Microplastic intelligent identification, positioning and size calculation system based on artificial intelligence

By using SVMD and SO-SloEn feature extraction methods and KPE-YOLOv5 algorithms, the problem of low accuracy of feature extraction and object detection in the intelligent recognition and positioning system of microplastics is solved, and the accuracy of microplastic particles with small size and complex shapes is achieved. Identification and real-time processing improve the robustness and efficiency of the system.

CN118351113BActive Publication Date: 2024-12-06XIAMEN HAIHONGXING INSTR

Patent Information

Application Number
CN202410772741.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-06-17
Publication Date
2024-12-06
Estimated Expiration
2044-06-17

AI Technical Summary

Technical Problem

The existing microplastic intelligent identification, positioning and size calculation system has problems such as low accuracy, slow efficiency and poor robustness in feature extraction and object detection, especially the poor identification of microplastic particles with small sizes and complex shapes, resulting in the system Low performance.

Method used

Feature extraction is performed using S-RNS feature extraction methods based on SVMD and SO-SloEn, and boundary and shape positioning are combined with the KPE-YOLOv5 algorithm to improve the identification accuracy and processing efficiency of microplastic particles.

Benefits of technology

Improve the identification and positioning accuracy of microplastic particles, realize real-time or near-real-time processing, enhance the robustness and generalization capabilities of the system, and are suitable for microplastic identification tasks in different scenarios, reduce errors, and improve System efficiency and accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118351113B_ABST
    Figure CN118351113B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of artificial intelligence, and specifically refers to an artificial intelligence-based microplastic intelligent identification, positioning and size calculation system, comprising a data acquisition module, a data preprocessing module, a feature extraction module, an intelligent identification module, a positioning and segmentation module, a size calculation module and a result display and output module. The present invention proposes to use an S-RNS feature extraction method based on SVMD and SO-SloEn to extract features from preprocessed data, thereby improving the accuracy of identifying and positioning microplastic particles; using a KPE-YOLOv5 algorithm to determine the boundaries and shapes of microplastic particles, accurately positioning the boundary box and segmenting the shape of the microplastic particles, thereby achieving accurate identification of the microplastic particles, maintaining a high accuracy rate while achieving real-time or near real-time processing, thereby improving the work efficiency of the artificial intelligence-based microplastic intelligent identification, positioning and size calculation system, reducing errors and improving accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and specifically refers to an artificial intelligence-based microplastic intelligent identification, positioning and size calculation system. Background Art

[0002] With the rapid development of artificial intelligence technology, image recognition, target detection and size calculation technologies have been widely used in the field of environmental monitoring. The existing feature extraction methods have the problem of low accuracy in identifying and positioning microplastic particles, which makes it impossible to capture key information such as the complex texture and shape of microplastic particles. The feature representation is not sufficient, and highly correlated features are generated, which leads to increased feature redundancy. At the same time, microplastic particles of different sizes and shapes cannot be effectively identified and located, resulting in low system recognition accuracy, efficiency and robustness. Traditional target detection algorithms have the problem of low recognition accuracy, especially for small-sized and complex-shaped microplastic particles. The recognition accuracy is easily affected, the processing speed is slow, and the recognition performance of microplastic particles in new scenes or complex environments is poor, resulting in poor performance in terms of accuracy, processing speed, generalization ability and algorithm complexity of the microplastic intelligent identification, positioning and size calculation system, affecting the practicality of the system. Summary of the invention

[0003] In view of the above situation, the present invention provides an artificial intelligence-based microplastic intelligent identification, positioning and size calculation system. The existing feature extraction methods have the problem of low accuracy in identifying and positioning microplastic particles, which makes it impossible to capture the complex texture, shape and other key information of microplastic particles, and the feature representation is not sufficient. This scheme proposes to use the S-RNS feature extraction method based on SVMD and SO-SloEn to extract features from the preprocessed data, so as to better capture the texture, shape and other key information of microplastic particles, thereby improving the accuracy of identifying and positioning microplastic particles. The SO-SloEn method combines non-subsampled Contourlet transform and second-order analysis method, and can extract the features of microplastic particles at multiple scales. The accuracy, efficiency and robustness of the system are improved. In view of the problem that traditional target detection algorithms have low recognition accuracy, especially for microplastic particles of small size and complex shape, the recognition accuracy is easily affected and the processing speed is slow. This scheme proposes to use the KPE-YOLOv5 algorithm to determine the boundaries and shapes of microplastic particles, accurately locate the bounding box and segment the shape of microplastic particles, so as to achieve accurate recognition of microplastic particles, maintain high accuracy while achieving real-time or near real-time processing, and have strong generalization ability and adaptability, which can be applied to microplastic particle identification tasks in different scenarios, improving the work efficiency of the microplastic intelligent identification, positioning and size calculation system based on artificial intelligence, reducing errors and improving accuracy.

