u003csub\u003e2\u003c/sub\u003e) of the milk fat content [44].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcrylic resins\u003c/strong\u003e are synthetic polymers composed of polymerized esters of acrylic acid or methacrylic acid (e.g. polymethylmethacrylate). The main bands of acrylic resin are based on the combination bands of \u0026nu;(CH\u003csub\u003e2\u003c/sub\u003e) + \u0026delta;(CH\u003csub\u003e2\u003c/sub\u003e) at around 2260 nm and 2300 nm and the combination band of \u0026nu;(CH) + \u0026nu;(C=O) at 2130 nm [41].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eKetone resins\u003c/strong\u003e are synthetic resins, typically derived from cyclohexanone or acetophenone. The spectrum of ketone resin shows its main absorptions between approximately 1700 and 1850 nm, assigned to combinations and 2\u003csup\u003end\u003c/sup\u003e overtones of symmetric and asymmetric CH\u003csub\u003e2\u003c/sub\u003e and CH\u003csub\u003e3\u003c/sub\u003e stretching modes. The signal at 1930 nm is believed to be based on O-H and/or C=O bonds. Between 2000 and 2130 nm, the combinations of bending modes of CH\u003csub\u003e2\u003c/sub\u003e and O-H absorptions are visible. The most intense signals lie at approximately 2300 nm and 2345 nm. These are attributed to combinations of stretching and bending modes of CH\u003csub\u003e2\u003c/sub\u003e and CH\u003csub\u003e3\u003c/sub\u003e-based absorptions [45].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 Classification Case Study\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u003cbr\u003e\u003c/strong\u003eAs a case study, three fragments of Baumeister (V_004, V_006, and V_013) were selected for the application of the spectral library for binder classification in paintings.\u0026nbsp;Previous analyses using GC-MS have provided information on the binding media present, based on a limited number of samples taken from accessible and less aesthetically critical areas of the fragments.\u0026nbsp;(unpublished analyses, done within the framework of Baumeister Research Project). According to these results, the main binders in these fragments are drying oil (all fragments), CN (V_006), and PVAc (V_013). Based on these previous analyses, reference spectra of linseed oil, alkyd resin, CN, and PVAc were selected as classification endmembers for the case study. Linseed oil and alkyd resin were treated as one material class, due to their high spectral similarity.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAs a first and more general approach, a unified classification threshold was used for all classes to minimize bias and the risk of overfitting. The thresholds were selected empirically for each case to result in relatively complete classifications of the paintings\u0026apos; surfaces, except for areas with too low signal intensities. As this approach proved insufficient to resolve multi-component systems, the individual thresholds were adjusted, as discussed in detail in the following sections.For evaluation of the classifications, ROIs of the paints in question were manually selected, and their mean spectra were computed. The spectral features are discussed and compared to the reference library. \u0026nbsp;\u003c/p\u003e\n\u003cp id=\"_Toc199254065\"\u003e\u003cstrong\u003e3.2.1 Willi Baumeister Fragment \u0026ldquo;V_004\u003c/strong\u003e\u003cstrong\u003e\u0026rdquo;\u0026nbsp;\u003cbr\u003e\u003c/strong\u003eBased on optical observation, Fragment V_004 (35.5 cm x 46 cm, dated to ca. 1948, Willi Baumeister Stiftung, Stuttgart) is executed on fiberboard, followed by a red ground layer (or possibly remains of an older composition), partially visible through some \u003cem\u003elacunae\u003c/em\u003e (loss of material). The visible paint layers are composed of strongly structured, off-white-colored, and black areas. The white areas can be visually separated into a greyish-white layer, which is partially covered with a slightly yellow-toned layer.