A method for identifying and classifying microplastics in water samples based on deep learning
By employing deep learning methods, the problems of high reliance on manual labor and low efficiency in microplastic detection have been solved, achieving efficient and automated microplastic identification and classification, adapting to different water samples and imaging conditions, and suitable for various water body scenarios.
Patent Information
- Application Number
- CN202610415404.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-31
- Publication Date
- 2026-06-26
AI Technical Summary
Existing technologies for microplastic detection rely on manual observation, which is time-consuming and labor-intensive, with low automation, low recognition accuracy, and poor generalization ability, making it difficult to meet the high-throughput and standardized detection needs in different scenarios.
We employ a deep learning-based approach, which involves image data acquisition and adaptive preprocessing, saliency detection, lightweight network construction, and multi-scale feature fusion, combined with transfer learning and data augmentation, to achieve automated identification and classification of microplastics.
It has achieved full automation of microplastic detection, improving efficiency by more than 10 times, with high identification accuracy, adapting to different water samples and imaging conditions, and reducing detection costs and overall environmental monitoring costs.
Smart Images
Figure CN122290090A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of microplastic identification technology, specifically to a method for identifying and classifying microplastics in water samples based on deep learning. Background Technology
[0002] Currently, the detection and identification of microplastic samples in aquatic environments mainly rely on traditional manual observation and conventional digital image processing methods. In actual testing, laboratory personnel typically use optical microscopes to manually count, classify, and statistically analyze the samples. This method is not only time-consuming, labor-intensive, and physically demanding, but it is also prone to missed or false detections due to visual fatigue and subjective judgment differences, making it difficult to guarantee data consistency and reliability.
[0003] In terms of automated detection, existing technologies mostly employ image segmentation based on fixed thresholds, morphological filtering, edge extraction, and manually designed features (such as size, shape, grayscale, and texture) to achieve microplastic identification. These methods heavily rely on manually preset parameters and sample imaging quality. When microplastic particles are adhered, overlapping, or irregularly shaped, or when the water sample background is complex and contains bubbles, impurities, or filter membrane texture interference, problems such as target breakage, artifacts, and noise-induced misjudgments easily occur, resulting in low identification accuracy. Furthermore, the entire process requires significant manual intervention, including repeated parameter adjustments, manual result correction, and manual category classification, leading to low automation, low detection efficiency, and highly subjective results. In addition, existing algorithms and parameters are typically customized for specific water samples and imaging conditions, exhibiting poor generalization and adaptability. Changing the water type, sampling environment, or microscopic imaging conditions necessitates readjusting parameters or even reconstructing the algorithm, making it difficult to meet the high-throughput, standardized, and intelligent detection needs of microplastics in different scenarios. Summary of the Invention
[0004] This invention provides a deep learning-based method for identifying and classifying microplastics in water samples, addressing the technical problems in traditional microplastic detection such as high reliance on manual labor, low detection efficiency, strong background interference, easy omission of small particles, high annotation costs, and poor model generalization.
[0005] To address the aforementioned problems, this invention provides a method for identifying and classifying microplastics in water samples based on deep learning, employing the following technical solution: Includes the following steps: S1. Image Data Acquisition and Adaptive Preprocessing: Acquire raw microscopic images containing microplastics, perform color space conversion, brightness component enhancement, and color normalization on the raw images to eliminate background differences and highlight the characteristics of microplastic particles.
[0006] S2. Determine if the current process is in the training phase. If yes, execute steps S3-S6; otherwise, execute step S6 directly.
[0007] S3. Semi-automatic dataset construction based on saliency detection: A saliency detection algorithm is used to generate potential target candidate regions for microplastics. Threshold segmentation and noise removal are performed on the candidate regions to generate initial pseudo-labels. After manual verification and correction, the dataset size is expanded through data augmentation.
[0008] S4. Construct a lightweight microplastic detection network: Build a target detection model based on convolutional neural networks. The model includes a backbone feature extraction network, a spatial attention mechanism module, and a multi-scale feature fusion module, which are used to extract microplastic features and adapt to the detection requirements of microplastics of different sizes.