[0004] The microplastic intelligent identification, positioning and size calculation system based on artificial intelligence provided by the present invention includes a data acquisition module, a data preprocessing module, a feature extraction module, an intelligent identification module, a positioning and segmentation module, a size calculation module and a result display and output module, specifically including the following contents:

[0005] The data acquisition module collects samples to be tested, uses a camera to collect image and video data from the samples to be tested, and annotates them;

[0006] The data preprocessing module preprocesses the collected image and video data, including denoising, grayscale conversion, and contrast enhancement, to obtain preprocessed data;

[0007] The feature extraction module extracts features related to microplastics from the preprocessed data, and extracts features from the preprocessed data by using the S-RNS feature extraction method based on SVMD and SO-SloEn to obtain feature-extracted data;

[0008] The intelligent recognition module uses machine learning to train a microplastic recognition model, uses the microplastic recognition model to classify the features extracted by the feature extraction module, identifies whether there are microplastic particles in the image, and obtains a recognition result;

[0009] The positioning and segmentation module determines the boundary and shape of the microplastic particles through the KPE-YOLOv5 algorithm according to the recognition results to obtain the positioning results;

[0010] The size calculation module calculates the diameter, area and perimeter of the microplastic particles according to the boundaries and shapes of the microplastic particles using image processing technology to obtain the size results of the microplastic particles;

[0011] The result display and output module displays the recognition results, positioning results and size results in a visual form and outputs them to the database.

[0012] Furthermore, the feature extraction module extracts features related to microplastics from the preprocessed image, specifically including the following steps:

[0013] Step S1: Boundary extraction, using the Canny edge detection algorithm to extract the boundaries of microplastic particles;

[0014] Step S2: SVMD feature extraction, using the SVMD method to describe the data points on the boundary of the microplastic particles, and extracting the shape features of the microplastic particles by finding the hyperplane that maximizes the distance between the boundary and the data;

[0015] Step S3: SO-SloEn feature extraction, using the SO-SloEn method to describe the internal data of the boundary of the microplastic particles, capture the texture distribution and structural information inside the boundary, and extract the texture features of the microplastic particles;

[0016] Step S4: color feature extraction, using the color space conversion method to extract the color features of microplastic particles;

[0017] Step S5: feature fusion, using the feature weighting method to fuse the shape features, texture features and color features of the extracted microplastic particles to obtain a feature vector of multi-faceted information;

[0018] Further, step S2 specifically includes the following steps:

[0019] Step S21: data point selection, uniformly selecting a set of data points on the boundary of the microplastic particles;

[0020] Step S22: data point conversion, representing the selected data points in the form of feature vectors, each data point is represented as a multi-dimensional vector, which includes the position information of the data point in the image coordinate system;

[0021] Step S23: constructing an SVDD model, using the SVDD method to construct an SVDD model that can describe the distribution of data points on the boundary, and describing the distribution of the data by finding the smallest ball or smallest hypersphere that can surround the data points on the boundary;

[0022] Step S24: shape feature extraction, using the SVDD model to extract the geometric shape information of the boundary of the microplastic particles, including the shape, size and curvature of the boundary;

[0023] Step S25: extracting hyperplane parameters, extracting hyperplane parameters from the SVDD model, including the normal vector and the intercept, and combining the geometric shape information of the microplastic particle boundary to obtain the shape characteristics of the microplastic particle;

[0024] Further, step S3 specifically includes the following steps:

[0025] Step S31: selecting data inside the boundary, using a uniform sampling method to select a group of data points from the inside of the boundary of the microplastic particles as the data points inside the boundary to describe the texture characteristics of the microplastic particles;

