A microplastic aging process traceability and factor intelligent identification system and method

Abstract

A system and method for tracing the source of microplastic aging process and intelligently identifying factors, the system: the sample conveying and primary screening device includes a sample receiving tank, a vibrating screen and a conveyor belt; the particle dispersion and electrostatic dust removal device includes a dispersing stirrer, a drying pipeline, a dry air ejector and an electrostatic eliminator; the microplastic enrichment device includes a cyclone separator and a collection tank; the microplastic separation unit includes an electrostatic separation channel, a high-voltage electrostatic charging device and a microplastic collection device; the microplastic identification and aging detection unit includes a high-resolution scanning electron microscope, a Fourier transform infrared spectrometer and an aging characteristic multimodal analysis module. Methods: rough sample selection; drying, dust removal and anti-static treatment; secondary selection; multi-level sorting; obtaining the surface morphology image and infrared spectrum of the sample; constructing a prediction model for tracing the source of microplastic aging process and identifying factors; and real-time analysis and prediction of microplastics. The system and method can realize the identification and analysis of microplastics.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent identification technology, and specifically relates to a system and method for tracing the source of microplastic aging process and intelligent identification of factors. Background Art

[0002] Microplastics are widely present in the global environment. They refer to plastic particles with a particle size of less than 5 mm, including original particles produced by industrial production and tiny fragments produced by the decomposition of plastic waste. With the rapid development of industry, a large amount of plastic waste is continuously generated and continues to enter the natural world. At the same time, microplastics continue to accumulate in the ocean, rivers, soil, air and other environments, and have become a global pollution problem that threatens the ecological environment and human health. Due to the tiny size, complex composition and wide range of sources of microplastics, it is extremely challenging to accurately detect and track them. However, the detection, separation and analysis of microplastics in environmental samples are of great significance for assessing the impact of environmental pollution, identifying pollution sources and formulating governance strategies.

[0003] Existing microplastic detection equipment and methods mainly focus on type identification and material composition analysis, making it difficult to effectively track the changes in microplastics under the influence of different aging factors. Under the long-term influence of different environmental factors such as ultraviolet radiation, thermal oxidation, and mechanical wear, the surface characteristics and chemical structure of microplastics will change significantly, and the specific action modes and proportions of these aging factors are usually difficult to accurately quantify. Due to the lack of accurate judgment equipment and methods for the degree of influence of aging factors, the application of existing technologies in tracing the source of microplastic pollution is limited, and it is difficult to fully track the diffusion path of microplastics in the environment and the evolution of their aging state.

[0004] In order to effectively solve the above problems, it is urgently necessary to provide an identification system and method that can achieve in-depth tracking of the aging behavior of microplastics and identify the degree of effect of different aging factors, so as to improve the traceability of microplastic pollution. Summary of the invention

[0005] In response to the problems existing in the above-mentioned prior art, the present invention provides a system and method for tracing the source of microplastic aging process and intelligently identifying factors. The system has reliable performance and high identification accuracy. It can realize intelligent identification and analysis of microplastics, and help to quickly and accurately infer the aging sources and possible environmental pathways of microplastics. The method can effectively distinguish the intensity of the effects of various aging factors in the aging process of microplastics, and can significantly enhance the ability to identify and monitor the dynamic distribution of microplastics in the environment. It can provide a scientific basis for the tracing analysis of microplastic pollution and the identification of pollution diffusion pathways, and can provide strong technical support for the governance of microplastic pollution.