[0009] S5. Model Training and Optimization: The pre-trained weights are loaded using a transfer learning strategy, and combined with online HSV domain enhancement and loss function optimization, the detection model is trained to obtain a high-precision microplastic detection model.
[0010] S6. Inference and Post-processing Screening: The trained detection model is used to predict new input microscopic images. Through confidence screening, non-maximum suppression, and optional morphological / geometric screening, accurate identification and classification of microplastic particles are achieved.
[0011] Furthermore, in S1, the adaptive preprocessing specifically includes: S11, Color Space Conversion: Converts RGB images to HSV or LAB color space, separating the luminance and color components.
[0012] S12, Luminance Component Enhancement: Only the luminance component is subjected to limited contrast adaptive histogram equalization to enhance the local contrast between microplastic particles and the background and suppress low-frequency noise in the background.
[0013] S13. Color Standardization: Eliminate background color shifts caused by different shooting environments through grayscale world algorithms or white balance correction, and unify the background features of training images.
[0014] Furthermore, in S3, the semi-automatic dataset construction specifically includes: S31. Saliency Target Extraction: A saliency detection algorithm based on spectral residuals or local contrast is used to automatically generate candidate regions for potential microplastic targets.
[0015] S32. Pseudo-label generation: Adaptive binarization is performed on the candidate region, and morphological opening and closing operations are used to remove small noise such as bubbles and dust to generate initial pseudo-labels.
[0016] S33. Data Augmentation: Perform geometric transformation, photometric transformation, and Mosaic enhancement on the manually verified and corrected image. The geometric transformation includes random rotation, flipping, and cropping. The photometric transformation includes random adjustment of brightness, contrast, and saturation.
[0017] Furthermore, in S4, the lightweight microplastic detection network specifically comprises: S41. The backbone feature extraction network adopts a lightweight network such as CSPDarknet or MobileNet.
[0018] S42. The spatial attention mechanism module is embedded at the end of the feature extraction layer to guide the network to focus on high-contrast pixel areas and suppress interference from complex backgrounds.
[0019] S43. The multi-scale feature fusion module adopts an FPN+PAN structure, which integrates shallow detail features and deep semantic features to meet the detection needs of microplastics of different sizes.
[0020] S44. The target detection model is a lightweight version of the YOLO series or SSD.
[0021] Furthermore, in S5, model training and optimization specifically include: S51. Pre-trained weight loading: Load weights pre-trained on the COCO large general dataset and utilize the learned edge and texture general features.
[0022] S52, HSV Domain Online Enhancement: During the training iteration, random HSV tone shifts and saturation perturbations are applied to the input image to enable the model to learn the essential morphological features of microplastics.
[0023] S53. Loss Function Optimization: Use CIoU or DIoU as the bounding box regression loss function to improve the accuracy of detection box localization.
[0024] Furthermore, in S6, post-processing filtering specifically includes: S61, Confidence Filtering: Set a confidence threshold to filter out low-probability targets.
[0025] S62, Non-maximum suppression (NMS): Removes redundant detection boxes with excessive overlap, retaining the optimal detection position.
[0026] S63. Optional morphological / geometric screening: Set aspect ratio or area thresholds based on the physical properties of microplastics to remove non-microplastic interferences such as long fibers and extremely small noise.
[0027] Furthermore, the microplastics have a size range of 20μm-5mm, which is compatible with microplastic detection standards in the field of water monitoring.
[0028] The beneficial effects of the deep learning-based method for identifying and classifying microplastics in water samples provided by this invention are: This invention automates the entire process of image preprocessing, recognition, classification, and statistics through deep learning, completely replacing traditional manual visual inspection. It completes the analysis of a single image in seconds, improving efficiency by more than 10 times compared to manual methods, and enabling high-throughput, batch detection. At the same time, it uses saliency detection to assist semi-automatic annotation, significantly reducing the workload of annotation. Combined with transfer learning and data augmentation, it effectively solves the pain points of scarce microplastic samples, difficult annotation, and difficult model training, significantly reducing the overall cost of environmental monitoring.