[0026] Step S32: data representation, representing the selected boundary internal data points in the form of feature vectors, each data point is represented as a multi-dimensional vector containing texture information, and the texture information includes gray value, color and xx;

[0027] Step S33: constructing a SO-SloEn model, using the selected boundary internal data points to construct the SO-SloEn model, setting the goal of the SO-SloEn model to capture the texture distribution and structural information inside the data points, and measuring the correlation degree and structural characteristics between the data points based on the energy function;

[0028] Step S34: Extracting texture distribution and structural information, using the SO-SloEn model to extract the texture distribution and structural information of microplastic particles;

[0029] Step S35: Texture feature extraction, using a local binary pattern algorithm to describe the texture information inside the boundary of the microplastic particles and extract texture features;

[0030] Further, step S33 specifically includes the following steps:

[0031] Step S331: Initialization, setting the initialization range of the SO-SloEn threshold ;

[0032] Step S332: Set parameters and initialize the parameters of SO-SloEn, including individuals, population size, and maximum number of iterations;

[0033] Step S333: Calculate the fitness, calculate the fitness function of each individual, and the formula used is as follows:

[0034] ;

[0035] in, For the The fitness function of an individual is is the number of misclassified samples, is the number of correctly classified samples;

[0036] Step S334: Find the best individual, and find the individual with the best fitness in the population;

[0037] Step S335: Output the result. When the maximum number of iterations is reached, the optimization process is completed and the optimal value is output. ;

[0038] Furthermore, the boundaries and shapes of microplastic particles are determined by the KPE-YOLOv5 algorithm in the positioning and segmentation module, which specifically includes the following steps:

[0039] Step K1: Data preparation: collect image datasets containing microplastic particles and annotate them, including the bounding boxes and key point locations of the microplastic particles, to obtain annotated image datasets;

[0040] Step K2: Model selection, select the YOLOv5 model as the basic architecture, use the K-means++ clustering algorithm to optimize the anchor box, integrate the scSE attention module into the backbone network, and add the prediction feature layer of the shallow network to build a positioning model;

[0041] Step K3: Target detection and key point estimation: for the data extracted from the feature extraction module, the KPE-YOLOv5 algorithm is used to perform target detection and key point estimation to determine the bounding box and key point positions of the microplastic particles;

[0042] Step K4: key point connection and shape recognition, splicing the detected key points, identifying the shape of microplastic particles, and further extracting the specific characteristics of microplastic particles;

[0043] Step K5: Post-processing and output, post-processing the bounding boxes and key point positions of the detected microplastic particles, removing duplicate detections and optimizing the bounding boxes, and finally outputting the boundaries and shape information of the determined microplastic particles;

[0044] Step K6: Evaluation and tuning: Evaluate and tune the algorithm according to actual needs to improve the accuracy and performance of microplastic particle boundary and shape recognition;

[0045] Furthermore, step K2 specifically includes the following contents:

[0046] K-means++ clustering algorithm optimizes anchor box: a random sample point is selected from the annotated image data set as the first initialized cluster center, and the possibility of each point becoming the next cluster centroid is determined by calculating the shortest distance between each sample point. The formula used is as follows:

[0047] ;

[0048] in, is the sample point in the image dataset, is the index of the sample, are all sample points in the image dataset, Represents the square of the distance from each sample point to the nearest cluster center. For sample The probability of being selected as the next cluster centroid, is the total number of sample points;

[0049] choose The highest point is used as the next cluster center, and the calculation is repeated until k cluster centers are selected. For each sample point in the data set , calculate the distance from each sample point to the K cluster centers, assign the sample point to the class corresponding to the cluster center with the smallest distance, and update the cluster center until the position of the cluster center no longer changes;

[0050] scSE attention module: including spatial compression channel excitation sSE module and channel compression spatial excitation cSE module, adding scSE attention module to the third layer of the backbone network;

[0051] Among them, the cSE module introduces the attention mechanism from the channel perspective, weights the features of each channel starting from optimizing the channel dimension, and inputs the feature map, which is , including channels, each channel corresponds to a feature map, and the feature map is globally embedded into the vector through the global pooling layer. In the above example, the number of channels is kept unchanged and the feature map is transformed from The spatial dimension is converted to a 1×1 vector , the formula used is as follows:

[0052] ;

