A low-cost machine-vision inspection system for RSS1--RSS5 rubber sheet grading using standardised illumination and HSV representation

A low-cost machine-vision inspection system for RSS1--RSS5 rubber sheet grading using standardised illumination and HSV representation | Authorea try { document.documentElement.classList.add('js'); } catch (e) { } var _gaq = _gaq || []; _gaq.push(['_setAccount', 'G-8VDV14Y67G']); _gaq.push(['_trackPageview']); (function() { var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true; ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js'; var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s); })(); Skip to main content Preprints Collections Wiley Open Research IET Open Research Ecological Society of Japan All Collections About About Authorea FAQs Contact Us Quick Search anywhere Search for preprint articles, keywords, etc. Search Search ADVANCED SEARCH SCROLL This is a preprint and has not been peer reviewed. Data may be preliminary. 31 March 2026 V1 Latest version Share on A low-cost machine-vision inspection system for RSS1--RSS5 rubber sheet grading using standardised illumination and HSV representation Authors : Thisakya Ransarani , K.T. Lakshitha Priyashan , R.A.N.C. Wijesinghe , I.P.T.S Wickramasooriya , Ruchire Eranga Wijesinghe , Ilya Kavalchuk 0000-0003-4560-4281 [email protected] , and Bhagya Silva Authors Info & Affiliations https://doi.org/10.22541/au.177497136.63415473/v1 150 views 73 downloads Contents Abstract Information & Authors Metrics & Citations View Options References Figures Tables Media Share Abstract Manual grading of Ribbed Smoked Sheets (RSS) is labour-intensive and experience-dependent, which can lead to inconsistent grade assignment and reduced transparency in quality evaluation. This paper presents a low-cost machine-vision inspection system for full-spectrum RSS grading (RSS1–RSS5) that integrates standardised image acquisition and HSV-based colour representation with a lightweight convolutional neural network (CNN) classifier. Images were captured using a custom-built acquisition setup with fixed illumination and camera geometry to improve repeatability across samples. To enhance grading-relevant cue representation, RGB images were transformed into the Hue–Saturation–Value (HSV) colour space, separating chromatic information from brightness prior to classification. Using a balanced dataset of expert-labelled images, the HSV-based system achieved 91.87% classification accuracy, outperforming an RGB baseline (87.81%). Confusion-matrix analysis indicated strong performance across all five grades, with remaining errors concentrated between visually similar adjacent grades. The results demonstrate the feasibility of combining low-cost standardised illumination with HSV-based colour processing and lightweight learning for practical RSS inspection and grading. 1. Introduction The natural rubber (NR) industry is a major component of the global agricultural economy that supplies a strategic raw material for manufacturing and international supply chains [1], [2], [3], [4]. In recent years, global NR output has been over 10 million tonnes, produced largely in tropical regions [5], with harvested area reported at roughly 12 million hectares worldwide [6]. Production is concentrated in Southeast Asia, which accounts for around 80% of global output, led by countries such as Thailand, Indonesia, Malaysia, and Sri Lanka [7], [8]. NR is used across a wide range of industries, including tyre manufacturing, automotive components, and latex-based medical products, supporting extensive downstream value chains [2], [5], [9]. Among the traded forms of raw NR, Ribbed Smoked Sheets (RSS) are widely utilised due to their relatively consistent quality and suitability for downstream industrial processing [10]. RSS quality is classified into five grades (RSS1–RSS5) based on visual appearance, purity, and processing integrity [11], [12]. Grading criteria are defined in recognised standards and guidelines, including the international standards for quality and packing of natural rubber grades (“Green Book”) and SNI 06-001-1987 [13]. In practice, however, RSS grading is still performed predominantly through manual visual inspection and assessor judgement, making it labour-intensive and prone to variability across graders, facilities, and operating conditions [12], [13], [14]. This leads to misclassification, reduced transparency in pricing, and variability in material quality for downstream industrial users [15]. These challenges disproportionately affect small-scale producers, who constitute a large share of RSS suppliers. Subjective grading, improper sorting practices, and intermediary exploitation reduce income and discourage the production of higher-grade RSS sheets [1], [12], [15]. Quality degradation in RSS can arise from processing-related factors such as improper drying, suboptimal smoking conditions, and uncontrolled production environments [12], [16], [17], which may introduce defects that are difficult to interpret consistently near grade boundaries. In practice, a substantial proportion