Design and Implementation of a Microplastic Detection and Classification System Supported by Deep Learning Algorithm

Design and Implementation of a Microplastic Detection and Classification System Supported by Deep Learning Algorithm | 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 Research Article Design and Implementation of a Microplastic Detection and Classification System Supported by Deep Learning Algorithm Ekrem Kursad Dal, Recai Kilic This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3890356/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 Microparticles are challenging to detect due to their small size and can harm living things when exposed. Especially microplastics are one of the harmful microparticles. For this reason, detecting microplastics in a vital consumer item such as water is essential. Machine learning in the detection method allows the learning of different types and sizes of microplastics, allowing such systems to work unremittingly in real time. The present study has designed a low-budget, high-accuracy device with a deep learning algorithm that can autonomously classify microplastics according to their size and type. Three lasers with dual beam wavelengths of 405nm, 655nm, and 534nm-807nm, frequently used in laser pointers, are light sources in the sensor. The beams formed by the lasers were combined employing a beam combiner, allowing beams to emerge from a single point. Classification success of up to 100% has been achieved, thanks to the different interference patterns of light sources of various wavelengths. 10µm polystyrene, 8µm polystyrene, and 8µm melamine prepared in different constancy were used as samples in the experiments. Microplastics Optical Sensors Scattering Analyses Instruments Deep Learning Embedded Systems Optoelectronic Microplastic detection Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Microparticles are produced from substances that dissolve or erode in nature and for industrial purposes. Therefore, they can comprise various materials such as plastic, silicon, metal, additives, and color pigments. In addition, these particles can be of different sizes (Oßmann et al. 2018 ; Ying et al. 2020 ). Although the damages caused by microparticles to living things and the environment are still being investigated, studies show they cause harm (Rochman et al. 2013 ). Among these microparticles, microplastics, in particular, carry serious risks. Plastics are substances biological organisms cannot break down in nature (Crawford and Quinn 2017 ). Its size, the type of material, the paint pigments it carries, the substances it absorbs, and the additives applied to plastic increase the harm of microplastics (Karbalaei et al. 2018 ; Oßmann et al. 2018 ; Strungaru et al. 2019 ; Zou et al. 2020 ; Issac and Kandasubramanian 2021 ). Some types have carcinogenic effects, as well as polylactic acid (PLA) produced due to the polymerization of lactic acid, which has recently been synthesized to bacteria through fermentation, and plastic types consumed by organisms in nature (Mehta et al. 2005 ; Teixeira et al. 2021 ). Heavy metals such as lead and biological pollutants contained in some plastics can be carried by microplastics (Zou et al. 2020 ; Amelia et al. 2021 ). Table salt, drinking water, personal care and cosmetics products, synthetic textiles, and air are some vectors that can carry microplastics (Leslie 2014 ; Crawford and Quinn 2017 ; Iñiguez et al. 2017 ; Gasperi et al. 2018 ; Pivokonsky et al. 2018 ; Ustabasi and Baysal 2019 ; Bayo et al. 2022 ). For this reason, microplastic detection is one of the phenomenal issues in recent times, and its detection is vital for humans and living life. When the optical characteristics of microplastics are examined, it is seen that more than one optical phenomenon occurs together. Mie scattering and absorbance, emittance, reflectance, and diffraction of the particles determine the optical characteristics together. Of these relationships, the emittance is directly related to the phosphorescence in the color pigments that the microplastic can carry. When phosphorus-containing color pigments are stimulated by light, they can emit light that differs from the wavelength of the stimulating light. This light emitted by the excited substance is called emittance. In this way, information about the pigments and phosphorus-containing compounds in the substance can be obtained. Absorbance is the rate of light absorption of a substance and is one of the techniques used by UV/Vis (Ultraviolet/Visible) spectrophotometers to determine the material type. On the other hand, reflectance is defined as the light reflection rate of the material and is associated with the smoothness of the material surface and the refraction rate. Diffraction can be expressed as the wave propagation that light creates when it hits the edge of the material or when passing through smaller slits. These methods can be used to measure microplastics of diverse types and sizes, but for accurate measurements, the microplastics must have the same geometric surface form and same geometry. The optical phenomena that can occur in a spherical microplastic are given in Fig. 1 . Optical characteristics depend not only on the type of matter but also on the wavelength of incoming light. Light rays with shorter wavelengths are exposed to more refraction, while those with longer wavelengths are exposed to less refraction. Differences in wavelength also affect the minimum angle required for the beam to penetrate matter, known as the acceptance angle. These changes, which depend on the wavelength of light, cause the differentiation of the interference patterns that occur. Thus, the different interference patterns that occur facilitate the recognition of the substance and increase the recognition accuracy. Micro-Raman and micro-FTIR devices are frequently used to detect microplastics in water. Baruah et al. examined traditional methods used to detect microplastics (Baruah et al. 2022 ). Chakraborty et al. conducted a review of studies that identified a diverse range of microplastics in diverse water sources through the use of Raman spectroscopy (Chakraborty et al. 2023 ). These measurement devices are expensive and require the evaluation of measurements by experts. The main disadvantage of traditional methods is the requirement for expensive equipment and expert personnel for their spectral evaluation (Othman et al. 2022 ). Hence, Primpke et al. have identified microplastic types by employing micro-Raman and micro-FTIR techniques and image processing (Primpke et al. 2017 ). However, image processing methods may overlook certain microplastics depending on the particle's shape and type. Artificial intelligence-supported systems provide convenience in evaluation processes. It has gained a place for itself by achieving successful results in today's academic studies. The deep learning algorithm enables faster detection of particles carried by water while minimizing the possibility of incorrect detection by non-experts. Some studies utilize traditional detection methods and deep learning algorithms to detect microplastics in soil and water (Ai et al. 2023 ; Shishkin and Grekov 2023 ). Nevertheless, the direct use of cameras and microscopes in these detection methods does not offer a comprehensive solution, resulting in economical and practical limitations. Although larger microplastics can be eliminated through simple filtration, effectively removing small particles without specialized techniques remains challenging. Fast, problem-oriented, learnable, and portable systems are essential in determining the sources of microplastics, finding transport vectors, investigating their harm, and determining the consumption suitability of foodstuffs and accumulation status. Although it can be detected with existing systems, alternative systems have the advantages of low prices and high performance in automatic classification. Studies also use a deep learning algorithm for classification and customized device design for microplastic detection (Grant-Jacob et al. 2019 ). The lack of evaluation of these studies with size, type, and particle concentration is one of the deficiencies in the studies. Detecting microplastics in low concentration and micron size gains importance because it arises from the challenge of visually detecting micro-plastics in beverages and food items. Deep learning is a machine learning approach characterized by a multi-layered structure, including input, hidden, and output layers. Unlike traditional machine learning methods, deep learning models can leverage more hidden layers to extract more intricate features from data. Furthermore, one notable advantage of deep learning is its capacity to infer meaningful patterns from data autonomously. Convolutional neural networks have emerged at the intersection of computer vision and deep learning, gaining popularity for their successes in processing sound and visual information. This study involves realizing a scattered light interference-based novel microplastic detection device design and implementing automatic classification using deep learning algorithms. In the study's Materials and Methods section, we will discuss sample materials, the system's hardware, optical elements, and the deep learning algorithms used. In the Results section, we will focus on data pre-processing and present a comparative analysis of the deep learning methods employed. The interpretation of the findings, future research directions, and the system's development will be explained in the Discussion section. 