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFigure 2\u0026nbsp;shows the results of the classification and the investigation of spectral features. For classification with SAM, a maximum spectral angle of 0.25 radians as a uniform threshold for all classes was selected. This resulted in the classification of the white-painted areas as mainly alkyd resin/linseed oil, as shown in Figure 2a.\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eSome pixels, mainly bordering the black shapes, were classified as CN. While this could point towards a different paint system, the classification could also be affected by differing signal intensities \u0026nbsp; due to surface features or varying degrees of opacity. The black paint remains unclassified due to low reflectance. Figure 2b\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eshows the mean spectra of both white paints and the linseed oil reference. The spectra of the white areas show well-resolved features, most prominently between 1600-1800 nm and 2200-2400 nm.\u0026nbsp;These bands are attributed to 2\u0026nu;(CH₂) and to\u0026nbsp;\u0026nu;(CH₂) +\u0026nbsp;\u0026delta;(CH₂) combination bands. This interpretation is consistent with the reference spectra of linseed oil and alkyd resin. The \u0026nu;(C=O)\u0026nbsp;+\u0026nbsp;\u0026nu;(CH\u003csub\u003e2\u003c/sub\u003e) centered at approx. 2140 nm is not visible in the spectra of V_004. While the general shape of the spectra appears very similar, the position of the absorption band at approximately 1900-1920 nm appears to be shifted to longer wavelengths, possibly due to humidity in the systems. The overall flatter appearance of the case study spectra is attributed to overlaying features of multiple secondary components in the paints. The spectrum of the greyish-white shows slightly higher overall absorption. The SWIR spectra do not allow a clear differentiation of the two paints. The result is in accordance with previous analyses of the white paints, which determined a drying oil as the main component of both paints (unpublished analyses, done within the framework of the Baumeister Research Project).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2.2 Willi Baumeister Fragment \u0026ldquo;V_006\u003c/strong\u003e\u003cstrong\u003e\u0026rdquo;\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe painting V_006 (25 x 34.9 cm, dated ca. 1945, Willi Baumeister Stiftung, Stuttgart) was executed on cardboard that was covered with a thin, white ground layer. Subsequently, Baumeister applied a light, pink-orange paint layer as background, followed by a composition of white, grey, and brown lines, framed by a light-grey, translucent paint along the borders of the painting. The composition was crossed out by the artist with a few quick brushstrokes with a thin, dark paint to mark it as discarded. Along the brushstrokes, this paint is very thin and almost transparent. More opaque areas are visible only at the centers and ends of the brushstrokes, where more of the paint accumulated. For classification, SAM was used initially with a maximum angle of 0.4 radians for all classes. As expected, the darker pixels in the center of the brushstrokes remained unclassified, as well as the darker borders. Using the initial unified threshold, almost all classified pixels were labeled as CN. This conflicts with the results of the previous analysis, in which CN was detected only in the paint used for crossing out, and a drying oil as main component of the underlying black, white, and pink-orange paints (unpublished analyses, done within the framework of the Baumeister Research Project).