of sheets may fall into lower grades; for example, one study reports that approximately 70% of graded sheets were classified as low quality [18]. In addition to unintentional variability, deliberate malpractice has been reported, where ungraded sheets are mixed with graded batches and sold as higher-grade RSS to obtain higher prices [11]. Collectively, these factors can undermine trust in grade-based valuation and contribute to financial losses across the supply chain. Several studies have proposed improvements to RSS production, including increasing line efficiency [19], reducing processing time [16], and enhancing or automating machinery of the workflow [20], [21]. Quality enhancement has also been investigated through process-level interventions such as improving smoking conditions [12], [16] and evaluating physicochemical properties under different drying processes and conditions [17]. Efforts to reduce costs and resource requirements associated with RSS production have also been reported [16], [22]. However, despite these advances, ensuring consistent and high-quality RSS in practice remains challenging because assessment and grade assignment still largely depend on subjective visual inspection. Simultaneously, the availability of skilled grading labour is declining, and many producers lack access to reliable assessment methods to support consistent compliance with grading standards [12]. RSS grading standards emphasise factors such as cleanliness and dryness and describe surface defects including air bubbles, rust, mould, holes, and surface markings [13], [23], [24], underscoring the need for objective methods that can evaluate these cues consistently. Motivated by these limitations, recent work has explored automated image-based assessment of RSS quality. However, the literature remains limited, and many reported approaches focus on detecting specific defect types or classify only a subset of grades rather than performing full RSS grade classification. Fibriani et al. [24] proposed a colour-based image processing approach in which captured images were converted to grayscale and binarised to quantify clean surface regions, reporting 88% overall accuracy and 100% accuracy for RSS3. Nevertheless, performance for RSS2 deteriorated under uneven illumination, and the method was limited to RSS1–RSS3. Rahmat et al. [13] introduced a Learning Vector Quantization (LVQ) approach involving image enhancement and feature extraction, achieving 89% accuracy but only for two grades (RSS1 and RSS3). Pornpanomchai and Chantharangsikul [11] proposed an RSS grading system combining image processing and machine learning. Images were captured under controlled conditions and pre-processed using grayscale conversion, noise removal, and RGB (Red, Green, Blue) to L*a*b* colour space transformation, and five-grade classification using k-means clustering achieved 80.9% precision on untrained samples. However, reported performance was constrained by reliance on controlled acquisition conditions and a limited dataset, and intermediate grades, particularly RSS2 remained challenging due to dependence on colour-based features. Overall, the literature indicates a clear gap, the absence of a standardised and scalable machine-vision inspection pipeline that combines repeatable acquisition and colour-space processing to support accurate RSS1–RSS5 grading under practical operating conditions. To address this gap, this study proposes a lightweight CNN-based grading system for multi-class classification of RSS into all five standard grades, combined with a low-cost image acquisition setup. Unlike earlier RSS assessment approaches that focus on selected defect categories or limited grade subsets, the proposed framework implements an end-to-end five-grade classification pipeline trained on a newly acquired, domain-specific dataset collected using a standardised imaging protocol. The key contributions of this study are: • An end-to-end, multi-class grading pipeline that predicts all five standard RSS grades. • Integration of HSV (Hue, Saturation, Value)-based colour conversion to enhance separability of grading-relevant visual cues and reduce sensitivity to illumination-related variation. • A low-cost, standardised image acquisition unit that enables consistent capture conditions while preserving grading-relevant differences in texture and surface appearance. • A newly collected RSS image dataset acquired using a standardised capture protocol to support training and evaluation under production conditions. The remainder of this paper is organised as follows. Section 2 describes the proposed methodology, including the image acquisition setup, HSV-based pre-processing, and the CNN architecture. Section 3 presents the experimental results and performance evaluation, and Section 4 discusses the findings and practical implications. Section 5 concludes the paper and outlines directions for future work on RSS grading. 