[0053] in, represents the Kth value in the vector after global pooling, represents the height and width of the feature map, is the pixel index in the vertical and horizontal directions of the feature map, is the first The channels are located in the image Eigenvalue at pixel location, vector By including weights and The fully connected layer is processed, and then the ReLU activation function and sigmoid normalization are processed to obtain the i-th channel The feature importance of , The formula used to calculate the value of is as follows:

[0054] ;

[0055] in, represents the feature importance vector obtained after the full connection layer, ReLU activation function and Sigmoid normalization. is the sigmoid function, is the ReLU function, and is the weight function of the fully connected layer, Represents the vector before being processed by the fully connected layer;

[0056] Prediction feature layer: Add a small target detection layer to detect shallower feature maps.

[0057] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0058] (1) The existing feature extraction methods have the problem of low accuracy in identifying and locating microplastic particles, which makes it impossible to capture the complex texture, shape and other key information of microplastic particles, and the feature representation is not sufficient. This scheme proposes to use the S-RNS feature extraction method based on SVMD and SO-SloEn to extract features from the preprocessed data, so as to better capture the texture, shape and other key information of microplastic particles, thereby improving the accuracy of identifying and locating microplastic particles. The SO-SloEn method combines the non-subsampled Contourlet transform and the second-order analysis method, which can extract the features of microplastic particles at multiple scales, thereby improving the accuracy, efficiency and robustness of the system.

[0059] (2) The traditional target detection algorithm has the problem of low recognition accuracy, especially for microplastic particles of small size and complex shape, the recognition accuracy is easily affected and the processing speed is slow. This scheme proposes to use the KPE-YOLOv5 algorithm to determine the boundary and shape of microplastic particles, accurately locate the boundary box and segment the shape of microplastic particles, so as to achieve accurate recognition of microplastic particles, maintain a high accuracy rate while achieving real-time or near real-time processing. At the same time, it has strong generalization ability and adaptability, and can be applied to microplastic particle recognition tasks in different scenarios, which improves the work efficiency of the microplastic intelligent recognition, positioning and size calculation system based on artificial intelligence, reduces errors and improves accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of a module of an artificial intelligence-based microplastic intelligent identification, positioning and size calculation system provided by the present invention;

[0061] Figure 2 Schematic diagram of the process for extracting features related to microplastics;

[0062] Figure 3 is a schematic flow chart of step S2;

[0063] Figure 4 is a schematic flow chart of step S3;

[0064] Figure 5 is a flow chart of step S33;

[0065] Figure 6 This is a flowchart of the KPE-YOLOv5 algorithm.

[0066] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. Implementation

[0067] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0068] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0069] Example 1, see Figure 1 The microplastic intelligent identification, positioning and size calculation system based on artificial intelligence provided by the present invention includes a data acquisition module, a data preprocessing module, a feature extraction module, an intelligent identification module, a positioning and segmentation module, a size calculation module and a result display and output module, specifically including the following contents:

[0070] The data acquisition module collects samples to be tested, uses a camera to collect image and video data from the samples to be tested, and annotates them;

[0071] The data preprocessing module preprocesses the collected image and video data, including denoising, grayscale conversion, and contrast enhancement, to obtain preprocessed data;

[0072] The feature extraction module extracts features related to microplastics from the preprocessed data, and extracts features from the preprocessed data by using the S-RNS feature extraction method based on SVMD and SO-SloEn to obtain feature-extracted data;

[0073] The intelligent recognition module uses machine learning to train a microplastic recognition model, uses the microplastic recognition model to classify the features extracted by the feature extraction module, identifies whether there are microplastic particles in the image, and obtains a recognition result;

[0074] The positioning and segmentation module determines the boundary and shape of the microplastic particles through the KPE-YOLOv5 algorithm according to the recognition results to obtain the positioning results;

[0075] The size calculation module calculates the diameter, area and perimeter of the microplastic particles according to the boundaries and shapes of the microplastic particles using image processing technology to obtain the size results of the microplastic particles;

[0076] The result display and output module displays the recognition results, positioning results and size results in a visual form and outputs them to the database.