[0006] In order to achieve the above-mentioned purpose, the present invention provides a microplastic aging process traceability and factor intelligent identification system, including a sample pretreatment unit, a microplastic separation unit, and a microplastic identification and aging detection unit; The sample pretreatment unit includes a sample conveying and preliminary screening device, a particle dispersion and electrostatic dust removal device, and a microplastic enrichment device; the sample conveying and preliminary screening device includes a sample receiving tank, a vibrating screen and a conveyor belt; the upper end of the sample receiving tank has a feed port, and the lower end thereof has a discharge port, which is used to receive samples and quantitatively output samples; the vibrating screen and the sample receiving tank are distributed at intervals for screening samples; the conveyor belt is arranged between the sample receiving tank and the vibrating screen, and its feed end is located below the discharge port of the sample receiving tank, and its discharge end is located above the feed port of the vibrating screen, which is used to uniformly transport the sample output from the sample receiving tank to the vibrating screen; the particle dispersion and electrostatic dust removal device includes a dispersing agitator, a drying pipeline, a dry air ejector and an electrostatic eliminator; the interior of the dispersing agitator is provided with a stirring mechanism and a multi-stage screening mechanism, which are used to stir, disperse and perform multi-stage screening on the sample, and realize multi-stage particle size range The sample is output in multiple stages; a multi-stage discharge port is arranged at the bottom of the disperser and is respectively connected to a plurality of discharge pipelines; the discharge port of the drying pipeline is connected to the feed port at the top of the disperser, and the feed port is connected to a receiving funnel, and the receiving funnel is arranged below the discharge port of the vibrating screen; the air outlet pipeline of the dry air ejector is sleeved on the outside of the feed section of the drying pipeline, and the two are connected by a plurality of connecting pipelines evenly distributed circumferentially; the static eliminator is sleeved on the outside of the discharge section of the drying pipeline, and is used to eliminate or reduce the accumulation of static electricity in the sample in the drying pipeline; the microplastic enrichment device includes a cyclone separator and a collecting tank; a plurality of cyclone separators are respectively arranged corresponding to a plurality of discharge pipelines, and are respectively connected to the outlet ends of the plurality of discharge pipelines, and at the same time, the divided particle size classification of the plurality of cyclone separators is arranged; a plurality of collecting tanks are correspondingly arranged below the outlet ends of the plurality of discharge pipelines, the upper end of the collecting tank is an open structure, and a discharge port is arranged at the bottom thereof; The microplastic separation unit comprises an electrostatic separation channel, a high-voltage electrostatic charging device and a microplastic collection device; the electrostatic separation channel comprises an electrostatic deflection plate, a collecting electrode and an electric field regulator; the electrostatic separation channel is arranged below a plurality of collecting slots, and a feed channel, an electrostatic separation channel and a plurality of collecting chambers are arranged inside the channel from top to bottom, a plurality of feed ports connected to the feed channel are arranged at the top corresponding to the discharge ports of the plurality of collecting slots, and a plurality of discharge ports are opened at the bottom corresponding to the plurality of collecting chambers; the electrostatic deflection plate is installed in the electrostatic separation channel, and the electrostatic deflection plate is used to form a transverse electric field to cause the falling trajectory of the charged sample to deviate under the action of the transverse electric field, and move to the area where different collecting chambers are located along the deviation direction; the collecting The collecting electrode is installed in the collecting chamber, and is used to adsorb and introduce the charged sample into the collecting chamber; the electric field regulator is connected to the electrostatic deflection plate, and is used to adjust the strength of the transverse electric field; the high-voltage electrostatic charging device includes a high-voltage power supply, a charging electrode and a grounding electrode, and the high-voltage power supply is respectively connected to the charging electrode and the grounding electrode for supplying electricity; the charging electrode and the grounding electrode are relatively arranged on both sides of the feed channel in the electrostatic separation channel, and are used to charge the sample passing through the feed channel; the microplastic collection device includes a microplastic collection tank and an isolation screen; a plurality of microplastic collection tanks are correspondingly arranged below a plurality of discharge ports of the electrostatic separation channel, and the upper end of the microplastic collection tank is an open structure; a plurality of isolation screens are correspondingly assembled at the opening ends of a plurality of microplastic collection tanks; The microplastic identification and aging detection unit includes a high-resolution scanning electron microscope, a Fourier transform infrared spectrometer and an aging characteristic multimodal analysis module; the high-resolution scanning electron microscope is used to bombard the sample surface with an electron beam and generate secondary electrons and backscattered electrons on the sample surface, and at the same time, the secondary electron detector captures the surface secondary electron signal and converts it into image brightness information, thereby generating a high-resolution surface morphology image, and then transmitting the generated surface morphology image to the aging characteristic multimodal analysis module; The Fourier transform infrared spectrometer is used to emit interference light to the sample, and receive the light intensity signal with sample information through the detector, and then convert the interference light with sample information into an infrared spectrum through Fourier transform, and then transmit the generated infrared spectrum of the sample to the aging characteristic multimodal analysis module; The aging characteristic multimodal analysis module is respectively connected to a high-resolution scanning electron microscope and a Fourier transform infrared spectrometer, and is used for identifying and tracing microplastics based on surface morphology images and infrared spectra.