[0029] This invention effectively eliminates interference from uneven illumination, color shift, and complex background noise through adaptive preprocessing, highlighting the contour features of microplastics. Combined with lightweight networks, spatial attention mechanisms, and multi-scale feature fusion, it can accurately identify microplastics of different sizes and morphologies ranging from 20μm to 5mm. Further multi-stage post-processing and screening significantly reduce false detections of impurities and missed detections of small particles, enabling the model to maintain high recognition accuracy and consistent detection performance under different water samples and imaging conditions. The data reliability is far superior to traditional methods.
[0030] This invention is not a simple improvement on existing technologies, but rather a deep integration of adaptive preprocessing, semi-automatic annotation, lightweight model design, transfer learning, and targeted enhancement to form a robust, efficient, and data-friendly complete solution. The lightweight model is easy to deploy and adaptable to various water body scenarios and microplastic types, requiring no frequent parameter tuning for universal applicability. The system can automatically output standardized data such as particle size distribution and classification statistics, directly supporting water environment assessment and drinking water safety monitoring. It boasts low deployment costs, wide applicability, and high practical value and promising prospects for widespread adoption. Attached Figure Description
[0031] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein: Figure 1 This is a logic block diagram of the present invention. Detailed Implementation
[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Those skilled in the art should understand that the embodiments described below are only some, not all, of the embodiments disclosed. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0033] The principles and spirit of the present invention will be explained in detail below with reference to several representative embodiments.
[0034] Example 1 of a deep learning-based method for identifying and classifying microplastics in water samples provided by this invention: In this embodiment, the microplastic size ranges from 20μm to 5mm, which is compatible with the microplastic detection standards in the field of water monitoring. Specifically, it is applied to the high-throughput detection of microplastics in filter membrane samples from a tributary of the Yangtze River by an environmental testing laboratory, and includes the following steps: The hardware and software facilities in this embodiment include: 1. Imaging equipment: An Olympus BX53 upright optical microscope was used, equipped with a color CCD camera. A 10× objective lens was selected to capture microscopic images of the filter membrane samples, and the image resolution was set to 1920×1080 pixels.
[0035] 2. Computing Platform: Workstation equipped with NVIDIA GTX 3080 GPU and 16GB of memory to ensure high efficiency in model training and inference.
[0036] 3. Based on the Python 3.8 environment, integrate libraries such as OpenCV 4.5.5, PyTorch 1.12.0, and Ultralytics YOLOv5, and write an automatic microplastic detection system according to the process of this invention.
[0037] This embodiment specifically includes: S1. Image data acquisition and adaptive preprocessing.
[0038] Specifically, for image acquisition, the filter membrane samples from the Yangtze River tributaries were placed on the microscope stage, and 50 fields of view were randomly selected for shooting, resulting in 50 original RGB microscopic images, which included microplastics of different shapes (fragments, fibers, particles) and sizes, as well as background interferences (bubbles, dust, filter membrane fibers).
[0039] Adaptive preprocessing specifically includes: S11, Color Space Conversion: Convert the original RGB image to LAB color space, separate the L (luminance) component, A (red-green difference) component and B (blue-yellow difference) component, and perform subsequent enhancement processing only on the luminance component to avoid interference from the color components.
[0040] S12 Luminance Component Enhancement: The L component is processed by Limit Contrast Adaptive Histogram Equalization (CLAHE) with clipLimit=2.5 and tileGridSize=(16,16) to enhance the local contrast between microplastic particles and the background, suppress low-frequency noise in the background, and highlight the edge features of microplastics.
[0041] S13 Color Standardization: The grayscale world algorithm is used to correct the color of the A and B components, eliminating the differences in light source color temperature at different shooting times and the background color deviation caused by the filter film, so that the background brightness and tone of the 50 original images are unified.
[0042] S2. In this embodiment, the training phase process is executed first. After the model is built and trained, the inference phase process is executed. That is, steps S3-S5 are executed first, and then step S6 is executed.
[0043] S3. Semi-automatic dataset construction based on saliency detection.
[0044] Using 300 training images as the target, the specific operation is as follows: S31. Randomly select 40 images from the 50 preprocessed images, and use a saliency detection algorithm based on spectral residuals to calculate the saliency value of each region in the image, generate candidate regions for potential microplastic targets, and initially locate the pixel range of suspected microplastics.