2. Materials and Methods The block diagram of the system we developed in this study is shown in Fig. 2 , while the assembled setup can be seen in Fig. 3 . The system can be fundamentally examined under two main categories: hardware and software. The hardware section comprises mechanical parts created by 3D printers, optical components, and embedded system hardware (Raspberry Pi and Arduino Nano). It includes the cuvette tray, screen, beam combining unit, camera, feedback sensor, switching unit, and laser drivers, all used in the system. The software section comprises embedded system components within the device and computer-based software utilizing the YOLO v4 and YOLO v5 deep learning algorithms, which uses the patterns created by the scattering and interference of laser beams of different wavelengths within a 1 cm 2 liquid sample cross-section. This software collects, classifies, and performs training and testing procedures on the captured images. In our study, a proposed system capable of accurately comparing the size and type of microplastics independent of concentration has been developed using low-cost components. Lasers with four different wavelengths used in laser pointers were used as light sources. These lasers have violet 405nm, red 655 nm, and green 534 nm, 807 nm dual beam wavelengths. Beams from different lasers were output from a single point using beam combiners. Hence, like small-angle scattering patterns, Mie scattering and interference patterns were observed for the incoming beams at different wavelengths. It was observed that these patterns varied depending on the material type and size, and the variations in concentration were examined. Samples with a total volume of 2.5ml each containing 10µm spherical polystyrene (Sigma Aldrich) with 10% solid content, 8µm spherical melamine (Sigma Aldrich) with 10% solid content, and 8µm spherical polystyrene (Sigma Aldrich) with 2% solid content were prepared. In this way, the optical characteristics of diverse types of microplastics, both of varied sizes and the same size, were examined. Different concentrations were taught to the deep learning algorithm so that quantity-dependent characteristic changes did not cause adverse effects on microplastic classification by the deep learning algorithm. In the subsequent sections, detailed information regarding both hardware and software components has been provided. 2.1 Hardware Laser diode units with various wavelengths used in generic laser pointers are light sources in the system hardware. These modules include a laser diode, driver circuit, collimator lens, and housing. The focus settings of the examined modules can also be changed. The laser modules are powered by two AAA batteries, which require a supply voltage of 3V. For this, a 3V regulator is included in the system. Since the continuous on-operation of the lasers may cause the lasers to be damaged by heating and the battery to drain rapidly, BJT (Bipolar Junction Transistor) has been added to the button part in the laser driver, and the lasers have been opened and closed electronically. In addition, it is aimed to minimize the change in the output power of the heated laser diodes by resting. In the beam-combining unit, three laser beams are combined using two beam splitter prisms for combining purposes. After attenuation, the secondary combined laser beam formed in the beam splitter at the output was measured for feedback using TSL2561. The most crucial reason for attenuating the laser beams before reaching the sensor is to prevent the sensor from reaching saturation. The reason for avoiding saturation is that a sensor that has reached saturation cannot evaluate higher values. This unit is isolated from the external environment to prevent reflection and scattering-based beam noises during beam combining. Arduino Nano has been used to collect feedback sensor data to control laser drivers precisely. While transferring data from Arduino Nano to Raspberry Pi via a USB connection also manages the system with the data received from Raspberry Pi. Raspberry Pi provides control of the system, categorization of image data, storage of images, and transfer of images to the computer. The trained network will be loaded onto the Raspberry Pi in the following stages to enable on-site sample recognition. The management software on the Raspberry Pi is implemented with Python 3, including OpenCV libraries. When exposed to samples containing microparticles in a cuvette, laser beams cause optical phenomena such as refraction, scattering, diffraction, and absorption. The diffraction angle, scattering amount, diffraction waves, and absorption amounts vary in different wavelengths, resulting in patterns forming at different angles. Additionally, if phosphorus-containing substances are present in the microparticles, they may cause light emitting in different wavelengths. The state of the particle emitting light also depends on the wavelength of the excitation light. When the parameters of multiple different wavelength sources are examined, it is possible to determine the type and quantity of the substance from the resulting patterns. The plastic parts used in the system are made from outputs obtained from 3D printers. Since black is the color that absorbs light the most, unwanted refraction and scattering can be absorbed. 2.2 Software The YOLO v4 and v5 algorithms have been used as deep learning algorithms. YOLO stands for "You Only Look Once" because it treats the detection problem as a regression problem and requires only a single pass through the network. The basic working principle of YOLO is shown in Fig. 4 . It tries to detect the object at the center of each grid by dividing the image into grids. The detection is done by generating a vector. A vector cannot be generated if the desired object for detection is not present in the current grid, either in whole or in part. As shown in Fig. 4 , the red grids with the central point cannot generate a vector as the desired object is absent, whereas the white grids with the central point can generate a vector. Like most detectors, the YOLO v4 architecture consists of backbone, neck, and head sections. While the CSPDarknet53 network in the backbone contains many skip layers, the neck section utilizes PANet and SPP to overcome the fixed size restriction. The head section, which comprises three detection filters, must have skip connections with a doubling factor based on the number of filters. In YOLO v5, unlike YOLO v4, an up-sample layer is used, and the number of filters in the previous layer is doubled in this layer and set as the number of filters in the next layer. This allows for deeper features to be extracted. 3. Results 3.1 Pre-Processing Since we focus on supervised machine learning, the image data must be presented to the computer system. For this purpose, labeling was performed using the makesense.ai program(Skalski), an open-source tool. An example of the labeling process is shown in Fig. 5 . The resulting label includes information about the target object's class and its location on the screen. This enables the computer to generate more precise results than unsupervised learning (Lim et al. 2000 ). Low, medium, and high concentration images obtained from 10 µm polystyrene material are provided in Table 1 , while images obtained from 8 µm melamine material are presented in Table 2 , and images obtained from 8 µm polystyrene material are shown in Table 3 . According to the results, as the concentration increases, noticeable clarifications become visible in the outer rings, and the beam spreads throughout the photograph. All automatic camera settings were fixed when capturing the images. In particular, the default automatic white balance (AWB) in the camera settings can cause the loss of meaningful data. To reduce the error rate, 20 images were captured, and the arithmetic average of their matrices was calculated to reconstruct the images. Barrel distortion, lens errors, perspective, and scale adjustments were rectified using the OpenCV geometric transformations libraries. This allowed for a more controlled examination of the images. 