[0077] Example 2, see Figure 2 and Figure 3 This embodiment is based on the above embodiment, and the feature extraction module extracts features related to microplastics from the preprocessed image, specifically including the following steps:

[0078] Step S1: Boundary extraction, using the Canny edge detection algorithm to extract the boundaries of microplastic particles;

[0079] Step S2: SVMD feature extraction, using the SVMD method to describe the data points on the boundary of the microplastic particles, and extracting the shape features of the microplastic particles by finding the hyperplane that maximizes the distance between the boundary and the data;

[0080] Step S2 specifically includes the following steps:

[0081] Step S21: data point selection, uniformly selecting a set of data points on the boundary of the microplastic particles;

[0082] Step S22: data point conversion, representing the selected data points in the form of feature vectors, each data point is represented as a multi-dimensional vector, which includes the position information of the data point in the image coordinate system;

[0083] Step S23: constructing an SVDD model, using the SVDD method to construct an SVDD model that can describe the distribution of data points on the boundary, and describing the distribution of the data by finding the smallest ball or smallest hypersphere that can surround the data points on the boundary;

[0084] Step S24: shape feature extraction, using the SVDD model to extract the geometric shape information of the boundary of the microplastic particles, including the shape, size and curvature of the boundary;

[0085] Step S25: extracting hyperplane parameters, extracting hyperplane parameters from the SVDD model, including the normal vector and the intercept, and combining the geometric shape information of the microplastic particle boundary to obtain the shape characteristics of the microplastic particle;

[0086] Step S3: SO-SloEn feature extraction, using the SO-SloEn method to describe the internal data of the boundary of the microplastic particles, capture the texture distribution and structural information inside the boundary, and extract the texture features of the microplastic particles;

[0087] Step S4: color feature extraction, using the color space conversion method to extract the color features of microplastic particles;

[0088] Step S5: Feature fusion: Use the feature weighting method to fuse the shape features, texture features and color features of the extracted microplastic particles to obtain a feature vector of multi-faceted information.

[0089] The existing feature extraction methods have the problem of low accuracy in identifying and locating microplastic particles, which makes it impossible to capture the complex texture, shape and other key information of microplastic particles, and the feature representation is not sufficient. This scheme proposes to use the S-RNS feature extraction method based on SVMD and SO-SloEn to extract features from the preprocessed data, so as to better capture the texture, shape and other key information of microplastic particles, thereby improving the accuracy of identifying and locating microplastic particles. The SO-SloEn method combines the non-subsampled Contourlet transform and the second-order analysis method, which can extract the features of microplastic particles at multiple scales, thereby improving the accuracy, efficiency and robustness of the system.

[0090] Example 3, see Figure 4 and Figure 5 This embodiment is based on the above embodiment, and step S3 specifically includes the following steps:

[0091] Step S31: selecting data inside the boundary, using a uniform sampling method to select a group of data points from the inside of the boundary of the microplastic particles as the data points inside the boundary to describe the texture characteristics of the microplastic particles;

[0092] Step S32: data representation, representing the selected boundary internal data points in the form of feature vectors, each data point is represented as a multi-dimensional vector containing texture information, and the texture information includes gray value, color and xx;

[0093] Step S33: constructing a SO-SloEn model, using the selected boundary internal data points to construct the SO-SloEn model, setting the goal of the SO-SloEn model to capture the texture distribution and structural information inside the data points, and measuring the correlation degree and structural characteristics between the data points based on the energy function;

[0094] Step S33 specifically includes the following steps:

[0095] Step S331: Initialization, setting the initialization range of the SO-SloEn threshold

[0096] Step S332: Set parameters and initialize the parameters of SO-SloEn, including individuals, population size, and maximum number of iterations;

[0097] Step S333: Calculate the fitness, calculate the fitness function of each individual, and the formula used is as follows:

[0098] ;

[0099] in, For the The fitness function of an individual is is the number of misclassified samples, is the number of correctly classified samples;

[0100] Step S334: Find the best individual, and select the individual with the best fitness in the population;

[0101] Step S335: Output the result. When the maximum number of iterations is reached, the optimization process is completed and the optimal value is output. ;

[0102] Step S34: Extracting texture distribution and structural information, using the SO-SloEn model to extract the texture distribution and structural information of microplastic particles;

[0103] Step S35: Texture feature extraction, using the local binary pattern algorithm to describe the texture information inside the boundary of the microplastic particles and extract the texture features.