\u003c/p\u003e\n\u003cp\u003eDue to analytical information available, the maximum angles were gradually modified, arriving at a threshold of 0.325 radians for CN and 0.475 radians for linseed oil/alkyd resin. This resulted in a classification of the pixels from thinner areas of the black paint as CN (red in Figure 3b). Pixels of the surrounding areas were classified as oil/alkyd resin (green in Figure 3b). This distribution was confirmed by additional non-invasive spot analysis using extended range FT-IR in reflectance mode [Angelin et al. 2025, manuscript submitted to the journal Heritage Science npj]. All spectra in Figure 3a show a pronounced contribution from the underlying cardboard support, with cellulose absorptions dominating the primary spectral components [46]. The differences between the black overpaint and the underlying pink-orange paint are subtle. For both paint layers, the most intense absorption features are the C\u0026ndash;H combination bands between 2200 and 2400 nm, and an intense absorption band centered near 1925 nm, which is likely associated with O\u0026ndash;H vibrations, originating from cellulose (cardboard support). For the pink-orange colored paint, the C\u0026ndash;H overtone region around 1700\u0026ndash;1800 nm is slightly more pronounced than for the black overpaint, while the latter shows more features between 2000 and 2200 nm. SAM classification successfully distinguishes the paint layers, despite the strong spectral contribution from the cellulose of the cardboard support.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2.3 Willi Baumeister Fragment \u0026ldquo;V_013\u0026rdquo;\u0026nbsp;\u003cbr\u003e\u003c/strong\u003eFragment V_013 (54.7 cm x 33 cm, dated to 1955, Willi Baumeister Stiftung, Stuttgart, figure 4a) was executed on a hardboard support, covered with a thin white ground layer. This was followed by translucent black glaze. In the middle, a composition with red, green, and dark blue paints was executed. This was later overpainted with an opaque, black, oval shape. The underlying colors remain visible in lacunae. Due to the low reflectance, no valid spectra could be measured in the black central region. Figure 4 shows the classification result and evaluation spectra of the grey areas. The spectra show a pronounced slope towards shorter wavelengths, starting at approximately 1550 nm, probably due to noise and low signal intensities. For classification, the spectral region was therefore limited to wavelengths between 1600 and 2445 nm. A uniform maximum angle for SAM classification of 1.0 radians was selected, resulting in a relatively low number of unclassified pixels in the grey paint (Figure 4a). While the black oval shape in the center remains unclassified, the grey painted regions are predominantly labeled as PVAc, with a few pixels labeled as CN. A mean spectrum of 100 points of the grey paint in the corners closely matches the reference of PVAc, with the prominent absorption triplets centered on absorption bands at 2300 nm and 1720 nm (Figure 4b). This matches the previous analysis of the grey paint (unpublished analyses, done within the framework of the Baumeister Research Project).\u003c/p\u003e"},{"header":"4. Conclusions","content":"\u003cp\u003eIn this work, the potential of SWIR-HSI as a tool for the classification and mapping of major components of organic binding media in paintings was studied. A spectral library of binding media was presented, based on pure, naturally dried materials on matted aluminum sheet metal. This mockup design proved to be critical in the production of reference spectra with a high apparent signal-to-noise ratio. The comparison with spectra measured with a PE 1050 spectrophotometer demonstrated that the spectral resolution of the HSI system was sufficient for this task. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eA tailored pre-processing workflow was applied to use the references as endmembers in a classification case study. Using SAM, the main binding media component of three painting fragments by Willi Baumeister were classified. The results were evaluated via the discussion of paint spectra and compared with previous micro-invasive research. This proved the strong potential of a transferable spectral library for binder classification and mapping in a non-invasive analytical workflow. As is apparent from the case study application, HSI could be employed, e.g., as a screening method to guide possible sampling strategies even without further knowledge. The mappings can provide spatial information, complementing and extending spot-based analytical information.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe study also stresses the evident limitations of HSI. Without additional information, setting individual classification thresholds for each material can be challenging, considering the complex material mixtures and distributions in paintings. In paintings with multiple binders in mixture or superposition, or with strong spectral contribution of other materials (cardboard support), this can lead to misinterpretations. For validation and optimization of the classification settings, additional analysis using reflectance FT-IR effectively improves results while keeping to a non-invasive approach. However, for a concise identification of multi-component paints, chromatographic methods like GC/MS remain necessary.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements \u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe present work had the financial support of the InsiTUMlab - Analytical infrastructure for non-destructive in-situ studies of Cultural Heritage, founded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) \u0026ndash; Project No. 492717225, including a PhD position (Simon Mindermann). The Hyperspectral system has been purchased with funding from the Major Instrumentation Program of the DFG \u0026ndash; Project n. 450461305. We thank Hadwig Goez of the Willi Baumeister Stiftung GmbH (Stuttgart, Germany) for the opportunity to study the Baumeister fragments.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll the authors have contributed to the current work. S.M.: Formal Analysis, Investigation, Writing \u0026ndash; Original Draft and Review \u0026amp; Editing, Visualization of supervised HSI data, data acquisition, planned and conducted all data processing and interpretation, and wrote the manuscript. K.L.: Methodology, Investigation, Writing \u0026ndash; Review \u0026amp; Editing. E.A.: Methodology, Validation, Writing \u0026ndash; Review \u0026amp; Editing. M.P.: Investigation and Review. C.C.: Validation, Writing \u0026ndash; Original Draft and Review \u0026amp; Editing, Supervision. C.K.: Validation, Writing \u0026ndash; Review \u0026amp; Editing, W.N.: Resources, Writing \u0026ndash; Review \u0026amp; Editing. C.S.: Conceptualization, Methodology, Writing \u0026ndash; Review \u0026amp; Editing, Supervision, Project Administration, Funding Acquisition.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCorresponding author: Correspondence to [email protected].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors declare no competing financial or non-financial interests. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability Statement\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and analysed during the current study will be provided by the corresponding author upon request. Data will be made available via the mediaTUM repository (https://mediatum.ub.tum.de).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eCucci, C., Delaney, J. K. \u0026amp; Picollo, M. Reflectance Hyperspectral Imaging for Investigation of Works of Art: Old Master Paintings and Illuminated Manuscripts. \u003cem\u003eAccounts of chemical research\u003c/em\u003e \u003cstrong\u003e49\u003c/strong\u003e, 2070-9; 10.1021/acs.accounts.6b00048 (2016).\u003c/li\u003e\n\u003cli\u003eStriova, J., Dal Fovo, A. \u0026amp; Fontana, R. Reflectance imaging spectroscopy in heritage science. \u003cem\u003eLa Rivista del Nuovo Cimento\u003c/em\u003e \u003cstrong\u003e43\u003c/strong\u003e, 515-66; 10.1007/s40766-020-00011-6 (2020).