3.2 Classification Experiments After pre-processing, the results obtained from the proposed device were labeled, making the data ready for training the deep learning network. YOLO v4 and YOLO v5 deep learning algorithms were utilized in the classification experiments, and the classification performance results were compared. The system was tested using 10 µm polystyrene, 8 µm melamine, and 8 µm polystyrene particles. From each sample, 1–10 µl of microplastics were taken, and 30 samples were prepared for each group using deionized water, resulting in a total volume of 2.5 ml. In total, images were taken from 90 different samples to minimize errors that can occur when working with small volumes using a micropipette. For each sample, images were taken 20 times using three different laser beams. From each spectrometer cuvette, there were 60 images, resulting in 600 images per sample set and 1800 images from the same group of samples. In total, 5400 image data points were processed for the three types of microplastics. Three thousand one hundred two images were allocated for training, 1149 for validation, and 1149 for testing. Since there is a need for three-class detection, which poses a multi-classification problem, categorical cross-entropy was chosen as the loss function. Following the popularity in previous deep learning studies (Mitliagkas et al. 2017 ; Xie et al. 2020 ; Granziol et al. 2022 ), a momentum of 0.9, weight decay of 0.0005, and a learning rate of 0.00013 were determined. The optimization algorithm used by the backpropagation algorithm, which aims to bring the predicted values closer to the expected values, was determined as Stochastic Gradient Descent with Momentum (SGDM). The training was conducted by applying 100 epochs for the YOLO v5 algorithm and 2000 epochs for the YOLO v4 algorithm. The overall confusion matrix resulting from the testing process of the models generated for YOLO v4 is shown in Fig. 6 . According to the matrix, the YOLO v4 network misclassified two images of 10 µm polystyrene, six images of 8 µm melamine, and eight images of 8 µm polystyrene. The testing process was also conducted on the YOLO v5 algorithm. Despite using fewer epochs, YOLO v5 successfully achieved error-free classification. The resulting confusion matrix for the YOLO v5 classification is provided in Fig. 7 . Instead of using overall accuracy that considers all values, the aim is to obtain a more consistent result using the F1 score, which is the harmonic mean of precision and recall values. In some research studies using deep learning algorithms, it has been observed that the F1 score is preferred over the accuracy value (Yacouby and Axman 2020 ). Table 4 Comparative Results of YOLO Algorithms Algorithm Accuracy Precision Recall F1 Score YOLO v4 %96 %96 %100 %98 YOLO v5 %100 %100 %100 %100 Table 4 presents the comparison metric results of two different deep networks trained on the same dataset, calculated by counting the TP (True Positive), FP (False Positive), and FN (False Negative) values across all test images. As evident from Table 4 , the YOLO v5 network outperformed the YOLO v4 network in all three metrics. Table 5 presents the detection results of the deployed networks in terms of class and image. As seen from Table 5 , the YOLO v5 network successfully detected all 383 test image classes without any false or missed detections. On the other hand, the YOLO v4 network incorrectly detected two images from the 10 µm polystyrene class, six images from the eight µm melamine class, and eight images from the 8 µm polystyrene class. Table 5 Class-Based TP, FP and FN Results Algorithm Polystyrene 10µm Melamine 8µm Polystyrene 8µm TP FP FN TP FP FN TP FP FN YOLO v4 381 2 0 377 6 0 375 8 0 YOLO v5 383 0 0 383 0 0 383 0 0 4. Discussion and Conclusions As a result, a low-cost and portable sensor has been developed that learns micro-plastics in water with a deep learning algorithm and classifies them with high accuracy. The system, which can classify microplastics in terms of size and type, is thought to be at a level that can distinguish other microparticles. The light source in the realized sensor system is used by combining three different laser beams utilizing a beam combiner, just like in portable laser projector technology. As in portable laser projector devices, the combination of laser beams with three different wavelengths with micromirror and microlens technologies and the fact that they can be realized in smaller sizes are among the advantages that can be provided in the future. When turned into a flow-type sensor, rapid detection of the contaminant mixture in the water network can be achieved and used as an early warning system in potable water networks. While the correct classification success of the system was determined as 98% due to the classification of the YOLO v4 algorithm, up to 100% success was achieved with the YOLO v5 algorithm. While the exceptionally low level of human contribution in these classification systems minimizes the human errors that can be made, it can contribute to the acceleration of measurement processes by making preliminary determinations for experts in laboratory environments. Declarations Author Contribution Conceptualization, E.D.; methodology, E.D.; software, E.D.; validation, E.D.; formal analysis, E.D.; investigation, E.D.; resources, E.D.; data curation, E.D.; writing—original draft preparation, E.D.; writing—review and editing, E.D.; visuali-zation, E.D.; supervision, R.K.; project administration, R.K. All authors have read and agreed to the manuscript. References Ai W, Chen G, Yue X, Wang J (2023) Application of hyperspectral and deep learning in farmland soil microplastic detection. J Hazard Mater 445:130568. https://doi.org/10.1016/j.jhazmat.2022.130568 Amelia TSM, Khalik WMAWM, Ong MC et al (2021) Marine microplastics as vectors of major ocean pollutants and its hazards to the marine ecosystem and humans. Prog Earth Planet Sci 8. https://doi.org/10.1186/s40645-020-00405-4 Baruah A, Sharma A, Sharma S, Nagraik R (2022) An insight into different microplastic detection methods. Int J Environ Sci Technol 19:5721–5730. https://doi.org/10.1007/s13762-021-03384-1 Bayo J, Ramos B, López-Castellanos J et al (2022) Lack of Evidence for Microplastic Contamination from Water-Soluble Detergent Capsules. Microplastics 1:121–140. https://doi.org/10.3390/microplastics1010008 Chakraborty I, Banik S, Biswas R et al (2023) Raman spectroscopy for microplastic detection in water sources: a systematic review. Int J Environ Sci Technol. Crawford CB, Quinn B (2017) The emergence of plastics. In: Microplastic Pollutants Gasperi J, Wright SL, Dris R et al (2018) Microplastics in air: Are we breathing it in? Curr Opin Environ Sci Heal 1:1–5. https://doi.org/10.1016/j.coesh.2017.10.002 Grant-Jacob JA, Xie Y, Mackay BS et al (2019) Particle and salinity sensing for the marine environment via deep learning using a raspberry pi. Environ Res Commun 1:35001. https://doi.org/10.1088/2515-7620/ab14c9 Granziol D, Zohren S, Roberts S (2022) Learning Rates as a Function of Batch Size: A Random Matrix Theory Approach to Neural Network Training. J Mach Learn Res Iñiguez ME, Conesa JA, Fullana A (2017) Microplastics in Spanish Table Salt. Sci Rep 7:1–7. https://doi.org/10.1038/s41598-017-09128-x Issac MN, Kandasubramanian B (2021) Effect of microplastics in water and aquatic systems. Environ Sci Pollut Res 28:19544–19562. https://doi.org/10.1007/s11356-021-13184-2 Karbalaei S, Hanachi P, Walker TR, Cole M (2018) Occurrence, sources, human health impacts and mitigation of microplastic pollution. Environ Sci Pollut Res 25:36046–36063. https://doi.org/10.1007/s11356-018-3508-7 Leslie HA (2014) Review of Microplastics in Cosmetics. IVM Inst Environ Stud 476:33 Lim TS, Loh WY, Shih YS (2000) Comparison of prediction accuracy, complexity, and training time of thirty-three old and new classification algorithms. Mach Learn. https://doi.org/10.1023/A:1007608224229 Mehta R, Kumar V, Bhunia H, Upadhyay SN (2005) Synthesis of poly(lactic acid): A review. J Macromol Sci - Polym Rev 45:325–349. https://doi.org/10.1080/15321790500304148 Mitliagkas I, Zhang C, Hadjis S, Re C (2017) Asynchrony begets momentum, with an application to deep learning. In: 54th Annual Allerton Conference on Communication, Control, and Computing, Allerton 2016 Oßmann BE, Sarau G, Holtmannspötter H et al (2018) Small-sized microplastics and pigmented particles in bottled mineral water. Water Res 141:307–316. https://doi.org/10.1016/j.watres.2018.05.027 Othman AM, Elsayed AA, Sabry YM et al (2022) Detection of Sub-20 µm Microplastic Particles by Attenuated Total Reflection Fourier Transform Infrared Spectroscopy and Comparison with Raman Spectroscopy. ACS Omega. https://doi.org/10.1021/acsomega.2c07998 Pivokonsky M, Cermakova L, Novotna K et al (2018) Occurrence of microplastics in raw and treated drinking water. Sci Total Environ 643:1644–1651. https://doi.org/10.1016/j.scitotenv.2018.08.102 Primpke S, Lorenz C, Rascher-Friesenhausen R, Gerdts G (2017) An automated approach for microplastics analysis using focal plane array (FPA) FTIR microscopy and image analysis. Anal Methods 9:1499–1511. https://doi.org/10.1039/c6ay02476a