[0007] Furthermore, in order to facilitate the centralized collection of waste, the particle dispersion and electrostatic dust removal device also includes a waste box, and the feed port of the waste box is respectively connected to the waste ports of multiple cyclone separators through multiple connecting pipes.

[0008] Preferably, the aging characteristic multimodal analysis module is an industrial computer.

[0009] As a preferred embodiment, the aging characteristic multimodal analysis module includes a data preprocessing module, a multimodal fusion and analysis module, a traceability and path tracking module, a model evaluation and optimization module, and a data storage and database module; the data preprocessing module, the multimodal fusion and analysis module, the traceability and path tracking module, and the model evaluation and optimization module are connected in sequence, and the data storage and database module is respectively connected to the data preprocessing module, the multimodal fusion and analysis module, the traceability and path tracking module, and the model evaluation and optimization module.

[0010] As a preferred embodiment, the multimodal fusion and analysis module includes a feature extraction layer, a feature splicing layer, a feature compression and fusion layer, and a classification prediction layer; the traceability and path tracking module includes a factor tracing unit and a path simulation unit; the model evaluation and optimization module includes an evaluation index calculation unit, an error analysis unit, an optimization algorithm unit and a feedback unit; the model evaluation and optimization module includes an evaluation index calculation unit, an error analysis unit, an optimization algorithm unit and a feedback unit; the data storage and database module includes an original data storage unit, a feature data storage unit, an analysis result storage unit, a model evaluation and performance recording unit, a model version control unit, a historical data management unit and an environment database.

[0011] As a preferred embodiment, the feature extraction layer adopts a multimodal fusion feature extraction architecture combining a ResNeXt structure and a CNN structure, wherein the ResNeXt structure is used to extract image features and the CNN structure is used to extract spectral features; the classification prediction layer includes two-level fully connected layers, dual activation functions and multiple output layers, which are used to accurately identify the main aging factors of microplastics and their combined effects, and at the same time, to realize the quantification of the overall aging process due to different aging factors.

[0012] In the present invention, a conveyor belt is arranged between the sample receiving tank and the vibrating screen, so that the quantitative unloading of the sample receiving tank can be continuously and uniformly transported to the vibrating screen, thereby facilitating the continuous transport and screening operation. A dry air ejector and an electrostatic eliminator are connected in sequence on the drying pipeline, so that the drying, dust removal and anti-static treatment of the sample can be synchronously achieved in the process of transporting the discharge of the vibrating screen to the dispersing agitator, thereby being conducive to improving the effect and efficiency of the dispersing agitator on the dispersion and multi-stage screening of the sample. Connecting multiple cyclone separators on multiple discharge pipelines of the dispersing agitator can achieve a more refined output of samples in a multi-level particle size range. The charging electrode and the grounding electrode in the high-voltage electrostatic charging device are arranged on both sides of the feed channel, which can facilitate the charging treatment of microplastics and other impurities in the sample, and is conducive to the use of the characteristics that microplastics are charged while other impurities are not easily charged to achieve impurity removal. An electrostatic deflection plate is arranged in the electrostatic separation channel, which can facilitate the use of the formed transverse electric field to achieve the classification of microplastics with different contamination degrees according to the different charges of microplastics. By setting up a collection electrode in the collection chamber, it is possible to further ensure the rapid and accurate classification of microplastics with different degrees of contamination, which is conducive to ensuring the accuracy of subsequent identification and analysis results. The surface morphology image of the sample is obtained using a high-resolution scanning electron microscope, and the infrared spectrum of the sample is obtained using a Fourier transform infrared spectrometer. The aging characteristic multimodal analysis module is then used to identify and analyze microplastics based on the surface morphology image and infrared spectrum. The surface morphology characteristics and chemical functional group characteristics of microplastics can be fully combined to ensure the accuracy of identification and analysis.