\u003c/li\u003e\n\u003cli\u003eDelaney, J. K., Dooley, K. A., van Loon, A. \u0026amp; Vandivere, A. Mapping the pigment distribution of Vermeer\u0026rsquo;s Girl with a Pearl Earring. \u003cem\u003eHeritage Science\u003c/em\u003e 202010.1186/s40494-019-0348-9.\u003c/li\u003e\n\u003cli\u003evan der Snickt, G., et al. Exploring a Hidden Painting Below the Surface of Ren\u0026eacute; Magritte\u0026rsquo;s Le Portrait. \u003cem\u003eApplied spectroscopy\u003c/em\u003e \u003cstrong\u003e70\u003c/strong\u003e, 57-67; 10.1177/0003702815617123 (2016).\u003c/li\u003e\n\u003cli\u003eCucci, C., Picollo, M., Stefani, L. \u0026amp; Jim\u0026eacute;nez, R. The man who became a Blue Glass. Reflectance hyperspectral imaging discloses a hidden Picasso painting of the Blue period. \u003cem\u003eJournal of Cultural Heritage\u003c/em\u003e \u003cstrong\u003e62\u003c/strong\u003e, 484-92; 10.1016/j.culher.2023.07.003 (2023).\u003c/li\u003e\n\u003cli\u003eCucci, C., et al. Short-wave infrared reflectance hyperspectral imaging for painting investigations: A methodological study. \u003cem\u003eJournal of the American Institute for Conservation\u003c/em\u003e \u003cstrong\u003e58\u003c/strong\u003e, 16-36; 10.1080/01971360.2018.1543102 (2019).\u003c/li\u003e\n\u003cli\u003eCucci, C., et al. Remote-sensing hyperspectral imaging for applications in archaeological areas: Non-invasive investigations on wall paintings and on mural inscriptions in the Pompeii site. \u003cem\u003eMicrochemical Journal\u003c/em\u003e \u003cstrong\u003e158\u003c/strong\u003e, 105082; 10.1016/j.microc.2020.105082 (2020).\u003c/li\u003e\n\u003cli\u003eNodari, L. \u0026amp; Ricciardi, P. Non-invasive identification of paint binders in illuminated manuscripts by ER-FTIR spectroscopy: a systematic study of the influence of different pigments on the binders\u0026rsquo; characteristic spectral features. \u003cem\u003eHerit Sci\u003c/em\u003e 201910.1186/s40494-019-0249-y.\u003c/li\u003e\n\u003cli\u003eRosi, F., et al. Noninvasive Analysis of Paintings by Mid‐infrared Hyperspectral Imaging. \u003cem\u003eAngewandte Chemie\u003c/em\u003e \u003cstrong\u003e125\u003c/strong\u003e, 5366-9; 10.1002/ange.201209929 (2013).\u003c/li\u003e\n\u003cli\u003eRosi, F., et al. Broad Range Mid-IR Reflection Spectroscopy for Macroscale Standoff Hyperspectral Imaging of Paintings. \u003cem\u003eACS sensors\u003c/em\u003e 202510.1021/acssensors.5c00865.\u003c/li\u003e\n\u003cli\u003eAngelin, E. M., et al. Application of Infrared Reflectance Spectroscopy on Plastics in Cultural Heritage Collections: A Comparative Assessment of Two Portable Mid-Fourier Transform Infrared Reflection Devices. \u003cem\u003eAppl Spectrosc\u003c/em\u003e \u003cstrong\u003e75\u003c/strong\u003e, 818-33; 10.1177/0003702821998777 (2021).\u003c/li\u003e\n\u003cli\u003eFabbri, M., Picollo, M., Porcinai, S. \u0026amp; Bacci, M. Mid-Infrared Fiber-Optics Reflectance Spectroscopy: A Noninvasive Technique for Remote Analysis of Painted Layers. Part I: Technical Setup. \u003cem\u003eAppl Spectrosc\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 420-7; 10.1366/0003702011952172 (2001).\u003c/li\u003e\n\u003cli\u003eSessa, C., Bag\u0026aacute;n, H. \u0026amp; Garc\u0026iacute;a, J. F. Influence of composition and roughness on the pigment mapping of paintings using mid-infrared fiberoptics reflectance spectroscopy (mid-IR FORS) and multivariate calibration. \u003cem\u003eAnalytical and bioanalytical chemistry\u003c/em\u003e \u003cstrong\u003e406\u003c/strong\u003e, 6735-47; 10.1007/s00216-014-8091-2 (2014).\u003c/li\u003e\n\u003cli\u003eMiliani, C., Rosi, F., Daveri, A. \u0026amp; Brunetti, B. G. Reflection infrared spectroscopy for the non-invasive in situ study of artists\u0026rsquo; pigments. \u003cem\u003eAppl. Phys. A\u003c/em\u003e \u003cstrong\u003e106\u003c/strong\u003e, 295-307; 10.1007/s00339-011-6708-2 (2012).