[0104] Example 4, see Figure 6 Based on the above embodiment, this embodiment determines the boundary and shape of microplastic particles by using the KPE-YOLOv5 algorithm in the positioning and segmentation module, and specifically includes the following steps:

[0105] Step K1: Data preparation: collect image datasets containing microplastic particles and annotate them, including the bounding boxes and key point locations of the microplastic particles, to obtain annotated image datasets;

[0106] Step K2: Model selection, select the YOLOv5 model as the basic architecture, use the K-means++ clustering algorithm to optimize the anchor box, integrate the scSE attention module into the backbone network, and add the prediction feature layer of the shallow network to build a positioning model;

[0107] Step K3: Target detection and key point estimation: for the data extracted from the feature extraction module, the KPE-YOLOv5 algorithm is used to perform target detection and key point estimation to determine the bounding box and key point positions of the microplastic particles;

[0108] Step K4: key point connection and shape recognition, splicing the detected key points, identifying the shape of microplastic particles, and further extracting the specific characteristics of microplastic particles;

[0109] Step K5: Post-processing and output, post-processing the bounding boxes and key point positions of the detected microplastic particles, removing duplicate detections and optimizing the bounding boxes, and finally outputting the boundaries and shape information of the determined microplastic particles;

[0110] Step K6: Evaluation and tuning: Evaluate and tune the algorithm according to actual needs to improve the accuracy and performance of microplastic particle boundary and shape recognition;

[0111] Embodiment 5: This embodiment is based on the above embodiment, and step K2 specifically includes the following contents:

[0112] K-means++ clustering algorithm optimizes anchor box: a random sample point is selected from the annotated image data set as the first initialized cluster center, and the possibility of each point becoming the next cluster centroid is determined by calculating the shortest distance between each sample point. The formula used is as follows:

[0113] ;

[0114] in, is the sample point in the image dataset, is the index of the sample, are all sample points in the image dataset, Represents the square of the distance from each sample point to the nearest cluster center. For sample The probability of being selected as the next cluster centroid, is the total number of sample points;

[0115] choose The highest point is used as the next cluster center, and the calculation is repeated until k cluster centers are selected. For each sample point In the data set, calculate the distance between each sample point and K cluster centers, assign the sample point to the class corresponding to the cluster center with the smallest distance, and update the cluster center until the position of the cluster center no longer changes;

[0116] scSE attention module: including spatial compression channel excitation sSE module and channel compression spatial excitation cSE module, adding scSE attention module to the third layer of the backbone network;

[0117] Among them, the cSE module introduces the attention mechanism from the channel perspective, weights the features of each channel starting from optimizing the channel dimension, and inputs the feature map, which is , including channels, each channel corresponds to a feature map, and the feature map is globally embedded into the vector through the global pooling layer. In the above example, the number of channels is kept unchanged and the feature map is transformed from The spatial dimension is converted to a 1×1 vector , the formula used is as follows:

[0118] ;

[0119] in, represents the Kth value in the vector after global pooling, represents the height and width of the feature map, is the pixel index in the vertical and horizontal directions of the feature map, is the first The channels are located in the image Eigenvalue at pixel location, vector By including weights and The fully connected layer is processed, and then the ReLU activation function and sigmoid normalization are processed to obtain the i-th channel The feature importance of , The formula used to calculate the value of is as follows:

[0120] ;

[0121] in, represents the feature importance vector obtained after the full connection layer, ReLU activation function and Sigmoid normalization. is the sigmoid function, is the ReLU function, and is the weight function of the fully connected layer, Represents the vector before being processed by the fully connected layer;

[0122] Prediction feature layer: Add a small target detection layer to detect shallower feature maps.

[0123] In view of the problem that traditional target detection algorithms have low recognition accuracy, especially for microplastic particles of small size and complex shape, the recognition accuracy is easily affected and the processing speed is slow. This scheme proposes to use the KPE-YOLOv5 algorithm to determine the boundaries and shapes of microplastic particles, accurately locate the bounding box and segment the shape of microplastic particles, so as to achieve accurate recognition of microplastic particles, maintain high accuracy while achieving real-time or near real-time processing, and have strong generalization ability and adaptability, which can be applied to microplastic particle identification tasks in different scenarios, improving the work efficiency of the microplastic intelligent identification, positioning and size calculation system based on artificial intelligence, reducing errors and improving accuracy.