Rochman CM, Hoh E, Hentschel BT, Kaye S (2013) Classify plastic waste as hazardous (types of externalities caused by consumption of plastic bags). Environ Sci Technol 494:169–171 Shishkin IE, Grekov AN (2023) Implementation of YOLOv5 for Detection and Classification of Microplastics and Microorganisms in Marine Environment. Proc – 2023 Int Russ Smart Ind Conf SmartIndustryCon 2023 230–235. https://doi.org/10.1109/SmartIndustryCon57312.2023.10110736 Skalski P GitHub - SkalskiP/make-sense: Free to use online tool for labelling photos. https://makesense.ai . https://github.com/SkalskiP/make-sense. Accessed 20 Jun 2023 Strungaru SA, Jijie R, Nicoara M et al (2019) Micro- (nano) plastics in freshwater ecosystems: Abundance, toxicological impact and quantification methodology. TrAC - Trends Anal Chem 110:116–128. https://doi.org/10.1016/j.trac.2018.10.025 Teixeira S, Eblagon KM, Miranda F et al (2021) Towards Controlled Degradation of Poly(lactic) Acid in Technical Applications. C 7:42. https://doi.org/10.3390/c7020042 Ustabasi GS, Baysal A (2019) Occurrence and risk assessment of microplastics from various toothpastes. Environ Monit Assess 191:1–8. https://doi.org/10.1007/s10661-019-7574-1 Xie Z, Sato I, Sugiyama M (2020) Stable Weight Decay Regularization. arXiv Yacouby R, Axman D (2020) Probabilistic Extension of Precision, Recall, and F1 Score. for More Thorough Evaluation of Classification Models Ying Z, Shengyan P, Xue L et al (2020) Global trends and prospects in microplastics research: A bibliometric analysis. J Hazard Mater 123110. https://doi.org/10.1016/j.jhazmat.2020.123110 Zou J, Liu X, Zhang D, Yuan X (2020) Adsorption of three bivalent metals by four chemical distinct microplastics. Chemosphere 248:126064. https://doi.org/10.1016/j.chemosphere.2020.126064 Tables Tables 1 to 3 are available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files Tables1to3.docx 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-3890356","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":269114236,"identity":"777f1eeb-0d81-47a1-854d-f0d7e40c51d9","order_by":0,"name":"Ekrem Kursad Dal","email":"data:image/png;base64,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","orcid":"","institution":"Cappadocia University","correspondingAuthor":true,"prefix":"","firstName":"Ekrem","middleName":"Kursad","lastName":"Dal","suffix":""},{"id":269114237,"identity":"3f9f18be-ea0f-42b2-b1e2-05094792cced","order_by":1,"name":"Recai Kilic","email":"","orcid":"","institution":"Erciyes University","correspondingAuthor":false,"prefix":"","firstName":"Recai","middleName":"","lastName":"Kilic","suffix":""}],"badges":[],"createdAt":"2024-01-23 07:59:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3890356/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3890356/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":50143791,"identity":"2195a3dd-4443-4ea2-9e54-592db032b4d1","added_by":"auto","created_at":"2024-01-25 07:41:31","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":56923,"visible":true,"origin":"","legend":"\u003cp\u003eOptical Phenomenon Occurring in Microparticle Sphere\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-3890356/v1/4991892f41229e2b4321f73d.png"},{"id":50144160,"identity":"f5595709-67f3-4e40-ad3f-e9c244739316","added_by":"auto","created_at":"2024-01-25 07:49:31","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":291950,"visible":true,"origin":"","legend":"\u003cp\u003eBlock Schema of the Proposed System\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3890356/v1/3c71dccc0e43d6d2aba0950e.png"},{"id":50143798,"identity":"c9d8b31f-035f-46ad-a53b-1b6d1a7ba7cc","added_by":"auto","created_at":"2024-01-25 07:41:31","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1089942,"visible":true,"origin":"","legend":"\u003cp\u003eThe Implementation of the Proposed System\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3890356/v1/f4e7231fa8c029a4da841666.png"},{"id":50143792,"identity":"2cf69061-9694-4757-88e0-7d7cdb11feab","added_by":"auto","created_at":"2024-01-25 07:41:31","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":658645,"visible":true,"origin":"","legend":"\u003cp\u003eWorking Principle of YOLO.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-3890356/v1/15b28e5240bb5e6001f17e2b.png"},{"id":50143799,"identity":"7984c099-0ddb-4db8-9978-ec20d55da8db","added_by":"auto","created_at":"2024-01-25 07:41:31","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":591413,"visible":true,"origin":"","legend":"\u003cp\u003eAn Example of the Labeling Process\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-3890356/v1/af3ff855bdad5b58646c1822.png"},{"id":50144412,"identity":"f9bc06bb-7295-410a-8a0a-d7c08f415d64","added_by":"auto","created_at":"2024-01-25 07:57:31","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":59973,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion Matrix of YOLO v4 Classification Results\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-3890356/v1/0f35c46ea20aea910aad45f9.png"},{"id":50143796,"identity":"50ceb02f-5557-482a-809c-377d739a3223","added_by":"auto","created_at":"2024-01-25 07:41:31","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":62986,"visible":true,"origin":"","legend":"\u003cp\u003eConfusion Matrix of YOLO v5 Classification Results\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-3890356/v1/23aebe528db07e59fb3b17c2.png"},{"id":71327192,"identity":"41c422bc-4701-4013-b18b-292488bc392c","added_by":"auto","created_at":"2024-12-13 11:09:26","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2995964,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3890356/v1/1ccc1fb1-1bf1-4b70-bb40-b980b2ad0ea7.pdf"},{"id":50144158,"identity":"c75e81ac-8395-44f8-81d1-5fdbb713fc24","added_by":"auto","created_at":"2024-01-25 07:49:31","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":160577,"visible":true,"origin":"","legend":"","description":"","filename":"Tables1to3.docx","url":"https://assets-eu.researchsquare.com/files/rs-3890356/v1/3d5b7e7bb4c5d52e15598796.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Design and Implementation of a Microplastic Detection and Classification System Supported by Deep Learning Algorithm","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eMicroparticles are produced from substances that dissolve or erode in nature and for industrial purposes. Therefore, they can comprise various materials such as plastic, silicon, metal, additives, and color pigments. In addition, these particles can be of different sizes (O\u0026szlig;mann et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Ying et al. \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Although the damages caused by microparticles to living things and the environment are still being investigated, studies show they cause harm (Rochman et al. \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2013\u003c/span\u003e). Among these microparticles, microplastics, in particular, carry serious risks. Plastics are substances biological organisms cannot break down in nature (Crawford and Quinn \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Its size, the type of material, the paint pigments it carries, the substances it absorbs, and the additives applied to plastic increase the harm of microplastics (Karbalaei et al. \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; O\u0026szlig;mann et al. \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Strungaru et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Zou et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Issac and Kandasubramanian \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Some types have carcinogenic effects, as well as polylactic acid (PLA) produced due to the polymerization of lactic acid, which has recently been synthesized to bacteria through fermentation, and plastic types consumed by organisms in nature (Mehta et al. \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Teixeira et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Heavy metals such as lead and biological pollutants contained in some plastics can be carried by microplastics (Zou et al. \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Amelia et al. \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Table salt, drinking water, personal care and cosmetics products, synthetic textiles, and air are some vectors that can carry microplastics (Leslie \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Crawford and Quinn \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; I\u0026ntilde;iguez et al. \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Gasperi et al. \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Pivokonsky et al. \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Ustabasi and Baysal \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Bayo et al. \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). For this reason, microplastic detection is one of the phenomenal issues in recent times, and its detection is vital for humans and living life.