[0013] The system has reliable performance and high recognition accuracy. It can realize intelligent identification and analysis of microplastics, helping to quickly and accurately infer the aging sources and possible environmental pathways of microplastics.

[0014] The present invention also provides a method for tracing the source of microplastics and identifying factors during aging, using a system for tracing the source of microplastics and identifying factors intelligently, comprising the following steps: Step 1: Initial rough selection of sample particle size range; The samples to be tested are placed in the sample receiving tank in advance; the conveyor belt and the vibrating screen are started, and the sample is quantitatively lowered from the discharge port of the sample receiving tank, and the lowered samples are uniformly transported to the vibrating screen by the conveyor belt. The vibrating screen uses a screen with a specific aperture to preliminarily screen the samples and screen out samples with smaller particle sizes; Step 2: Dry, remove dust and remove static electricity from the sample; The receiving funnel receives the smaller particle size samples screened by the vibrating screen and conveys them to the disperser through the drying pipeline. During the conveying process, the dry air ejector and the static eliminator are started synchronously. The dry air ejector is used to eject high-speed dry airflow to the feed section of the drying pipeline to disperse and quickly dry the samples. The static eliminator is used to eliminate or reduce the accumulation of static electricity in the dried samples. Step 3: Secondary refinement of the sample particle size range; The samples are stirred, dispersed and sieved by using a dispersing agitator, and the samples of multiple particle size ranges obtained after sieving are respectively output to multiple discharge pipelines; the samples output from multiple discharge pipelines are respectively processed by using multiple cyclone separators, and the samples of the set particle size range are screened out and output under the centrifugal action of the cyclone separators; Step 4: Multi-level sorting of samples; Multiple collecting tanks are used to receive samples of multiple particle size ranges output by multiple cyclone separators, and output them to multiple feed ports of the electrostatic separation channel; a high-voltage electrostatic charging device is used to provide a high-voltage electric field to the samples passing through the feed channel. Under the action of the high-voltage electric field, the microplastic particles in the sample are strongly charged, and other non-microplastic particles are not charged or are weakly charged. The samples that have completed the charging process fall into the electrostatic separation channel; The electrostatic deflection plate is used to provide a transverse electric field, so that the trajectory of the charged sample falling is deflected under the action of the transverse electric field, and moves along the deflection direction to the area where different collection chambers are located. At the same time, the collection electrode is used to adsorb the charged sample in the collection chamber area and guide it into the corresponding sample collection chamber, wherein other particles that are not charged or have a weak charge fall in a straight line to the waste collection chamber, and each collection chamber discharges the collected sample through the corresponding discharge port; The plurality of microplastic collection slots in the microplastic collection device are used to receive samples of multiple particle size ranges discharged from the plurality of discharge ports; Step 5: Obtain the surface morphology image and infrared spectrum of the sample; Start a high-resolution scanning electron microscope, place a portion of the sample in the microplastic collection tank on the detection platform of the high-resolution scanning electron microscope, control the high-resolution scanning electron microscope to emit an electron beam and bombard the sample surface, and generate secondary electrons and backscattered electrons on the sample surface. At the same time, use a secondary electron detector to capture the surface secondary electron signal and convert it into image brightness information, thereby generating a high-resolution surface morphology image, and then transmit the generated initial surface morphology image to the aging feature multimodal analysis module and the data storage and database module; Start the Fourier transform infrared spectrometer, transfer the sample on the detection platform to the sample stage of the Fourier transform infrared spectrometer, use the Fourier transform infrared spectrometer to emit interference light to the sample, and at the same time, capture the light intensity signal passing through the sample or reflected by the sample through the detector, and then convert the light intensity signal with sample information into an initial infrared spectrum through Fourier transform, and then transmit the generated initial infrared spectrum of the sample to the aging characteristic multimodal analysis module and the data storage and database module; After receiving the initial surface morphology image and the initial infrared spectrum, the data storage and database module stores them in the original data storage unit and generates a unique sample ID to ensure the identification and consistency of the data; Step 6: Construct a prediction model for tracing the source and identifying factors of microplastic aging process; S61: After receiving the surface morphology image and infrared spectrum, the aging feature multimodal analysis module uses the data preprocessing module to perform data preprocessing. For the surface morphology image, the data preprocessing module normalizes all the obtained surface morphology images to a uniform size, and applies filtering technology to eliminate random noise in the image, reduce the model error caused by image noise, and make the texture features clearer. At the same time, the image is rotated, scaled, and cropped through image enhancement technology to generate more variable sample data. The processed surface morphology image is transmitted to the multimodal fusion and analysis module and the data storage and database module. For the infrared spectrum, a baseline calibration is first performed. Positive, use the multi-point baseline correction algorithm to automatically adjust the baseline, eliminate background noise and drift, and then use the maximum normalization method to normalize the spectral data, scale the spectral data of all samples to the same level, and at the same time, apply the Savitzky-Golay smoothing filter technology to reduce the noise and smooth the spectral data to reduce the