[0124] Embodiment 5: This embodiment is based on the above embodiment. The data acquisition module collects samples to be detected, and the camera models used are Logitech C920 and Sony Alpha a6000.

[0125] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0126] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0127] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An artificial intelligence-based microplastic intelligent identification, positioning and size calculation system, characterized by: It includes data acquisition module, data preprocessing module, feature extraction module, intelligent recognition module, positioning and segmentation module, size calculation module and result display and output module. Includes the following: The data acquisition module collects samples to be tested, uses a camera to collect image and video data from the samples to be tested, and annotates them; The data preprocessing module preprocesses the collected image and video data, including denoising, graying, and contrast enhancement, to obtain preprocessed data; The feature extraction module extracts features related to microplastics from the preprocessed data, and extracts features from the preprocessed data by using the S-RNS feature extraction method based on SVMD and SO-SloEn to obtain feature-extracted data; The intelligent recognition module uses machine learning to train a microplastic recognition model, uses the microplastic recognition model to classify the features extracted by the feature extraction module, identifies whether there are microplastic particles in the image, and obtains a recognition result; The positioning and segmentation module determines the boundary and shape of the microplastic particles through the KPE-YOLOv5 algorithm according to the recognition results to obtain the positioning results; The size calculation module calculates the diameter, area and perimeter of the microplastic particles according to the boundaries and shapes of the microplastic particles using image processing technology to obtain the size results of the microplastic particles; The result display and output module displays the recognition results, positioning results and size results in a visual form and outputs them to the database; The feature extraction module extracts features related to microplastics from the preprocessed images, specifically including the following steps: Step S1: Boundary extraction, using the Canny edge detection algorithm to extract the boundaries of microplastic particles; Step S2: SVMD feature extraction, using the SVMD method to describe the data points on the boundary of the microplastic particles, and extracting the shape features of the microplastic particles by finding the hyperplane that maximizes the distance between the boundary and the data; Step S2 specifically includes the following steps: Step S21: data point selection, uniformly selecting a set of data points on the boundary of the microplastic particles; Step S22: data point conversion, representing the selected data points in the form of feature vectors, each data point is represented as a multi-dimensional vector, which includes the position information of the data point in the image coordinate system; Step S23: constructing an SVDD model, using the SVDD method to construct an SVDD model that can describe the distribution of data points on the boundary, and describing the distribution of the data by finding the smallest ball or smallest hypersphere that can surround the data points on the boundary; Step S24: shape feature extraction, using the SVDD model to extract the geometric shape information of the boundary of the microplastic particles, including the shape, size and curvature of the boundary; Step S25: extracting hyperplane parameters, extracting hyperplane parameters from the SVDD model, including the normal vector and the intercept, and combining the geometric shape information of the microplastic particle boundary to obtain the shape characteristics of the microplastic particle; Step S3: SO-SloEn feature extraction, using the SO-SloEn method to describe the internal data of the boundary of the microplastic particles, capture the texture distribution and structural information inside the boundary, and extract the texture features of the microplastic particles; Step S3 specifically includes the following steps: Step S31: selecting data inside the boundary, using a uniform sampling method to select a group of data points from the inside of the boundary of the microplastic particles as the data points inside the boundary to describe the texture characteristics of the microplastic particles; Step S32: data representation, representing the selected data points inside the boundary in the form of feature vectors, each data point being represented as a multidimensional vector containing texture information, the texture information including grayscale value and color; Step S33: constructing a SO-SloEn model, using the selected boundary internal data points to construct the SO-SloEn model, setting the goal of the SO-SloEn model to capture the texture distribution and structural information inside the data points, and measuring the correlation degree and structural characteristics between the data points based on the energy function; Step S33 specifically includes the following steps: Step S331: Initialization, setting the initialization range of the SO-SloEn threshold ; Step S332: Set parameters and initialize the parameters of SO-SloEn, including individuals, population size, and maximum number of iterations; Step S333: Calculate the fitness, calculate the fitness function of each individual, and the formula used is as follows: ; in, For the The fitness function of an individual is is the number of misclassified samples, is the number of correctly classified samples; Step S334: Find the best individual, and find the individual with the best fitness in