\u003c/p\u003e \u003cp\u003eWhen the optical characteristics of microplastics are examined, it is seen that more than one optical phenomenon occurs together. Mie scattering and absorbance, emittance, reflectance, and diffraction of the particles determine the optical characteristics together. Of these relationships, the emittance is directly related to the phosphorescence in the color pigments that the microplastic can carry. When phosphorus-containing color pigments are stimulated by light, they can emit light that differs from the wavelength of the stimulating light. This light emitted by the excited substance is called emittance. In this way, information about the pigments and phosphorus-containing compounds in the substance can be obtained. Absorbance is the rate of light absorption of a substance and is one of the techniques used by UV/Vis (Ultraviolet/Visible) spectrophotometers to determine the material type.\u003c/p\u003e \u003cp\u003eOn the other hand, reflectance is defined as the light reflection rate of the material and is associated with the smoothness of the material surface and the refraction rate. Diffraction can be expressed as the wave propagation that light creates when it hits the edge of the material or when passing through smaller slits. These methods can be used to measure microplastics of diverse types and sizes, but for accurate measurements, the microplastics must have the same geometric surface form and same geometry. The optical phenomena that can occur in a spherical microplastic are given in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eOptical characteristics depend not only on the type of matter but also on the wavelength of incoming light. Light rays with shorter wavelengths are exposed to more refraction, while those with longer wavelengths are exposed to less refraction. Differences in wavelength also affect the minimum angle required for the beam to penetrate matter, known as the acceptance angle. These changes, which depend on the wavelength of light, cause the differentiation of the interference patterns that occur. Thus, the different interference patterns that occur facilitate the recognition of the substance and increase the recognition accuracy.\u003c/p\u003e \u003cp\u003eMicro-Raman and micro-FTIR devices are frequently used to detect microplastics in water. Baruah et al. examined traditional methods used to detect microplastics (Baruah et al. \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Chakraborty et al. conducted a review of studies that identified a diverse range of microplastics in diverse water sources through the use of Raman spectroscopy (Chakraborty et al. \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). These measurement devices are expensive and require the evaluation of measurements by experts. The main disadvantage of traditional methods is the requirement for expensive equipment and expert personnel for their spectral evaluation (Othman et al. \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Hence, Primpke et al. have identified microplastic types by employing micro-Raman and micro-FTIR techniques and image processing (Primpke et al. \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). However, image processing methods may overlook certain microplastics depending on the particle's shape and type.\u003c/p\u003e \u003cp\u003eArtificial intelligence-supported systems provide convenience in evaluation processes. It has gained a place for itself by achieving successful results in today's academic studies. The deep learning algorithm enables faster detection of particles carried by water while minimizing the possibility of incorrect detection by non-experts.\u003c/p\u003e \u003cp\u003eSome studies utilize traditional detection methods and deep learning algorithms to detect microplastics in soil and water (Ai et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Shishkin and Grekov \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Nevertheless, the direct use of cameras and microscopes in these detection methods does not offer a comprehensive solution, resulting in economical and practical limitations. Although larger microplastics can be eliminated through simple filtration, effectively removing small particles without specialized techniques remains challenging.\u003c/p\u003e \u003cp\u003eFast, problem-oriented, learnable, and portable systems are essential in determining the sources of microplastics, finding transport vectors, investigating their harm, and determining the consumption suitability of foodstuffs and accumulation status. Although it can be detected with existing systems, alternative systems have the advantages of low prices and high performance in automatic classification. Studies also use a deep learning algorithm for classification and customized device design for microplastic detection (Grant-Jacob et al. \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). The lack of evaluation of these studies with size, type, and particle concentration is one of the deficiencies in the studies. Detecting microplastics in low concentration and micron size gains importance because it arises from the challenge of visually detecting micro-plastics in beverages and food items.\u003c/p\u003e \u003cp\u003eDeep learning is a machine learning approach characterized by a multi-layered structure, including input, hidden, and output layers. Unlike traditional machine learning methods, deep learning models can leverage more hidden layers to extract more intricate features from data. Furthermore, one notable advantage of deep learning is its capacity to infer meaningful patterns from data autonomously. Convolutional neural networks have emerged at the intersection of computer vision and deep learning, gaining popularity for their successes in processing sound and visual information. This study involves realizing a scattered light interference-based novel microplastic detection device design and implementing automatic classification using deep learning algorithms.\u003c/p\u003e \u003cp\u003eIn the study's \u003cspan refid=\"Sec2\" class=\"InternalRef\"\u003eMaterials and Methods\u003c/span\u003e section, we will discuss sample materials, the system's hardware, optical elements, and the deep learning algorithms used. In the \u003cspan refid=\"Sec5\" class=\"InternalRef\"\u003eResults\u003c/span\u003e section, we will focus on data pre-processing and present a comparative analysis of the deep learning methods employed. The interpretation of the findings, future research directions, and the system's development will be explained in the Discussion section.\u003c/p\u003e"},{"header":"2. Materials and Methods","content":"\u003cp\u003eThe block diagram of the system we developed in this study is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, while the assembled setup can be seen in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. The system can be fundamentally examined under two main categories: hardware and software.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe \u003cspan refid=\"Sec3\" class=\"InternalRef\"\u003ehardware\u003c/span\u003e section comprises mechanical parts created by 3D printers, optical components, and embedded system hardware (Raspberry Pi and Arduino Nano). It includes the cuvette tray, screen, beam combining unit, camera, feedback sensor, switching unit, and laser drivers, all used in the system.\u003c/p\u003e \u003cp\u003eThe \u003cspan refid=\"Sec4\" class=\"InternalRef\"\u003esoftware\u003c/span\u003e section comprises embedded system components within the device and computer-based software utilizing the YOLO v4 and YOLO v5 deep learning algorithms, which uses the patterns created by the scattering and interference of laser beams of different wavelengths within a 1 cm\u003csup\u003e2\u003c/sup\u003e liquid sample cross-section. This software collects, classifies, and performs training and testing procedures on the captured images. In our study, a proposed system capable of accurately comparing the size and type of microplastics independent of concentration has been developed using low-cost components. Lasers with four different wavelengths used in laser pointers were used as light sources. These lasers have violet 405nm, red 655 nm, and green 534 nm, 807 nm dual beam wavelengths. Beams from different lasers were output from a single point using beam combiners. Hence, like small-angle scattering patterns, Mie scattering and interference patterns were observed for the incoming beams at different wavelengths. It was observed that these patterns varied depending on the material type and size, and the variations in concentration were examined. Samples with a total volume of 2.5ml each containing 10\u0026micro;m spherical polystyrene (Sigma Aldrich) with 10% solid content, 8\u0026micro;m spherical melamine (Sigma Aldrich) with 10% solid content, and 8\u0026micro;m spherical polystyrene (Sigma Aldrich) with 2% solid content were prepared. In this way, the optical characteristics of diverse types of microplastics, both of varied sizes and the same size, were examined. Different concentrations were taught to the deep learning algorithm so that quantity-dependent characteristic changes did not cause adverse effects on microplastic classification by the deep learning algorithm.