influence of random noise and smooth the spectral curve. The processed infrared spectral data is transmitted to the multimodal fusion and analysis module and the data storage and database module; after receiving the processed surface morphology image and the processed infrared spectral data, the data storage and database module stores them in the feature data storage unit; S62: Processing is performed using a multimodal fusion and analysis module; in the feature extraction layer, the ResNeXt structure is used to extract texture and morphological features from the surface morphology image, and the basic feature extraction of the infrared spectrum image is performed through the initial convolution layer of the ResNeXt structure to extract low-level features such as edges, simple textures, and color changes. At the same time, the hierarchical features are improved through multi-layer group convolutions. The group convolution divides the input channels into multiple groups, and each group performs convolution operations independently to obtain rich feature representations; residual connections are used between convolution layers to add the input and output to prevent gradient disappearance and improve training stability; after multi-layer convolutions, a feature map with high-level features is generated, and finally global average pooling is used to compress the feature map into a vector of fixed length to generate an image Feature vector, and pass it to the feature splicing layer and data storage and database module; at the same time, use the CNN architecture to extract the functional group features of the spectral data; extract the preliminary spectral peak features of the spectral data through the initial convolution layer of CNN; further extract the key spectral features through a multi-layer convolutional network, each layer of convolution focuses on the changes in different frequency ranges in the spectrum, and captures the functional group peak and intensity changes; use maximum pooling or average pooling to reduce the data dimension while retaining key feature information. After the last layer of convolution output, use global average pooling to generate a fixed-length spectral feature vector and pass it to the feature splicing layer and data storage and database module; after receiving the image feature vector and spectral feature vector, the data storage and database module stores them in the feature data storage unit; Next, after receiving the image feature vector and the spectral feature vector, the feature concatenation layer directly connects the image feature vector and the spectral feature vector adjacent to each other through the Concatenation layer to form a joint vector containing all the information, and then inputs the joint vector into the feature compression and fusion layer; Then, after receiving the joint vector, the feature compression and fusion layer performs the first layer of full connection operation, performs preliminary compression on the joint feature vector, reduces the feature dimension from high dimension to a lower dimension, and performs linear transformation through the weight matrix to recombine the image and spectral information and filter out some irrelevant features; after the output of the fully connected layer, a nonlinear activation function is applied to enable the model to learn the nonlinear relationship between the image and spectral features, so that the model can model the complex associations between different modalities to better understand the implicit patterns in multimodal information; after the first layer of full connection operation, the Dropout layer is used to randomly remove some nodes to prevent the model from overfitting in the high-dimensional feature space and improve the generalization ability of the model; repeat the above process, perform multi-layer full connection fusion, further compress the joint feature dimension, gradually reduce the output dimension of each layer, and finally compress the high-dimensional joint feature into a low-dimensional representation; the output of the last layer of the fully connected network is a low-dimensional fused feature vector, which contains the compressed multimodal information, which is the final multimodal representation of the model, and then the low-dimensional fused feature vector is passed to the classification prediction layer; Finally, in the classification prediction layer, the input low-dimensional fused feature vector is linearly transformed and feature mapped through the first-level fully connected layer, so that the model can identify significant feature patterns related to the classification of aging factors; the first-level fully connected layer transforms the input vector through the weight matrix to extract features with classification directionality, thus laying the foundation for the preliminary classification of aging factors; then, these preliminary extracted features are processed through the Softmax activation function to convert the output of the model into a probability distribution, which is used to identify the dominant aging factor of the sample; the output of this stage provides intuitive labels for the most likely aging factors, so that the model can quickly identify the main aging sources of the samples; after determining the main aging factors, the feature vectors are further passed to the second-level fully connected layer, in which the model performs quantitative analysis on multimodal features to extract the specific contributions of different aging factors to the aging process; at this time, the second-level fully connected layer processes the neuron nodes and weights layer by layer, The input features are refined; the secondary layer uses the Sigmoid activation function to generate probability values ​​for each aging factor one by one, and the probability value represents the possible contribution of each factor to the aging of the sample; the Sigmoid function converts the output value of each aging factor into an independent probability, thereby quantifying the influence of multiple aging factors in the combination mode; finally, the output of the classification prediction layer includes two parts: one is the dominant aging factor label generated by Softmax, and the other is the combined probability distribution of each aging factor; the dominant aging factor label and the combined probability distribution of each aging factor are sent to the traceability and path tracking module and the data storage and database module; after receiving the dominant aging factor label and the combined probability distribution of each aging factor, the data storage and database module stores them in the analysis result storage unit; the analysis structure storage unit associates the dominant aging factor label and the combined probability