[0045] S32. Pseudo-label generation: Adaptive binarization is performed on the candidate region, and small noise such as bubbles and dust is removed by morphological opening and closing operations to generate initial pseudo-labels (including the bounding box and category label of the candidate region).
[0046] The researchers manually verified and corrected the false labels in the system interface, eliminating falsely detected labels and supplementing missed targets. At the same time, they collected samples of interfering substances such as bubbles and filter membrane impurities, and finally obtained 40 high-quality labeled images.
[0047] S33 Data Augmentation: Data augmentation is performed on 40 manually verified and corrected images and their labels, specifically including: Geometric transformations: random rotation (±30°), horizontal and vertical flipping, random cropping (cropping ratio 0.8-1.0); Photometric adjustment: Randomly adjust brightness (±10%), contrast (±15%), and saturation (±20%). Advanced enhancement: Mosaic enhancement is used to randomly stitch and merge four images, increasing background complexity and target scale diversity; After augmentation, the dataset was expanded to 300 images, which served as the base data for both the training and validation sets.
[0048] S4: Construct a lightweight microplastic detection network.
[0049] Specifically as follows: S41. Backbone Feature Extraction Network: The CSPDarknet-slim lightweight network is selected as the backbone feature extraction network to simplify the convolutional layer parameters, reduce the computational cost and parameter count of the model, and adapt to the inference and training needs of the laboratory workstation.
[0050] S42. Spatial Attention Mechanism Module: A spatial attention mechanism module is embedded at the end of the backbone feature extraction network. This module learns the weight matrix to guide the network to focus on high-contrast microplastic pixel regions, suppress interference from complex backgrounds (such as filter membrane textures and impurities), and enhance the microplastic feature extraction effect.
[0051] S43. Multi-scale feature fusion module: The FPN+PAN structure is used as the multi-scale feature fusion module to fuse shallow features (including microplastic detail features, such as edges and textures) and deep semantic features (including microplastic category semantic information) to adapt to the detection needs of microplastics of different sizes from 20μm to 5mm and avoid the loss of features of small-sized microplastics.
[0052] S44. Overall architecture of the object detection model: The lightweight backbone network, spatial attention module, and multi-scale feature fusion module are combined to build a lightweight YOLOv5-slim object detection model. This model is a lightweight version of the YOLO series, which balances detection accuracy and inference speed.
[0053] S5. Model Training and Optimization.
[0054] S51. Loading pre-trained weights: Load YOLOv5-slim weights pre-trained on the COCO large general dataset, and use general features such as edges and textures learned from the COCO dataset to accelerate model convergence and improve the efficiency of microplastic feature extraction.
[0055] S52, Online HSV Domain Augmentation: 300 augmented images were divided into a training set (240 images) and a validation set (40 images) in an 8:2 ratio. During training iterations, the images in the input training set were subjected to random HSV hue shifts (±15°-±30°) and saturation perturbations (±10%), enabling the model to learn the essential morphological characteristics of microplastics rather than relying on specific color features, thus improving the model's adaptability to microplastics with different staining and lighting conditions.
[0056] S53. Loss Function Optimization: CIoU is adopted as the bounding box regression loss function. Compared with the traditional IoU loss, CIoU considers the overlapping area of the bounding box, the distance between the center points and the aspect ratio at the same time, which improves the positioning accuracy of the microplastic detection box.
[0057] The initial learning rate was set to 0.001, a cosine annealing learning rate decay strategy was adopted, 100 epochs were trained, the batch size was set to 16, and the AdamW optimizer was selected.
[0058] S6. Reasoning and post-processing filtering.
[0059] The remaining 10 unlabeled microscopic images of filter membranes in the Yangtze River tributaries, after adaptive preprocessing in step S1, are input into the trained lightweight microplastic detection model for post-processing screening.
[0060] S61. Confidence Filtering: Set the confidence threshold to 0.3-0.45 to filter out low-probability targets with a confidence level below 0.3, thus avoiding false detections.
[0061] S62. Non-maximum suppression (NMS): Set the IoU threshold to 0.5 to remove redundant detection boxes with excessive overlap, retain the optimal detection position for each microplastic, and avoid duplicate counting.