2. Methodology The proposed methodology implements a low-cost machine-vision inspection pipeline for RSS grading by integrating standardised image acquisition, colour-space processing, and lightweight classification. The acquisition setup was designed with fixed illumination and camera geometry to improve repeatability and reduce nuisance variation in colour and appearance, which is critical for visual grading tasks. Within this pipeline, HSV-based pre-processing was selected because RSS grade assignment depends strongly on colour uniformity and the visual appearance of defects described in grading standards and guidance. In RGB, brightness and chromaticity are not separated [25], which can reduce the consistency of colour-dependent cues. By contrast, HSV decouples them aligning more closely with human colour perception by using hue (dominant colour), saturation (colour purity), and value (brightness), thereby providing a more intuitive and robust representation of colour information [25], [26], [27]. This design choice is supported by prior comparative studies reporting improved performance with HSV relative to RGB in colour-feature analysis and visual classification tasks [28], [29]. Consequently, HSV has been widely adopted in related computer-vision pipelines [26], [30], [31]. Accordingly, RGB images captured using the proposed acquisition setup were transformed into HSV to better isolate variations associated with discoloration, surface contamination, and material consistency. The pre-processed images are then used as input to the classification model for RSS grade prediction. The model is trained to learn grading-relevant surface characteristics such as tar spots, air bubbles, flat roller prints, and discoloration, that collectively influence manual inspection outcomes. A high-level overview of the proposed workflow is shown in Figure 01. Figure 01: Overview of the proposed RSS grading framework 2.1 Dataset preparation A total of 500 pre-graded images of RSS were collected from a distribution centre that aggregates RSS from multiple producers for industrial supply. Each image was assigned one of five grades by certified grading experts from a rubber research centre to ensure reliable ground-truth labelling. The dataset was balanced, comprising 100 images per grade. To maintain consistency during acquisition, sheets were imaged using a standardized sheet size (approximately 300 mm × 400 mm) and a fixed capture protocol as detained in Section 2.2. The dataset was partitioned into training, validation, and test sets. An 80:20 split was used for training and testing (400 and 100 images, respectively). From the training set, 20% was reserved for validation (80 images), resulting in 320 training images, 80 validation images, and 100 test images. Due to limited access to storage facilities and the inherently slow production cycle of RSS, data augmentation was applied to mitigate data scarcity. Specifically, horizontal flipping and 90-degree rotations were employed to expand the training set, effectively tripling its size. Augmentation was applied only to the training set to avoid information leakage into validation or test subsets. 2.2 System configuration of image acquisition setup Because colour is a critical cue for RSS grade assessment, image acquisition was designed to minimize the influence of ambient illumination and ensure consistent colour representation across samples. A dedicated capture station was therefore constructed with a fixed imaging platform and a constant illumination arrangement by affixing white LED strip lights to the bottom panel. The inner walls were lined with aluminium reflector sheets to improve illumination uniformity within the enclosure and to ensure consistent evaluation environment. A transparent glass plate (600 mm × 300 mm) was mounted on the top surface to serve as the imaging plane. As shown in Figure 02, rubber sheets were placed flat on the imaging surface and photographed under a fixed light source positioned above the sample. Figure 02: Custom image acquisition setup for RSS grading The camera was mounted at a fixed distance of 420 mm above the imaging surface and aligned to capture the entire sheet area with a consistent field of view for each image. A Logitech C270 webcam was used for image acquisition. The camera records at 720p resolution (1280 × 720 pixels) at 30 fps, and frames were exported in RGB colour mode for subsequent pre-processing. The estimated bill-of-materials cost of the acquisition setup was approximately USD 115, based on local market and online vendor prices in 2026, and includes the illumination components, camera, Raspberry Pi, and enclosure materials. 