distribution of each aging factor with the original sample data according to the sample ID; S63: Processing using the traceability and path tracking modules; first, the factor traceability unit is used to prioritize the dominant aging factor labels and identify the dominant factor labels. At the same time, the factor traceability unit analyzes and processes the combined probability distribution of each aging factor, sorts them according to the probability values ​​of different factors, and confirms the priority of each factor in combination with the typical aging characteristics of the environmental database; the factor traceability unit matches the processed dominant labels and probability distributions with the environmental database, which contains the typical aging characteristics of microplastics in different environments; the factor traceability unit uses the dominant factor labels as preliminary screening conditions to initially narrow the traceability scope and find possible environmental types in the database, and then further compares the group Combined probability distribution is used to refine the selection; after matching the potential environment type, the factor tracing unit further calculates the weights of each aging factor to reflect the specific impact of each factor on the aging process of the current sample; the weight calculation is based on the value of each factor in the combined probability distribution, combined with the typical weight range in the environment in the database; the factor tracing unit adjusts these weights according to the specific values ​​in the combined probability distribution to ensure that the generated weights can reflect the actual aging pattern of the sample in the environment; the factor tracing unit generates one or more possible potential source environments based on the results of feature matching and the output of weight calculation, and sends them to the path simulation unit; the model gives priority to the environment type with the highest matching degree, and provides matching scores for different environment types; When the path simulation unit receives information about the potential source environment, the model loads the corresponding migration conditions and selects a suitable migration model, and then selects different path modeling methods according to the different source environments. During the model construction process, the path simulation unit uses nodes and edges to represent the possible path nodes of microplastics and the relationship between adjacent nodes to form a migration network; after the model is built, the path simulation unit integrates the aging factor weights received from the factor tracing unit into the migration model to adjust the priority of the migration path. The weight integration process is carried out through the edges or nodes in the weighted path model, so that the model can more accurately reflect the diffusion trend of microplastics under the influence of multiple factors; after integrating the weights of the aging factors, the path simulation unit The migration path is optimized, the total migration probability and matching degree of each path are calculated, and the path is dynamically adjusted according to the influence of the aging factor weight. The state of the nodes in the path is recursively calculated, and the node sequence on the path is optimized to make it more consistent with the actual migration pattern. The local node selection of the path is re-evaluated to ensure that the path meets the environmental conditions and the action mode of aging factors both globally and locally. The optimized path will be integrated into a path map data structure, which contains node and edge information. Each node represents a potential position of microplastics in the migration process, and the edge represents the migration direction and probability between adjacent positions. Finally, the path simulation unit sends the path map data structure to the model evaluation and optimization module. S64: When the model evaluation and optimization module receives the output data of the path simulation unit, it first uses the evaluation index calculation unit to compare the received data with the known true label to obtain the evaluation result data of the model, the evaluation result includes the calculated key performance indicators, and then sends the evaluation result to the data storage and database module; after receiving the evaluation result, the data storage and database module stores it in the model evaluation and performance recording unit after adding a timestamp; The error analysis unit further analyzes the model's error prediction and obtains the error analysis results. By associating the errors with specific environments and factor distribution characteristics, a specific direction is provided for the model's improvement. At the same time, the error analysis results are sent to the optimization algorithm unit. The optimization algorithm unit selects a suitable optimization strategy based on the error analysis results, adjusts the model parameters, weights and model structure, and uses an automated hyperparameter tuning process to improve the prediction accuracy of the model under a specific combination of aging factors to obtain an optimized model, and sends the optimized model information to the data storage and database module and the feedback unit; after receiving the optimized model information, the data storage and database module stores it in the model version control unit; the feedback unit feeds the optimized model back to the multimodal fusion and analysis module and the traceability and path tracking module, and performs data analysis and traceability tests again until the error analysis result obtained by the error analysis unit is less than the set threshold, and a prediction model for tracing the microplastic aging process and identifying factors is obtained; Step 7: Conduct real-time analysis and prediction of microplastics, and output the analysis and prediction results; According to the process in step one to step five, an initial surface morphology image and an initial infrared spectrum are obtained, and the initial surface morphology image and the initial infrared spectrum are input as input data to a prediction model for tracing the source and identifying factors of the microplastic aging process. The prediction model for tracing the source and identifying factors of the microplastic aging process is used to perform analysis and prediction to obtain analysis and prediction results. At the same time, the analysis and prediction results are output and sent to the data storage and database module. After receiving the analysis and prediction results, the data storage and database module stores them in the historical data management unit.