\u003c/p\u003e \u003cp\u003eIn the subsequent sections, detailed information regarding both hardware and software components has been provided.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Hardware\u003c/h2\u003e \u003cp\u003eLaser diode units with various wavelengths used in generic laser pointers are light sources in the system hardware. These modules include a laser diode, driver circuit, collimator lens, and housing. The focus settings of the examined modules can also be changed. The laser modules are powered by two AAA batteries, which require a supply voltage of 3V. For this, a 3V regulator is included in the system. Since the continuous on-operation of the lasers may cause the lasers to be damaged by heating and the battery to drain rapidly, BJT (Bipolar Junction Transistor) has been added to the button part in the laser driver, and the lasers have been opened and closed electronically. In addition, it is aimed to minimize the change in the output power of the heated laser diodes by resting.\u003c/p\u003e \u003cp\u003eIn the beam-combining unit, three laser beams are combined using two beam splitter prisms for combining purposes. After attenuation, the secondary combined laser beam formed in the beam splitter at the output was measured for feedback using TSL2561. The most crucial reason for attenuating the laser beams before reaching the sensor is to prevent the sensor from reaching saturation. The reason for avoiding saturation is that a sensor that has reached saturation cannot evaluate higher values. This unit is isolated from the external environment to prevent reflection and scattering-based beam noises during beam combining.\u003c/p\u003e \u003cp\u003eArduino Nano has been used to collect feedback sensor data to control laser drivers precisely. While transferring data from Arduino Nano to Raspberry Pi via a USB connection also manages the system with the data received from Raspberry Pi.\u003c/p\u003e \u003cp\u003eRaspberry Pi provides control of the system, categorization of image data, storage of images, and transfer of images to the computer. The trained network will be loaded onto the Raspberry Pi in the following stages to enable on-site sample recognition. The management software on the Raspberry Pi is implemented with Python 3, including OpenCV libraries.\u003c/p\u003e \u003cp\u003eWhen exposed to samples containing microparticles in a cuvette, laser beams cause optical phenomena such as refraction, scattering, diffraction, and absorption. The diffraction angle, scattering amount, diffraction waves, and absorption amounts vary in different wavelengths, resulting in patterns forming at different angles. Additionally, if phosphorus-containing substances are present in the microparticles, they may cause light emitting in different wavelengths. The state of the particle emitting light also depends on the wavelength of the excitation light. When the parameters of multiple different wavelength sources are examined, it is possible to determine the type and quantity of the substance from the resulting patterns.\u003c/p\u003e \u003cp\u003eThe plastic parts used in the system are made from outputs obtained from 3D printers. Since black is the color that absorbs light the most, unwanted refraction and scattering can be absorbed.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Software\u003c/h2\u003e \u003cp\u003eThe YOLO v4 and v5 algorithms have been used as deep learning algorithms. YOLO stands for \"You Only Look Once\" because it treats the detection problem as a regression problem and requires only a single pass through the network.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe basic working principle of YOLO is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. It tries to detect the object at the center of each grid by dividing the image into grids. The detection is done by generating a vector. A vector cannot be generated if the desired object for detection is not present in the current grid, either in whole or in part. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the red grids with the central point cannot generate a vector as the desired object is absent, whereas the white grids with the central point can generate a vector.\u003c/p\u003e \u003cp\u003eLike most detectors, the YOLO v4 architecture consists of backbone, neck, and head sections. While the CSPDarknet53 network in the backbone contains many skip layers, the neck section utilizes PANet and SPP to overcome the fixed size restriction. The head section, which comprises three detection filters, must have skip connections with a doubling factor based on the number of filters. In YOLO v5, unlike YOLO v4, an up-sample layer is used, and the number of filters in the previous layer is doubled in this layer and set as the number of filters in the next layer. This allows for deeper features to be extracted.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\n \u003ch2\u003e3.1 Pre-Processing\u003c/h2\u003e\n \u003cp\u003eSince we focus on supervised machine learning, the image data must be presented to the computer system. For this purpose, labeling was performed using the makesense.ai program(Skalski), an open-source tool. An example of the labeling process is shown in Fig. \u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e. The resulting label includes information about the target object\u0026apos;s class and its location on the screen. This enables the computer to generate more precise results than unsupervised learning (Lim et al. \u003cspan class=\"CitationRef\"\u003e2000\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eLow, medium, and high concentration images obtained from 10 \u0026micro;m polystyrene material are provided in Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e, while images obtained from 8 \u0026micro;m melamine material are presented in Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e, and images obtained from 8 \u0026micro;m polystyrene material are shown in Table \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e. According to the results, as the concentration increases, noticeable clarifications become visible in the outer rings, and the beam spreads throughout the photograph. All automatic camera settings were fixed when capturing the images. In particular, the default automatic white balance (AWB) in the camera settings can cause the loss of meaningful data. To reduce the error rate, 20 images were captured, and the arithmetic average of their matrices was calculated to reconstruct the images. Barrel distortion, lens errors, perspective, and scale adjustments were rectified using the OpenCV geometric transformations libraries. This allowed for a more controlled examination of the images.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Classification Experiments\u003c/h2\u003e \u003cp\u003eAfter pre-processing, the results obtained from the proposed device were labeled, making the data ready for training the deep learning network. YOLO v4 and YOLO v5 deep learning algorithms were utilized in the classification experiments, and the classification performance results were compared.\u003c/p\u003e \u003cp\u003eThe system was tested using 10 \u0026micro;m polystyrene, 8 \u0026micro;m melamine, and 8 \u0026micro;m polystyrene particles. From each sample, 1\u0026ndash;10 \u0026micro;l of microplastics were taken, and 30 samples were prepared for each group using deionized water, resulting in a total volume of 2.5 ml. In total, images were taken from 90 different samples to minimize errors that can occur when working with small volumes using a micropipette. For each sample, images were taken 20 times using three different laser beams. From each spectrometer cuvette, there were 60 images, resulting in 600 images per sample set and 1800 images from the same group of samples. In total, 5400 image data points were processed for the three types of microplastics.\u003c/p\u003e \u003cp\u003eThree thousand one hundred two images were allocated for training, 1149 for validation, and 1149 for testing. Since there is a need for three-class detection, which poses a multi-classification problem, categorical cross-entropy was chosen as the loss function. Following the popularity in previous deep learning studies (Mitliagkas et al. \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Xie et al. \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Granziol et al. \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), a momentum of 0.9, weight decay of 0.0005, and a learning rate of 0.00013 were determined. The optimization algorithm used by the backpropagation algorithm, which aims to bring the predicted values closer to the expected values, was determined as Stochastic Gradient Descent with Momentum (SGDM). The training was conducted by applying 100 epochs for the YOLO v5 algorithm and 2000 epochs for the YOLO v4 algorithm.