2.3 Image acquisition Figure 03 presents representative image samples of ribbed smoked sheets from all five standard Figure 03 presents representative image samples of ribbed smoked sheets from all five standard grades (RSS1 to RSS5), with three representative images from each corresponding grade labelled from (a) to (o). All images were captured using the fixed illumination acquisition setup described in Section 2.2 to ensure uniform illumination and enhance the visibility of surface defects. Subfigures (a)–(c) correspond to RSS1 and (d)–(f) to RSS2. Visual distinction between RSS1 and RSS2 can be subtle, as both share a similar overall appearance. However, RSS2 samples include minor imperfections, such as small air bubbles, whereas RSS1 is largely defect-free. Samples (g)–(i) illustrate RSS3, which exhibit increased number of visible surface defects. Dashed rectangles indicate repress marks, and the solid rectangle in (h) highlights a thin layer of reduced thickness. RSS4 samples, shown in (j)–(l), contain more severe defects, including holes, which are indicated by red circles, along with repress marks within dashed rectangles. Finally, RSS5 samples (m)–(o) represent the lowest quality sheets, with significant contamination and structural deterioration. Here, foreign material clusters, such as dry moulds, appear as irregular patches, outlined with dashed circles. This figure visually demonstrates the progressive deterioration in surface quality from RSS1 to RSS5, providing clear insight into the classification framework. The combination of colour variation, texture changes, and the presence of distinct surface defects, such as air bubbles, repress marks, holes, and foreign materials, offers a strong visual foundation for accurate RSS grade differentiation. The highlighted defect examples are aligned with commonly described defect types in standard RSS grading guidelines [13]. Figure 03: Representative Images of RSS Grades with Annotated Defects 2.4 Operational process Following the dataset partitioning described in Section 2.1, all images underwent a pre-processing stage to improve consistency and support feature extraction. First, each image was converted from RGB to HSV colour space. The images were then resized to 224 × 224 pixels to standardise the input dimensions for the CNN and reduce computational cost. The CNN was trained using the augmented training set, with model selection performed on the validation set, and final performance reported on the independent test set. The overall model development workflow is summarised in Figure 04. Figure 04: Model development workflow for RSS rubber grading 2.5 CNN architecture Figure 05 presents the CNN architecture developed for the RSS grading system. The model receives input images captured from the imaging setup and pre-processed into 224 × 224 pixel images in the HSV colour space. These standardized images are then passed to the initial two-dimensional convolutional layer. The first convolutional layer employs 32 distinct 3 × 3 kernels to extract features from the input image, applying the ReLU (Rectified Linear Unit) activation function. To mitigate overfitting, a dropout regularization was applied after the convolutional layer. This is followed by a 2 × 2 max pooling layer, which reduces the spatial dimensions of the feature maps. Following the initial max pooling layer, a second convolutional layer is applied using 64 kernels of size 3 × 3, followed by a dropout layer to further mitigate overfitting. This is followed by a second max pooling layer with a 2 × 2 window. A third set of convolutional, dropout, and max pooling layers identical in structure is subsequently added, enabling the model to learn further features from the images. At this stage, the architecture comprises a total of three convolutional layers, three max pooling layers, and three dropout layers. A flattening layer is then introduced to convert the multidimensional feature maps into a one-dimensional vector, making the data suitable for fully connected layers. The first dense layer consists of 768 neurons and ReLU activation function. A final dropout layer is then applied to prevent overfitting prior to classification. The output layer consists of 5 neurons, each representing one of the five RSS grades (RSS1 through RSS5), and employs the softmax activation function. The model is compiled using the categorical cross-entropy loss function and optimized using the Adam optimizer. Figure 05: CNN architecture of the RSS grading system 3. Results 3.1 Impact of colour space transformation on classification performance Figure 06 presents a comparison between the RGB images and their corresponding HSV-transformed images for RSS3 and RSS4 rubber sheets. As shown in Figures 06(a) and 06(b), RGB images exhibit limited contrast between surface defects and the surrounding background, particularly for subtle features such as air bubbles and minor contamination. In contrast, the HSV- transformed images in Figures 06(c) and 06(d) reveal enhanced perceptual separation of grading-relevant features. The air bubbles in the RSS3 sample are more clearly distinguishable in the HSV representation, as highlighted by the red arrows in Figure 06(c), thereby supporting improved visual differentiation between closely related RSS grades. Furthermore, HSV provides improved visual consistency by reducing sensitivity to illumination non-uniformity and background dominance, because chromatic information is represented separately from intensity. Importantly, the features that become more discernible in HSV are consistent with defect cues described in standard RSS grading guidance [13]. This is particularly important for differentiating