[0062] S63. Morphological / Geometric Screening: Set thresholds based on the physical properties of microplastics to eliminate non-microplastic interfering substances. Among them, the area threshold is set to a lower limit of 50 pixels to eliminate extremely small noise. Aspect Ratio Threshold: Set the aspect ratio range to 0.2-10.0 to remove interference targets such as long strip fibers and irregular extreme shapes.
[0063] After processing using the method described in this embodiment, the system automatically outputs the quantity, location coordinates, and particle size distribution of microplastics in 30 microscopic images. After manual verification, the microplastic identification accuracy reached 94.2%, and the classification accuracy reached 88.5%. The detection efficiency is 15-20 times higher than traditional manual microscope counting, achieving high-throughput, accurate identification and classification of microplastics in water samples, meeting the detection standards requirements in the field of water body monitoring.
[0064] This invention provides a second embodiment of a deep learning-based method for identifying and classifying microplastics in water samples, which differs from embodiment 1 in that: The implementation scenarios are different.
[0065] This embodiment is applied to the detection of microplastics in the raw water of a waterworks. The test object is a filter membrane sample of raw Yellow River water at the water intake of the waterworks.
[0066] Step S1 parameter adjustment: The color space conversion uses the HSV color space. When enhancing the luminance component, CLAHE's clipLimit=2.0 and tileGridSize=(16,16). Color normalization uses a white balance correction algorithm to eliminate image color deviation caused by the turbidity of the raw water.
[0067] Step S3 Data Augmentation: Add MixUp data augmentation to fuse microplastic features from different images, further improving the model's ability to recognize low-contrast microplastics.
[0068] Step S4 Network Architecture: The backbone feature extraction network uses the MobileNetV3 lightweight network, the spatial attention mechanism module uses CBAM (Convolutional Block Attention Module), and the multi-scale feature fusion module uses the PANet structure.
[0069] Step S5 Loss Function: DIoU is used as the bounding box regression loss function, the initial learning rate is set to 0.001, the number of training epochs is 100, and the batch size is set to 8.
[0070] Step S6 Geometric screening: Increase the roundness threshold (≥0.5) to further eliminate irregularly shaped impurities.
[0071] This embodiment ultimately achieves a microplastic identification accuracy of 93.5% and a classification accuracy of 87.2% in raw water of waterworks, meeting the microplastic detection requirements for raw water quality monitoring and providing technical support for drinking water safety.
[0072] This invention provides a third embodiment of a deep learning-based method for identifying and classifying microplastics in water samples, which differs from embodiment 1 in that: The implementation scenarios are different.
[0073] It was applied to the detection of microplastics in the effluent of a wastewater treatment plant, and the test object was the filter membrane sample of the secondary sedimentation tank of the wastewater treatment plant.
[0074] Step S3: Significance Detection: A saliency detection algorithm based on local contrast is adopted, which is more suitable for scenarios where the contrast between microplastics and the background is low in wastewater effluent samples.
[0075] Step S5 Training Strategy: A combination of transfer learning and semi-supervised learning is adopted. First, pre-training is completed using 200 labeled samples, and then semi-supervised fine-tuning is performed using unlabeled effluent samples to improve the model's adaptability to microplastics in wastewater treatment effluent.
[0076] Post-processing step S6: Add particle size classification screening to classify microplastics into three grades: 20μm-50μm, 50μm-100μm, and 100μm-5mm, to meet the statistical requirements of microplastic particle size distribution in wastewater treatment plants.
[0077] This embodiment ultimately achieves a microplastic identification accuracy of 90.5% and a particle size classification accuracy of 84.3% in the effluent of wastewater treatment plants, providing accurate data support for the optimization of wastewater treatment processes.
[0078] In addition, in the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.