\u003c/li\u003e\n\u003cli\u003eSessa, C., Jim\u0026eacute;nez de Garnica, R., Rosi, F., Fontana, R. \u0026amp; Garcia, J. F. A Study of Picasso\u0026rsquo;s Painting Materials and Techniques in Six of His Early Portraits. \u003cem\u003eJournal of the American Institute for Conservation\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 198-216; 10.1080/01971360.2016.1235438 (2016).\u003c/li\u003e\n\u003cli\u003eGabrieli, F., Dooley, K. A., Zeibel, J. G., Howe, J. D. \u0026amp; Delaney, J. K. Standoff Mid-Infrared Emissive Imaging Spectroscopy for Identification and Mapping of Materials in Polychrome Objects. \u003cem\u003eAngewandte Chemie (International ed. in English)\u003c/em\u003e \u003cstrong\u003e57\u003c/strong\u003e, 7341-5; 10.1002/anie.201710192 (2018).\u003c/li\u003e\n\u003cli\u003eDooley, K. A., et al. Standoff chemical imaging finds evidence for Jackson Pollock\u0026rsquo;s selective use of alkyd and oil binding media in a famous \u0026lsquo;drip\u0026rsquo; painting. \u003cem\u003eAnalytical Methods\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 28-37; 10.1039/C6AY01795A (2017).\u003c/li\u003e\n\u003cli\u003eDooley, K. A., et al. Mapping of egg yolk and animal skin glue paint binders in Early Renaissance paintings using near infrared reflectance imaging spectroscopy. \u003cem\u003eThe Analyst\u003c/em\u003e \u003cstrong\u003e138\u003c/strong\u003e, 4838-48; 10.1039/c3an00926b (2013).\u003c/li\u003e\n\u003cli\u003eViguerie, L. de, et al. Mapping pigments and binders in 15\u003csup\u003eth\u003c/sup\u003e century Gothic works of art using a combination of visible and near infrared hyperspectral imaging. \u003cem\u003eMicrochemical Journal\u003c/em\u003e \u003cstrong\u003e155\u003c/strong\u003e, 104674; 10.1016/j.microc.2020.104674 (2020).\u003c/li\u003e\n\u003cli\u003eRicciardi, P., et al. Near infrared reflectance imaging spectroscopy to map paint binders in situ on illuminated manuscripts. \u003cem\u003eAngewandte Chemie (International ed. in English)\u003c/em\u003e \u003cstrong\u003e51\u003c/strong\u003e, 5607-10; 10.1002/anie.201200840 (2012).\u003c/li\u003e\n\u003cli\u003eCapobianco, G., et al. Methodological approach for the automatic discrimination of pictorial materials using fused hyperspectral imaging data from the visible to mid-infrared range coupled with machine learning methods. \u003cem\u003eSpectrochimica acta. Part A, Molecular and biomolecular spectroscopy\u003c/em\u003e \u003cstrong\u003e304\u003c/strong\u003e, 123412; 10.1016/j.saa.2023.123412 (2024).\u003c/li\u003e\n\u003cli\u003eLaureti, S., et al. Looking Through Paintings by Combining Hyper-Spectral Imaging and Pulse-Compression Thermography. \u003cem\u003eSensors (Basel, Switzerland)\u003c/em\u003e 201910.3390/s19194335.\u003c/li\u003e\n\u003cli\u003eL\u0026oacute;pez-Baldomero, A. B., et al. Hyperspectral Database of Synthetic Historical Inks. \u003cem\u003earchiving\u003c/em\u003e \u003cstrong\u003e21\u003c/strong\u003e, 11-6; 10.2352/issn.2168-3204.2024.21.1.3 (2024).\u003c/li\u003e\n\u003cli\u003ePicollo M, Basilissi G, Cucci C, Stefani L, Tsukada M. Database of Reflectance Spectra (DRS). https://spectradb.ifac.cnr.it/drs/. Accessed 6 Aug 2025.\u003c/li\u003e\n\u003cli\u003ePillay, R., Hardeberg, J. Y. \u0026amp; George, S. Hyperspectral imaging of art: Acquisition and calibration workflows. \u003cem\u003eJournal of the American Institute for Conservation\u003c/em\u003e \u003cstrong\u003e58\u003c/strong\u003e, 3-15; 10.1080/01971360.2018.1549919 (2019).\u003c/li\u003e\n\u003cli\u003eBeerhorst \u0026amp; Ursula. Studien zur Maltechnik von Willi Baumeister. \u003cem\u003eZeitschrift f\u0026uuml;r Kunsttechnologie und Konservierung\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 55-80 (1992).