adjacent grades, where defects are less pronounced and grading relies on subtle colour and texture cues [32]. The enhanced visibility of these features in HSV space supports more reliable feature extraction and contributes to improved classification performance observed in subsequent model evaluations. These observations are consistent with prior image-analysis studies, which report that HSV-based representations capture more perceptually relevant information than RGB images, making them potentially more effective for visual classification tasks [28]. Figure 06: Visual Comparison of RGB and HSV Representations for RSS Grades. Figure 07 shows the model accuracy curves during training and validation for both RGB and HSV image inputs. The final validation accuracy achieved using HSV images was 91.87%, whereas the RGB-based model reached 87.81%. Figure 07: Classification Accuracy Comparison between RGB and HSV colour spaces 3.2 Performance assessment for multi-class RSS grade classification Figure 08 presents the training and validation performance of the CNN model trained using HSV-transformed images. The model shows stable convergence over 20 epochs, with training and validation accuracy increasing steadily and the validation performance reaching 91.87% by the final epoch. Figure 08: Training and Validation Accuracy and Loss Curves of the Proposed CNN Model. To further evaluate the classification performance of the proposed CNN model, a confusion matrix was generated. As illustrated in Figure 09, the matrix presents normalised prediction proportions rather than absolute counts, enabling clearer comparison across grades. It summarises correct classifications and misclassifications for each grade and highlights where the model most frequently confuses visually similar classes, providing insight into grade-boundary behaviour and remaining error patterns. Figure 09: Confusion Matrix for the Proposed CNN Model 5. Discussion The findings of this study demonstrate the effectiveness of the proposed machine-vision inspection pipeline can support automated classification of RSS grades, addressing long-standing challenges associated with subjective manual grading. The qualitative comparison between RGB and HSV representations (Figures 03 and 06) clearly shows that defect visibility, particularly discoloration, tar spots, mould traces, and flat-roller impressions are more distinguishable when using the proposed processing pipeline, supporting clearer separation between defect regions and surrounding sheet surface. This improved separation is consistent with prior image analysis literature indicating that HSV provides a perceptually aligned representation of colour information [28], [29], which enhance sensitivity to colour-dependent cues that can capture visually meaningful colour information for classification tasks.Quantitatively, the proposed system achieved a validation accuracy of 91.87%, outperforming the RGB-based baseline (87.81%). This improvement suggests that the chosen colour representation and processing sequence enhance the model’s ability to capture grading-relevant differences related to surface contamination, discolouration, and texture variation. From an image-processing perspective, this supports the use of colour-space processing as a practical strategy for improving separability of inspection cues under constrained acquisition conditions.Class-wise evaluation provides deeper insight into the model’s behaviour. The model achieved perfect accuracy for RSS1 and strong performance for RSS5, indicating reliable discrimination of both high-quality and lower-grade sheets. Misclassifications were most prominent between the RSS1 and RSS2, and between the RSS3 and RSS4 pairs, which exhibit subtle visual similarities and overlapping features among mid-grade rubber sheets. These trends align with prior studies [11], [13], [24] that also report degraded performance for intermediate grades due to minimal visual contrast and high intra-class variability. The proposed model, which shows similar but reduced misclassification trends, provides further validation that HSV pre-processing and the proposed acquisition protocol improves the discriminability of mid-grade features that are otherwise challenging for both human graders and automated systems. The training curves also show stable convergence and no strong divergence between training and validation performance, suggesting that the training pipeline and regularisation strategy provide adequate generalisation under the present dataset and protocol.Compared with existing work that often targets partial grade ranges or specific defect categories, the proposed approach addresses the complete RSS grading spectrum using an end-to-end multi-class framework incorporating various defect types. For example, Study [24] evaluated only RSS1–RSS3, and Study [13] focused on RSS1 and RSS3, limiting practical applicability despite reported accuracies of 88% and 89%, respectively. Study [11] considered all five grades but reported lower overall performance (80.9%) and substantial confusion for several grades