Claims
1. A method for identifying and classifying microplastics in water samples based on deep learning, characterized in that, Includes the following steps: S1. Image data acquisition and adaptive preprocessing: Acquire the original microscopic image containing microplastics, perform color space conversion, brightness component enhancement and color normalization on the original image, eliminate background differences and highlight the characteristics of microplastic particles; S2. Determine whether the current process is in the training phase. If yes, execute steps S3-S6; otherwise, execute step S6 directly. S3. Semi-automatic dataset construction based on saliency detection: The saliency detection algorithm is used to generate potential target candidate regions for microplastics. Threshold segmentation and noise removal are performed on the candidate regions to generate initial pseudo-labels. After manual verification and correction, the dataset size is expanded through data augmentation. S4. Construct a lightweight microplastic detection network: Build a target detection model based on convolutional neural networks. The model includes a backbone feature extraction network, a spatial attention mechanism module, and a multi-scale feature fusion module, which are used to extract microplastic features and adapt to the detection requirements of microplastics of different sizes. S5. Model Training and Optimization: The pre-trained weights are loaded using a transfer learning strategy, and combined with online HSV domain enhancement and loss function optimization, the detection model is trained to obtain a high-precision microplastic detection model. S6. Inference and Post-processing Screening: The trained detection model is used to predict new input microscopic images. Through confidence screening, non-maximum suppression, and optional morphological / geometric screening, accurate identification and classification of microplastic particles are achieved.
2. The method for identifying and classifying microplastics in water samples based on deep learning according to claim 1, characterized in that, In S1, adaptive preprocessing specifically includes: S11, Color Space Conversion: Converts RGB images to HSV or LAB color space, separating the luminance and color components; S12, Luminance Component Enhancement: Only the luminance component is subjected to limited contrast adaptive histogram equalization to enhance the local contrast between microplastic particles and the background and suppress low-frequency noise in the background. S13. Color Standardization: Eliminate background color shifts caused by different shooting environments through grayscale world algorithms or white balance correction, and unify the background features of training images.
3. The method for identifying and classifying microplastics in water samples based on deep learning according to claim 1, characterized in that, In S3, the semi-automatic dataset construction specifically includes: S31. Saliency Target Extraction: A saliency detection algorithm based on spectral residuals or local contrast is used to automatically generate candidate regions for potential microplastic targets. S32. Pseudo-label generation: Adaptive binarization is performed on the candidate region, and small noise such as bubbles and dust is removed by morphological opening and closing operations to generate initial pseudo-labels. S33. Data Augmentation: Perform geometric transformation, photometric transformation, and Mosaic enhancement on the manually verified and corrected image. The geometric transformation includes random rotation, flipping, and cropping. The photometric transformation includes random adjustment of brightness, contrast, and saturation.
4. The method for identifying and classifying microplastics in water samples based on deep learning according to claim 1, characterized in that, In S4, the lightweight microplastic detection network is specifically as follows: S41. The backbone feature extraction network adopts a lightweight network such as CSPDarknet or MobileNet. S42. The spatial attention mechanism module is embedded at the end of the feature extraction layer to guide the network to focus on high-contrast pixel areas and suppress interference from complex backgrounds. S43. The multi-scale feature fusion module adopts an FPN+PAN structure, which integrates shallow detail features and deep semantic features to meet the detection needs of microplastics of different sizes. S44. The target detection model is a lightweight version of the YOLO series or SSD.
5. The method for identifying and classifying microplastics in water samples based on deep learning according to claim 1, characterized in that, In S5, model training and optimization specifically include: S51, Pre-trained weight loading: Load weights pre-trained on the COCO large general dataset and utilize the learned edge and texture general features; S52, HSV Domain Online Enhancement: During the training iteration, random HSV tone shifts and saturation perturbations are applied to the input image to enable the model to learn the essential morphological features of microplastics. S53. Loss Function Optimization: Use CIoU or DIoU as the bounding box regression loss function to improve the accuracy of detection box localization.
6. The method for identifying and classifying microplastics in water samples based on deep learning according to claim 1, characterized in that, In S6, post-processing filtering specifically includes: S61, Confidence Filtering: Set a confidence threshold to filter out low-probability targets; S62. Non-maximum suppression: Removes redundant detection boxes with excessive overlap and retains the optimal detection position; S63. Optional morphological / geometric screening: Set aspect ratio or area thresholds based on the physical properties of microplastics to remove non-microplastic interferences such as long fibers and extremely small noise.
7. The method for identifying and classifying microplastics in water samples based on deep learning according to claim 1, characterized in that, The microplastics are 20μm-5mm in size, which is compatible with microplastic detection standards in the field of water monitoring.