\u003c/li\u003e\n\u003cli\u003ePalm, U. \u0026amp; Neugebauer, W. Innovation or replacement? Investigating Willi Baumeister\u0026rsquo;s painting materials and technique using art technological research. In: ICOM-CC, editor; 24-25 November; Paris; 2022. p. 117\u0026ndash;125.\u003c/li\u003e\n\u003cli\u003eVagnini, M., et al. FT-NIR spectroscopy for non-invasive identification of natural polymers and resins in easel paintings. \u003cem\u003eAnalytical and bioanalytical chemistry\u003c/em\u003e \u003cstrong\u003e395\u003c/strong\u003e, 2107-18; 10.1007/s00216-009-3145-6 (2009).\u003c/li\u003e\n\u003cli\u003eWang, C. \u0026amp; Moretti, S. Fast and non-invasive identification of binding media in easel paintings by a portable hyperspectral sensor. \u003cem\u003eEstudos de Conserva\u0026ccedil;\u0026atilde;o e Restauro\u003c/em\u003e, 36-46 (2018).\u003c/li\u003e\n\u003cli\u003eLongoni, M., et al. FT-NIR Spectroscopy for the Non-Invasive Study of Binders and Multi-Layered Structures in Ancient Paintings: Artworks of the Lombard Renaissance as Case Studies. \u003cem\u003eSensors (Basel, Switzerland)\u003c/em\u003e 202210.3390/s22052052.\u003c/li\u003e\n\u003cli\u003eLorenz, K. Hyperspectral Imaging for Separation and Identification of Binders on Paintings. Stuttgart: Staatliche Akademie der Bildenden K\u0026uuml;nste Stuttgart; 2023.\u003c/li\u003e\n\u003cli\u003eKoinig, G., et al. Influence of reflective materials, emitter intensity and foil thickness on the variability of near-infrared spectra of 2D plastic packaging materials. \u003cem\u003eWaste Management\u003c/em\u003e \u003cstrong\u003e144\u003c/strong\u003e, 543-51; 10.1016/j.wasman.2021.12.019 (2022).\u003c/li\u003e\n\u003cli\u003eHarris, C. R., et al. Array programming with NumPy. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e585\u003c/strong\u003e, 357-62; 10.1038/s41586-020-2649-2 (2020).\u003c/li\u003e\n\u003cli\u003eVirtanen, P., et al. SciPy 1.0: fundamental algorithms for scientific computing in Python. \u003cem\u003eNature methods\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, 261-72; 10.1038/s41592-019-0686-2 (2020).\u003c/li\u003e\n\u003cli\u003eSmith, G. M. \u0026amp; Milton, E. J. The use of the empirical line method to calibrate remotely sensed data to reflectance. \u003cem\u003eInternational Journal of Remote Sensing\u003c/em\u003e \u003cstrong\u003e20\u003c/strong\u003e, 2653-62; 10.1080/014311699211994 (1999).\u003c/li\u003e\n\u003cli\u003eSavitzky, A. \u0026amp; Golay, M. J. E. Smoothing and Differentiation of Data by Simplified Least Squares Procedures. \u003cem\u003eAnalytical Chemistry\u003c/em\u003e \u003cstrong\u003e36\u003c/strong\u003e, 1627-39; 10.1021/ac60214a047 (1964).\u003c/li\u003e\n\u003cli\u003eRinnan, \u0026Aring;., van Berg, F. den \u0026amp; Engelsen, S. B. Review of the most common pre-processing techniques for near-infrared spectra. \u003cem\u003eTrAC Trends in Analytical Chemistry\u003c/em\u003e \u003cstrong\u003e28\u003c/strong\u003e, 1201-22; 10.1016/j.trac.2009.07.007 (2009).\u003c/li\u003e\n\u003cli\u003eDelaney, J. K., Dooley, K. A., Radpour, R. \u0026amp; Kakoulli, I. Macroscale multimodal imaging reveals ancient painting production technology and the vogue in Greco-Roman Egypt. \u003cem\u003eScientific reports\u003c/em\u003e \u003cstrong\u003e7\u003c/strong\u003e, 15509; 10.1038/s41598-017-15743-5 (2017).\u003c/li\u003e\n\u003cli\u003eKruse, F., et al. The Spectral Image Processing System (SIPS) ‐ Interactive Visualization and Analysis of Imaging Spectrometer Data. \u003cem\u003eRemote Sensing of Environment\u003c/em\u003e, 145-63 (1993).\u003c/li\u003e\n\u003cli\u003eWheeler, O. H. Near Infrared Spectra Of Organic Compounds. \u003cem\u003eChemical Reviews\u003c/em\u003e \u003cstrong\u003e59\u003c/strong\u003e, 629-66; 10.1021/cr50028a004 (1959).