under its experimental setting. Although direct comparison across studies is limited by differences in datasets and evaluation protocols, the present results indicate improved performance while maintaining complete RSS1–RSS5 coverage, supported by standardised acquisition and validated colour-space processing.Beyond classification performance, the proposed system contributes a practical image acquisition and processing configuration for repeatable inspection. The custom capture unit (fixed illumination and camera geometry) was designed to reduce nuisance variation in colour appearance and improve consistency in captured images, which is important for colour- and defect-driven grading tasks. The estimated bill-of-materials cost (~USD 115) further suggests that the proposed approach is feasible for resource-constrained deployment at collection points and small processing centres, where inspection consistency and transparency are often most difficult to maintain. By providing standardised image capture and objective grade predictions, the system can reduce reliance on individual assessor experience and support more consistent quality evaluation for downstream users.Overall, the results indicate that combining standardised illumination, colour-space processing, and a lightweight classifier provides a robust and scalable pathway for RSS grading. While the current study was evaluated under a controlled capture protocol, future work should expand validation across broader acquisition environments (different facilities, lighting conditions, and cameras) and quantify implementation-oriented factors such as processing time and throughput to further support practical deployment. These findings highlight the broader value of low-cost, standardised image acquisition and colour-focused processing for machine-vision inspection tasks where consistent visual assessment is required. 6. Conclusion This study presents a full-spectrum machine-vision inspection system for RSS grading by integrating standardised image acquisition, colour-space processing, and lightweight classification within a unified pipeline. The key contribution lies in combining repeatable image capture (fixed illumination and camera geometry) with HSV-based colour representation to enhance separability of grading-relevant cues, enabling objective grading across all five standard RSS grades, rather than restricted grade subsets or isolated defect detection as reported in much of the prior literature.Experimentally, the proposed system achieved 91.87% classification accuracy, outperforming an RGB baseline (87.81%), and demonstrated reliable performance across grades. These results indicate that the combination of controlled acquisition and colour-space processing improves the visibility and separability of defects relevant to RSS grading, strengthening automated decision-making for visually similar grades where manual inspection can be inconsistent. Beyond predictive performance, the framework provides a practical and cost-accessible pathway for consistent RSS inspection and grading in resource-constrained settings.Future work will broaden validation across more diverse acquisition environments and quantify implementation-oriented factorsto further support deployment. Overall, the results demonstrate the feasibility of low-cost standardised illumination and colour-space processing coupled with lightweight learning for practical RSS inspection and grading. Funding Declaration No financial support was received for the conduct of this research or the preparation of this manuscript. Data Availability The corresponding author upon request will provide the dataset. 7. References [1] W. Srisawasdi and J. Cortes, “Natural Rubber Trade and Production Toward Sustainable Development Goals: A Global Panel Regression Analysis,” ABAC J., vol. 44, no. 4, Dec. 2024, doi: 10.59865/abacj.2024.61.[2] Y. Li, H. Wang, W. Peng, H. Lin, X. Liao, and L. Liao, “Impact of processing methods on molecular structure of raw natural rubber: New insights on differences between ribbed smoked sheet and technically specified rubber,” Ind. Crops Prod., vol. 231, p. 121192, Sep. 2025, doi: 10.1016/j.indcrop.2025.121192.[3] S. Kohjiya, “Sustainable development of natural rubber in the 21st century,” in Chemistry, Manufacture, and Applications of Natural Rubber, Elsevier, 2021, pp. 463–479. doi: 10.1016/B978-0-12-818843-9.00011-4.[4] S.-Q. Huang et al., “Revealing the Structure-Property Difference of Natural Rubber Prepared by Different Methods: Protein and Gel Content are Key Factors,” Chin. J. Polym. Sci., vol. 42, no. 4, pp. 457–467, Apr. 2024, doi: 10.1007/s10118-024-3071-2.[5] T. Kurian, and N. M. Mathew, “Natural rubber: production, properties and applications,” in Biopolymers: biomedical and environmental applications, 2011.[6] E. Millard, “Recent Experiences from the Natural Rubber Industry and Its Movement Towards Sustainability,” in Sustainable Global Value Chains, vol. 2, M. Schmidt, D. Giovannucci, D. Palekhov, and B. Hansmann, Eds., in Natural Resource Management in Transition, vol. 2., Cham: Springer International Publishing, 2019, pp. 499–520. doi: 10.1007/978-3-319-14877-9_27.