[0015] As a preferred embodiment, in step four, an electric field regulator is used to adjust the electric field strength of the electrostatic deflection plate to meet the needs of separating microplastic particles with different charges and effectively ensure the separation accuracy.

[0016] Compared with the prior art, the present invention has the following advantages: 1. The present invention adopts a multimodal fusion analysis method, in conjunction with a deep learning model, through a comprehensive analysis of the surface morphology and chemical characteristics of microplastics, effectively combining the morphology and chemical functional groups of microplastics, and quantifying the contribution of different environmental factors to the aging process, so as to more comprehensively identify the different factors in the aging process of microplastics, and quickly and accurately infer the aging sources and possible environmental paths of microplastics. Compared with the traditional technology that relies only on a single data (such as an image or spectrum) for aging discrimination, the present invention achieves more accurate aging feature extraction through data fusion of a high-resolution scanning electron microscope and a Fourier transform infrared spectrometer, greatly improving the recognition accuracy of complex aging processes.

[0017] 2. The present invention is equipped with a special traceability and path tracking module. By matching the characteristics of aging factor labels and combined probability distributions, it can effectively narrow the traceability scope and identify the potential source environment of the sample by combining the aging feature templates of different environmental conditions in the environmental database. This traceability analysis based on multi-factor feature comparison greatly improves the accuracy of identifying the source of microplastic pollution, allowing this method to effectively distinguish the aging paths of microplastics in different environments (such as oceans, beaches, and industrial areas).

[0018] 3. Through the automated sample pretreatment unit, the present invention realizes automated operation in the links of sample transportation, electrostatic dust removal and particle enrichment. The coordination of devices such as vibrating screen, electrostatic eliminator and multi-stage cyclone separator can not only effectively remove unnecessary particles, but also enrich microplastic particles of a specific particle size, reducing the manual intervention links and reducing the labor intensity. At the same time, this highly automated pretreatment process greatly shortens the sample processing time and improves the overall processing efficiency, which is suitable for the rapid identification and analysis of batch samples.