the population; Step S335: Output the result. When the maximum number of iterations is reached, the optimization process is completed and the optimal value is output. ; Step S34: Extracting texture distribution and structural information, using the SO-SloEn model to extract the texture distribution and structural information of microplastic particles; Step S35: Texture feature extraction, using a local binary pattern algorithm to describe the texture information inside the boundary of the microplastic particles and extract texture features; Step S4: color feature extraction, using the color space conversion method to extract the color features of microplastic particles; Step S5: feature fusion, using the feature weighting method to fuse the shape features, texture features and color features of the extracted microplastic particles to obtain a feature vector of multi-faceted information; In the positioning and segmentation module, the boundaries and shapes of microplastic particles are determined by the KPE-YOLOv5 algorithm, which specifically includes the following steps: Step K1: Data preparation: collect image datasets containing microplastic particles and annotate them, including the bounding boxes and key point locations of the microplastic particles, to obtain annotated image datasets; Step K2: Model selection, select the YOLOv5 model as the basic architecture, use the K-means++ clustering algorithm to optimize the anchor box, integrate the scSE attention module into the backbone network, and add the prediction feature layer of the shallow network to build a positioning model; Step K2 specifically includes the following contents: K-means++ clustering algorithm optimizes anchor box: a random sample point is selected from the annotated image data set as the first initialized cluster center, and the possibility of each point becoming the next cluster centroid is determined by calculating the shortest distance between each sample point. The formula used is as follows: ; in, is a sample point in the image dataset, is the index of the sample, are all sample points in the image dataset, Represents the square of the distance from each sample point to the nearest cluster center. For sample The probability of being selected as the next cluster centroid, is the total number of sample points; choose The highest point is used as the next cluster center, and the calculation is repeated until k cluster centers are selected. For each sample point in the data set , calculate the distance from each sample point to the K cluster centers, assign the sample point to the class corresponding to the cluster center with the smallest distance, and update the cluster center until the position of the cluster center no longer changes; scSE attention module: including spatial compression channel excitation sSE module and channel compression spatial excitation cSE module, adding scSE attention module to the third layer of the backbone network; Among them, the cSE module introduces the attention mechanism from the channel perspective, weights the features of each channel starting from optimizing the channel dimension, and inputs the feature map, which is , including channels, each channel corresponds to a feature map, and the feature map is globally embedded into the vector through the global pooling layer. In the above example, the number of channels is kept unchanged and the feature map is transformed from The spatial dimension is converted to a 1×1 vector , the formula used is as follows: ; in, represents the Kth value in the vector after global pooling, represents the height and width of the feature map, is the pixel index in the vertical and horizontal directions of the feature map, is the first The channels are located in the image Eigenvalue at pixel location, vector By including weights and The fully connected layer is processed, and then the ReLU activation function and sigmoid normalization are processed to obtain the i-th channel The feature importance of , The formula used to calculate the value of is as follows: ; in, represents the feature importance vector obtained after the full connection layer, ReLU activation function and Sigmoid normalization. is the sigmoid function, is the ReLU function, and is the weight function of the fully connected layer, Represents the vector before being processed by the fully connected layer; Prediction feature layer: Add a small target detection layer to detect shallower feature maps; Step K3: Target detection and key point estimation: for the data extracted from the feature extraction module, the KPE-YOLOv5 algorithm is used to perform target detection and key point estimation to determine the bounding box and key point positions of the microplastic particles; Step K4: key point connection and shape recognition, splicing the detected key points, identifying the shape of microplastic particles, and further extracting the specific characteristics of microplastic particles; Step K5: Post-processing and output, post-processing the bounding boxes and key point positions of the detected microplastic particles, removing duplicate detections and optimizing the bounding boxes, and finally outputting the boundaries and shape information of the determined microplastic particles; Step K6: Evaluation and tuning, evaluate and tune the algorithm according to actual needs to improve the accuracy and performance of microplastic particle boundary and shape recognition.

Citation Information

Patent Citations

  • Method and system for identifying and detecting colored micro-plastics in soil based on deep learning

    CN116363440A

  • Waste plastic detection method based on improved M2Det

    CN116433599A

Cited By

  • Microplastic intelligent identification and measurement method based on machine vision

    CN121214432A

  • A micro-plastic intelligent identification and measurement method based on machine vision

    CN121214432B