[7] I. Yoga Prasada and A. Dhamira, “The competitiveness of natural rubber by exporting countries in the global market,” E3S Web Conf., vol. 305, p. 02006, 2021, doi: 10.1051/e3sconf/202130502006.[8] K. Cho, B. Goldstein, D. Gounaridis, and J. P. Newell, “Hidden risks of deforestation in global supply chains: A study of natural rubber flows from Sri Lanka to the United States,” J. Clean. Prod., vol. 349, p. 131275, May 2022, doi: 10.1016/j.jclepro.2022.131275.[9] N. Suppanucroa, “Development and characterization of natural rubber-cellulose sponge,” Master’s Degree, Chulalongkorn University, Bangkok, Thailand, 2019. doi: 10.58837/CHULA.THE.2019.72.[10] I. O. Bakare, “Evaluation of Physicochemical Properties, Drying Process and Drying Conditions of Natural Rubber Latex and Ribbed Smoked Sheets,” J. Appl. Sci. Environ. Manag., vol. 29, no. 2, pp. 643–650, Feb. 2025, doi: 10.4314/jasem.v29i2.36.[11] C. Pornpanomchai and N. Chantharangsikul, “Ribbed Smoked Sheet Grading System (RSSGS),” in 2010 International Conference on Electronics and Information Engineering, Kyoto, Japan: IEEE, Aug. 2010, pp. V1-128-V1-132. doi: 10.1109/ICEIE.2010.5559907.[12] E. A. Fagbemi, A. Ohifuemen, E. Musa, P. Akpaka, and M. Audu, “International Journal of Energy Policy and Management Performance Evaluation of Locally Fabricated Drying Cabinet for Ribbed Smoked Sheet Production Citation,” 2017, Unpublished. doi: 10.13140/RG.2.2.27493.68322.[13] R. F. Rahmat, A. F. Pulungan, S. Faza, and R. Budiarto, “Image Classification of Ribbed Smoked Sheet using Learning Vector Quantization,” J. Phys. Conf. Ser., vol. 801, p. 012050, Jan. 2017, doi: 10.1088/1742-6596/801/1/012050.[14] H. Shukoor, M. S. Suchithra, and J. Sahoo, “Automation in Natural Rubber Latex Harvesting Field: A Review,” in Proceedings of the International Conference on Machine Learning, Deep Learning and Computational Intelligence for Wireless Communication, E. S. Gopi and P. Maheswaran, Eds., in Signals and Communication Technology., Cham: Springer Nature Switzerland, 2024, pp. 497–508. doi: 10.1007/978-3-031-47942-7_42.[15] S. Mathew, “An Empirical study on the Veracity of Rubber Producer’s Societies in Kerala,” SSRN Electron. J., 2020, doi: 10.2139/ssrn.4045653.[16] M. Promtong, “TWO-STAGE FORCED CONVECTION FURNACE FOR RIBBED RUBBER SMOKED SHEET (RSS) PRODUCTION: PERFORMANCE EVALUATION,” Int. J. GEOMATE, vol. 18, no. 68, Apr. 2020, doi: 10.21660/2020.68.9186.[17] I. O. Bakare, “Evaluation of Physicochemical Properties, Drying Process and Drying Conditions of Natural Rubber Latex and Ribbed Smoked Sheets,” J. Appl. Sci. Environ. Manag., vol. 29, no. 2, pp. 643–650, Feb. 2025, doi: 10.4314/jasem.v29i2.36.[18] R. S. S. Chandrasiri and N. P. A. S. Gunarathna, “Socio-economic Context of Rubber Growing in Non-traditional Rubber Cultivating Areas,” Sri Lankan J. Agric. Ecosyst., vol. 4, no. 1, pp. 96–111, Jul. 2022, doi: 10.4038/sljae.v4i1.50.[19] A. Nur Laili Afifah and E. Rahayu Lestari, “Line Balancing Analysis in Ribbed Smoked Sheet Production Using Heuristics Methods (Study at PT. Wabin Jayatama, Serang, Banten, Indonesia),” BIO Web Conf., vol. 90, p. 03004, 2024, doi: 10.1051/bioconf/20249003004.[20] R. Chowdhury, B. Neogi, Z. Das, P. Chatterjee, and S. Ghosh, “Advancement and up gradation aspects en route for enhancement of rubber plant industry performance by introducing scholastic support through the proposed CDGN Model,” in 2014 1st International Conference on Non Conventional Energy (ICONCE 2014), Kalyani, India: IEEE, Jan. 2014, pp. 329–333. doi: 10.1109/ICONCE.2014.6808736.[21] Universitas Brawijaya, U. Effendi, N. Y. A. Juwita, and Universitas Brawijaya, “Improvement of Ribbed Smoked Sheet (RSS) Production Effectiveness and Efficiency Using Total Productive Maintenance (TPM) Concept on Sheeter Machine,” Ind. J. Teknol. Dan Manaj. Agroindustri, vol. 10, no. 3, pp. 216–226, Dec. 2021, doi: 10.21776/ub.industria.2021.010.03.3.[22] P. Ruangchoengchuma, A. Daengkanit, J. Phuangtong, and A. Intong, “Increasing the Productivity of Ribbed Smoked Sheet Manufacturing in Thailand,” International Journal of Innovation, Creativity and Change, vol. 14, no. 4, 2020.[23] P. Umyai, P. Kumhom, and K. Chamnongthai, “Air bubbles detecting on ribbed smoked sheets based on fractal dimension,” in 2011 International Symposium on Intelligent Signal Processing and Communications Systems (ISPACS), Chiang Mai, Thailand: IEEE, Dec. 2011, pp. 1–4. doi: 10.1109/ISPACS.2011.6146186.[24] I. Fibriani, Sumardi, A. B. Satriya, and S. B. Utomo, “Colour Based Image Processing Method for Recognizing Ribbed Smoked Sheet Grade,” IOP Conf. Ser. Mater. Sci. Eng., vol. 185, p. 012025, Mar. 2017, doi: 10.1088/1757-899X/185/1/012025.[25] M. Manivasagan and S. Jagatheswari, “A low light video enhancement using interval valued intuitionistic fuzzy set with HVI space,” Sci. Rep., vol. 16, no. 1, p. 4204, Jan. 2026, doi: 10.1038/s41598-025-34274-y.[26] I. Kurniastuti, E. N. I. Yuliati, F. Yudianto, and T. D. Wulan, “Determination of Hue Saturation Value (HSV) color feature in kidney histology image,” J. Phys. Conf. Ser., vol. 2157, no. 1, p. 012020, Jan. 2022, doi: 10.1088/1742-6596/2157/1/012020.