\u003c/li\u003e\n\u003cli\u003eRosi, F., Daveri, A., Moretti, P., Brunetti, B. G. \u0026amp; Miliani, C. Interpretation of mid and near-infrared reflection properties of synthetic polymer paints for the non-invasive assessment of binding media in twentieth-century pictorial artworks. \u003cem\u003eMicrochemical Journal\u003c/em\u003e \u003cstrong\u003e124\u003c/strong\u003e, 898-908; 10.1016/j.microc.2015.08.019 (2016).\u003c/li\u003e\n\u003cli\u003eZapata, F., Ferreiro-Gonz\u0026aacute;lez, M. \u0026amp; Garc\u0026iacute;a-Ruiz, C. Interpreting the near infrared region of explosives. \u003cem\u003eSpectrochimica acta. Part A, Molecular and biomolecular spectroscopy\u003c/em\u003e \u003cstrong\u003e204\u003c/strong\u003e, 81-7 (2018).\u003c/li\u003e\n\u003cli\u003eManley, M. Near-infrared spectroscopy and hyperspectral imaging: non-destructive analysis of biological materials. \u003cem\u003eChemical Society reviews\u003c/em\u003e \u003cstrong\u003e43\u003c/strong\u003e, 8200-14; 10.1039/c4cs00062e (2014).\u003c/li\u003e\n\u003cli\u003eSlobodan \u0026Scaron;a\u0026scaron;ić \u0026amp; Yukihiro Ozaki. Band Assignment of Near-Infrared Spectra of Milk by Use of Partial Least-Squares Regression. \u003cem\u003eAppl. Spectrosc.\u003c/em\u003e \u003cstrong\u003e54\u003c/strong\u003e, 1327-38 (2000).\u003c/li\u003e\n\u003cli\u003eBeć, K. B., Grabska, J. \u0026amp; Czarnecki, M. A. Spectra-structure correlations in NIR region: Spectroscopic and anharmonic DFT study of n-hexanol, cyclohexanol and phenol. \u003cem\u003eSpectrochimica acta. Part A, Molecular and biomolecular spectroscopy\u003c/em\u003e \u003cstrong\u003e197\u003c/strong\u003e, 176-84 (2018).\u003c/li\u003e\n\u003cli\u003eCichosz, S. \u0026amp; Masek, A. IR Study on Cellulose with the Varied Moisture Contents: Insight into the Supramolecular Structure. \u003cem\u003eMaterials (Basel, Switzerland)\u003c/em\u003e 202010.3390/ma13204573.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7347319/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7347319/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eHyperspectral imaging (HSI) has become an increasingly valuable non-invasive technique in heritage science, particularly for the documentation and identification of pigments. However, the classification of binding media\u0026mdash;especially organic, natural, semi-synthetic, and synthetic \u0026mdash;remains less explored. This study presents a workflow for the mapping of binding media using short-wave infrared (SWIR) HSI (1000\u0026ndash;2500 nm), supported by a reference spectral library of pure binding media films. To assess the reliability of the HSI system, spectral data were compared with measurements obtained using a benchtop spectrophotometer. A tailored data pre-processing and classification pipeline was developed and applied to three paintings by the German modernist Willi Baumeister. The results demonstrate successful spatial mapping of binding media, including alkyd resin/linseed oil, polyvinyl acetate and cellulose nitrate. This research underscores both the capabilities and limitations of SWIR-HSI as a non-invasive screening method to inform targeted sampling strategies. At the same time, it complements existing knowledge and contributes to a deeper understanding of the artist\u0026rsquo;s working methods.\u003c/p\u003e","manuscriptTitle":"Short-wave infrared hyperspectral imaging for organic media classification and mapping in works of art: the case of Willi Baumeister","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-11 10:46:18","doi":"10.21203/rs.3.rs-7347319/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a03d0bf9-0710-4a79-883e-66f0a5d01f4d","owner":[],"postedDate":"September 11th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-09-28T21:23:10+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-11 10:46:18","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7347319","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7347319","identity":"rs-7347319","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}