\u003c/p\u003e \u003cp\u003eThe overall confusion matrix resulting from the testing process of the models generated for YOLO v4 is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e. According to the matrix, the YOLO v4 network misclassified two images of 10 \u0026micro;m polystyrene, six images of 8 \u0026micro;m melamine, and eight images of 8 \u0026micro;m polystyrene.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe testing process was also conducted on the YOLO v5 algorithm. Despite using fewer epochs, YOLO v5 successfully achieved error-free classification. The resulting confusion matrix for the YOLO v5 classification is provided in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eInstead of using overall accuracy that considers all values, the aim is to obtain a more consistent result using the F1 score, which is the harmonic mean of precision and recall values. In some research studies using deep learning algorithms, it has been observed that the F1 score is preferred over the accuracy value (Yacouby and Axman \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparative Results of YOLO Algorithms\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlgorithm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF1 Score\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eYOLO v4\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e%96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e%96\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e%100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e%98\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eYOLO v5\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e%100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e%100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e%100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e%100\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e presents the comparison metric results of two different deep networks trained on the same dataset, calculated by counting the TP (True Positive), FP (False Positive), and FN (False Negative) values across all test images. As evident from Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the YOLO v5 network outperformed the YOLO v4 network in all three metrics.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e presents the detection results of the deployed networks in terms of class and image. As seen from Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, the YOLO v5 network successfully detected all 383 test image classes without any false or missed detections. On the other hand, the YOLO v4 network incorrectly detected two images from the 10 \u0026micro;m polystyrene class, six images from the eight \u0026micro;m melamine class, and eight images from the 8 \u0026micro;m polystyrene class.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eClass-Based TP, FP and FN Results\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"10\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlgorithm\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003ePolystyrene 10\u0026micro;m\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cp\u003eMelamine 8\u0026micro;m\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c10\" namest=\"c8\"\u003e \u003cp\u003ePolystyrene 8\u0026micro;m\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eFP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eFN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eTP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eFP\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eFN\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eYOLO v4\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e381\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e377\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e375\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eYOLO v5\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e383\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e383\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e383\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"4. Discussion and Conclusions","content":"\u003cp\u003eAs a result, a low-cost and portable sensor has been developed that learns micro-plastics in water with a deep learning algorithm and classifies them with high accuracy. The system, which can classify microplastics in terms of size and type, is thought to be at a level that can distinguish other microparticles. The light source in the realized sensor system is used by combining three different laser beams utilizing a beam combiner, just like in portable laser projector technology.\u003c/p\u003e \u003cp\u003eAs in portable laser projector devices, the combination of laser beams with three different wavelengths with micromirror and microlens technologies and the fact that they can be realized in smaller sizes are among the advantages that can be provided in the future.\u003c/p\u003e \u003cp\u003eWhen turned into a flow-type sensor, rapid detection of the contaminant mixture in the water network can be achieved and used as an early warning system in potable water networks.\u003c/p\u003e \u003cp\u003eWhile the correct classification success of the system was determined as 98% due to the classification of the YOLO v4 algorithm, up to 100% success was achieved with the YOLO v5 algorithm. While the exceptionally low level of human contribution in these classification systems minimizes the human errors that can be made, it can contribute to the acceleration of measurement processes by making preliminary determinations for experts in laboratory environments.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eConceptualization, E.D.; methodology, E.D.; software, E.D.; validation, E.D.; formal analysis, E.D.; investigation, E.D.; resources, E.D.; data curation, E.D.; writing\u0026mdash;original draft preparation, E.D.; writing\u0026mdash;review and editing, E.D.; visuali-zation, E.D.; supervision, R.K.; project administration, R.K. All authors have read and agreed to the manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAi W, Chen G, Yue X, Wang J (2023) Application of hyperspectral and deep learning in farmland soil microplastic detection. J Hazard Mater 445:130568. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jhazmat.2022.130568\u003c/span\u003e\u003cspan address=\"10.1016/j.jhazmat.2022.130568\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmelia TSM, Khalik WMAWM, Ong MC et al (2021) Marine microplastics as vectors of major ocean pollutants and its hazards to the marine ecosystem and humans. Prog Earth Planet Sci 8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s40645-020-00405-4\u003c/span\u003e\u003cspan address=\"10.1186/s40645-020-00405-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaruah A, Sharma A, Sharma S, Nagraik R (2022) An insight into different microplastic detection methods. Int J Environ Sci Technol 19:5721\u0026ndash;5730. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s13762-021-03384-1\u003c/span\u003e\u003cspan address=\"10.1007/s13762-021-03384-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBayo J, Ramos B, L\u0026oacute;pez-Castellanos J et al (2022) Lack of Evidence for Microplastic Contamination from Water-Soluble Detergent Capsules. Microplastics 1:121\u0026ndash;140. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/microplastics1010008\u003c/span\u003e\u003cspan address=\"10.3390/microplastics1010008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChakraborty I, Banik S, Biswas R et al (2023) Raman spectroscopy for microplastic detection in water sources: a systematic review. Int J Environ Sci Technol.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCrawford CB, Quinn B (2017) The emergence of plastics. In: Microplastic Pollutants\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGasperi J, Wright SL, Dris R et al (2018) Microplastics in air: Are we breathing it in? Curr Opin Environ Sci Heal 1:1\u0026ndash;5. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.coesh.2017.10.002\u003c/span\u003e\u003cspan address=\"10.1016/j.coesh.2017.10.002\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrant-Jacob JA, Xie Y, Mackay BS et al (2019) Particle and salinity sensing for the marine environment via deep learning using a raspberry pi. Environ Res Commun 1:35001. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1088/2515-7620/ab14c9\u003c/span\u003e\u003cspan address=\"10.1088/2515-7620/ab14c9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGranziol D, Zohren S, Roberts S (2022) Learning Rates as a Function of Batch Size: A Random Matrix Theory Approach to Neural Network Training. J Mach Learn Res\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eI\u0026ntilde;iguez ME, Conesa JA, Fullana A (2017) Microplastics in Spanish Table Salt. Sci Rep 7:1\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-017-09128-x\u003c/span\u003e\u003cspan address=\"10.1038/s41598-017-09128-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIssac MN, Kandasubramanian B (2021) Effect of microplastics in water and aquatic systems. Environ Sci Pollut Res 28:19544\u0026ndash;19562. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11356-021-13184-2\u003c/span\u003e\u003cspan address=\"10.1007/s11356-021-13184-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKarbalaei S, Hanachi P, Walker TR, Cole M (2018) Occurrence, sources, human health impacts and mitigation of microplastic pollution. Environ Sci Pollut Res 25:36046\u0026ndash;36063. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11356-018-3508-7\u003c/span\u003e\u003cspan address=\"10.1007/s11356-018-3508-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeslie HA (2014) Review of Microplastics in Cosmetics. IVM Inst Environ Stud 476:33\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLim TS, Loh WY, Shih YS (2000) Comparison of prediction accuracy, complexity, and training time of thirty-three old and new classification algorithms. Mach Learn. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1023/A:1007608224229\u003c/span\u003e\u003cspan address=\"10.1023/A:1007608224229\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMehta R, Kumar V, Bhunia H, Upadhyay SN (2005) Synthesis of poly(lactic acid): A review. J Macromol Sci - Polym Rev 45:325\u0026ndash;349. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/15321790500304148\u003c/span\u003e\u003cspan address=\"10.1080/15321790500304148\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMitliagkas I, Zhang C, Hadjis S, Re C (2017) Asynchrony begets momentum, with an application to deep learning. In: 54th Annual Allerton Conference on Communication, Control, and Computing, Allerton 2016\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eO\u0026szlig;mann BE, Sarau G, Holtmannsp\u0026ouml;tter H et al (2018) Small-sized microplastics and pigmented particles in bottled mineral water. Water Res 141:307\u0026ndash;316. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.watres.2018.05.027\u003c/span\u003e\u003cspan address=\"10.1016/j.watres.2018.05.027\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOthman AM, Elsayed AA, Sabry YM et al (2022) Detection of Sub-20 \u0026micro;m Microplastic Particles by Attenuated Total Reflection Fourier Transform Infrared Spectroscopy and Comparison with Raman Spectroscopy. ACS Omega. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1021/acsomega.2c07998\u003c/span\u003e\u003cspan address=\"10.1021/acsomega.2c07998\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePivokonsky M, Cermakova L, Novotna K et al (2018) Occurrence of microplastics in raw and treated drinking water. Sci Total Environ 643:1644\u0026ndash;1651. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.scitotenv.2018.08.102\u003c/span\u003e\u003cspan address=\"10.1016/j.scitotenv.2018.08.102\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePrimpke S, Lorenz C, Rascher-Friesenhausen R, Gerdts G (2017) An automated approach for microplastics analysis using focal plane array (FPA) FTIR microscopy and image analysis. Anal Methods 9:1499\u0026ndash;1511. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1039/c6ay02476a\u003c/span\u003e\u003cspan address=\"10.1039/c6ay02476a\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRochman CM, Hoh E, Hentschel BT, Kaye S (2013) Classify plastic waste as hazardous (types of externalities caused by consumption of plastic bags). Environ Sci Technol 494:169\u0026ndash;171\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShishkin IE, Grekov AN (2023) Implementation of YOLOv5 for Detection and Classification of Microplastics and Microorganisms in Marine Environment. Proc \u0026ndash;\u0026thinsp;2023 Int Russ Smart Ind Conf SmartIndustryCon 2023 230\u0026ndash;235. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/SmartIndustryCon57312.2023.10110736\u003c/span\u003e\u003cspan address=\"10.1109/SmartIndustryCon57312.2023.10110736\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSkalski P GitHub - SkalskiP/make-sense: Free to use online tool for labelling photos. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://makesense.ai\u003c/span\u003e\u003cspan address=\"https://makesense.ai\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. https://github.com/SkalskiP/make-sense. Accessed 20 Jun 2023\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStrungaru SA, Jijie R, Nicoara M et al (2019) Micro- (nano) plastics in freshwater ecosystems: Abundance, toxicological impact and quantification methodology. TrAC - Trends Anal Chem 110:116\u0026ndash;128. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.trac.2018.10.025\u003c/span\u003e\u003cspan address=\"10.1016/j.trac.2018.10.025\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTeixeira S, Eblagon KM, Miranda F et al (2021) Towards Controlled Degradation of Poly(lactic) Acid in Technical Applications. C 7:42. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/c7020042\u003c/span\u003e\u003cspan address=\"10.3390/c7020042\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUstabasi GS, Baysal A (2019) Occurrence and risk assessment of microplastics from various toothpastes. Environ Monit Assess 191:1\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10661-019-7574-1\u003c/span\u003e\u003cspan address=\"10.1007/s10661-019-7574-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXie Z, Sato I, Sugiyama M (2020) Stable Weight Decay Regularization. arXiv\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYacouby R, Axman D (2020) Probabilistic Extension of Precision, Recall, and F1 Score. for More Thorough Evaluation of Classification Models\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYing Z, Shengyan P, Xue L et al (2020) Global trends and prospects in microplastics research: A bibliometric analysis. J Hazard Mater 123110. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jhazmat.2020.123110\u003c/span\u003e\u003cspan address=\"10.1016/j.jhazmat.2020.123110\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZou J, Liu X, Zhang D, Yuan X (2020) Adsorption of three bivalent metals by four chemical distinct microplastics. Chemosphere 248:126064. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.chemosphere.2020.126064\u003c/span\u003e\u003cspan address=\"10.1016/j.chemosphere.2020.126064\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables 1 to 3 are available in the Supplementary Files section.\u003c/p\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":"Microplastics, Optical Sensors, Scattering Analyses, Instruments, Deep Learning, Embedded Systems, Optoelectronic, Microplastic detection","lastPublishedDoi":"10.21203/rs.3.rs-3890356/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3890356/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMicroparticles are challenging to detect due to their small size and can harm living things when exposed. Especially microplastics are one of the harmful microparticles. For this reason, detecting microplastics in a vital consumer item such as water is essential. Machine learning in the detection method allows the learning of different types and sizes of microplastics, allowing such systems to work unremittingly in real time. The present study has designed a low-budget, high-accuracy device with a deep learning algorithm that can autonomously classify microplastics according to their size and type. Three lasers with dual beam wavelengths of 405nm, 655nm, and 534nm-807nm, frequently used in laser pointers, are light sources in the sensor. The beams formed by the lasers were combined employing a beam combiner, allowing beams to emerge from a single point. Classification success of up to 100% has been achieved, thanks to the different interference patterns of light sources of various wavelengths. 10\u0026micro;m polystyrene, 8\u0026micro;m polystyrene, and 8\u0026micro;m melamine prepared in different constancy were used as samples in the experiments.\u003c/p\u003e","manuscriptTitle":"Design and Implementation of a Microplastic Detection and Classification System Supported by Deep Learning Algorithm","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-25 07:41:26","doi":"10.21203/rs.3.rs-3890356/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":"b516e9a4-800b-4a08-a13d-1ac398bc48f6","owner":[],"postedDate":"January 25th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-12-13T11:09:03+00:00","versionOfRecord":[],"versionCreatedAt":"2024-01-25 07:41:26","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3890356","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3890356","identity":"rs-3890356","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}