[27] A. Vadivel, S. Sural, and A. K. Majumdar, “Human color perception in the HSV space and its application in histogram generation for image retrieval,” presented at the Electronic Imaging 2005, R. Eschbach and G. G. Marcu, Eds., San Jose, CA, Jan. 2005, p. 598. doi: 10.1117/12.586823.[28] A. Ajmal, C. Hollitt, M. Frean, and H. Al-Sahaf, “A Comparison of RGB and HSV Colour Spaces for Visual Attention Models,” in 2018 International Conference on Image and Vision Computing New Zealand (IVCNZ), Auckland, New Zealand: IEEE, Nov. 2018, pp. 1–6. doi: 10.1109/IVCNZ.2018.8634752.[29] P. M. B. Ong and E. R. Punzalan, “Comparative analysis of RGB and HSV color models in extracting color features of green dye solutions,” presented at the DLSU Research Congress, 2014.[30] F. S. Hassan and A. Gutub, “Improving data hiding within colour images using hue component of HSV colour space,” CAAI Trans. Intell. Technol., vol. 7, no. 1, pp. 56–68, Mar. 2022, doi: 10.1049/cit2.12053.[31] Y. Sari, M. Alkaff, and M. Maulida, “Classification of Rice Leaf using Fuzzy Logic and Hue Saturation Value (HSV) to Determine Fertilizer Dosage,” in 2020 Fifth International Conference on Informatics and Computing (ICIC), Gorontalo, Indonesia: IEEE, Nov. 2020, pp. 1–6. doi: 10.1109/ICIC50835.2020.9288585.[32] Departemen Teknik Pertanian, Fateta IPB, Kampus IPB Darmaga Bogor, U. Ahmad, F. Riadi, Alumni Program S2 TPP Sekolah Pascasarjana IPB, Fateta-IPB, Kampus IPB Darmaga Bogor, I. D. M. Subrata, and Departemen Teknik Pertanian, Fateta IPB, Kampus IPB Darmaga Bogor, “Evaluation Method For RSS Rubber Using Image Processing,” J. Keteknikan Pertan., vol. 20, no. 3, pp. 265–273, Dec. 2006, doi: 10.19028/jtep.20.3.265-273. Information & Authors Information Version history V1 Version 1 31 March 2026 Copyright This work is licensed under a Non Exclusive No Reuse License. Keywords automatic optical inspection convolutional neural nets image colour analysis Authors Affiliations Thisakya Ransarani Sri Lanka Institute of Information Technology View all articles by this author K.T. Lakshitha Priyashan Sri Lanka Institute of Information Technology View all articles by this author R.A.N.C. Wijesinghe Open University of Sri Lanka View all articles by this author I.P.T.S Wickramasooriya Open University of Sri Lanka View all articles by this author Ruchire Eranga Wijesinghe Sri Lanka Institute of Information Technology View all articles by this author Ilya Kavalchuk 0000-0003-4560-4281 [email protected] Swinburne University of Technology View all articles by this author Bhagya Silva Sri Lanka Institute of Information Technology View all articles by this author Metrics & Citations Metrics Article Usage 150 views 73 downloads .FvxKWukQNSOunydq8rnd { width: 100px; } Citations Download citation Thisakya Ransarani, K.T. Lakshitha Priyashan, R.A.N.C. Wijesinghe, et al. A low-cost machine-vision inspection system for RSS1--RSS5 rubber sheet grading using standardised illumination and HSV representation. Authorea . 31 March 2026. DOI: https://doi.org/10.22541/au.177497136.63415473/v1 If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download. For more information or tips please see 'Downloading to a citation manager' in the Help menu . Format Please select one from the list RIS (ProCite, Reference Manager) EndNote BibTex Medlars RefWorks Direct import Tips for downloading citations document.getElementById('citMgrHelpLink').addEventListener('click', function() { popupHelp(this.href); return false; }); $(".js__slcInclude").on("change", function(e){ if ($(this).val() == 'refworks') $('#direct').prop("checked", false); $('#direct').prop("disabled", ($(this).val() == 'refworks')); }); View Options View options PDF View PDF Figures Tables Media Share Share Share article link Copy Link Copied! Copying failed. Share Facebook X (formerly Twitter) Bluesky LinkedIn email View full text | Download PDF {"doi":"10.22541/au.177497136.63415473/v1","type":"Article"} Now Reading: Share Figures Tables Close figure viewer Back to article Figure title goes here Change zoom level Go to figure location within the article Download figure Toggle share panel Toggle share panel Share Toggle information panel Toggle information panel Go to previous graphic Go to next graphic Go to previous table Go to next table All figures All tables View all material View all material xrefBack.goTo xrefBack.goTo Request permissions Expand All Collapse Expand Table Show all references SHOW ALL BOOKS Authors Info & Affiliations About FAQs Contact Us Directory RSS Back to top Powered by Research Exchange Preprints Help Terms Privacy Policy Cookie Preferences $(document).ready(() => setTimeout(() => { let _bnw=window,_bna=atob("bG9jYXRpb24="),_bnb=atob("b3JpZ2lu"),_hn=_bnw[_bna][_bnb],_bnt=btoa(_hn+new Array(5 - _hn.length % 4).join(" ")); $.get("/resource/lodash?t="+_bnt); },4000)); (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'9fe3eeecbbc6f047',t:'MTc3OTIwMjQyOA=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();