[0019] The present invention adopts the joint architecture of ResNeXt and CNN for multimodal feature extraction, and uses fully connected networks, Softmax and Sigmoid activation functions and other technologies in the subsequent feature splicing, compression and classification process to build a powerful multimodal analysis capability. Compared with the traditional aging recognition method based on rules or simple machine learning, the present invention can automatically learn complex nonlinear feature relationships through deep learning, ensuring higher adaptability and generalization ability under different samples and aging conditions, and effectively improving the recognition and analysis capabilities of various microplastic aging characteristics. The method uses multimodal fusion and analysis methods to effectively distinguish the intensity of the effects of various aging factors in the microplastic aging process, significantly enhances the recognition and monitoring capabilities of the dynamic distribution of microplastics in the environment, provides a scientific basis for the source analysis of microplastic pollution and the identification of pollution diffusion paths, and can provide strong technical support for the governance of microplastic pollution. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a structural schematic diagram of the identification system in the present invention; Figure 2 is a flow chart of the identification method of the present invention; Figure 3 It is a principle block diagram of the aging characteristic multimodal analysis module in the present invention.

[0021] In the figure: 1. Sample receiving tank; 2. Conveyor belt; 3. Vibrating screen; 4. Dry air ejector; 5. Static eliminator; 6. Disperser; 7. Cyclone separator; 8. Collecting tank; 9. High-voltage electrostatic charging device; 10. Electrostatic separation channel; 11. Microplastic collecting device; 12. Drying pipeline; 13. Receiving funnel; 14. Discharge pipeline. DETAILED DESCRIPTION

[0022] The present invention will be further described below in conjunction with the accompanying drawings.

[0023] like Figures 1 to 3 As shown, the present invention provides a microplastic aging process traceability and factor intelligent identification system, including a sample pretreatment unit, a microplastic separation unit, and a microplastic identification and aging detection unit; The sample pretreatment unit includes a sample conveying and preliminary screening device, a particle dispersion and electrostatic dust removal device, and a microplastic enrichment device; the sample conveying and preliminary screening device includes a sample receiving tank 1, a vibrating screen 3 and a conveyor belt 2; the upper end of the sample receiving tank 1 has a feed port, and the lower end thereof has a discharge port, which is used to receive samples and quantitatively output samples; the vibrating screen 3 and the sample receiving tank 1 are distributed at intervals, and are used to screen the samples; the conveyor belt 2 is arranged between the sample receiving tank 1 and the vibrating screen 3, and its feed end is located below the discharge port of the sample receiving tank 1, and its discharge end is located above the feed port of the vibrating screen 3, and is used to uniformly transport the sample output from the sample receiving tank 1 to the vibrating screen 3; the particle dispersion and electrostatic dust removal device includes a dispersing agitator 6, a drying pipeline 12, a dry air ejector 4 and an electrostatic eliminator 5; the dispersing agitator 6 is provided with a stirring mechanism and a multi-stage screening mechanism, which are used to stir, disperse and screen the samples, and realize multi-stage screening of samples with multi-stage particle size ranges Output; the bottom of the disperser 6 is provided with multiple discharge ports and connected to multiple discharge pipelines 14 respectively; the discharge port of the drying pipeline 12 is connected to the feed port at the top of the disperser 6, and its feed port is connected to a receiving funnel 13, and the receiving funnel 13 is arranged below the discharge port of the vibrating screen 3; the air outlet pipeline of the dry air ejector 4 is sleeved on the outside of the feed section of the drying pipeline 12, and the two are connected by multiple connecting pipelines evenly distributed in the circumferential direction; the static eliminator 5 is sleeved on the drying pipeline The outer side of the discharge section 12 is used to eliminate or reduce the accumulation of static electricity in the sample in the drying pipeline 12; the microplastic enrichment device includes a cyclone separator 7 and a collecting tank 8; a plurality of cyclone separators 7 are respectively arranged corresponding to a plurality of discharge pipelines 14, and are respectively connected to the outlet ends of the plurality of discharge pipelines 14, and at the same time, the divided particle sizes of the plurality of cyclone separators 7 are graded; a plurality of collecting tanks 8 are correspondingly arranged below the outlet ends of the plurality of discharge pipelines 14, the upper end of the collecting tank 8 is an open structure, and a discharge port is arranged at the bottom thereof; The separation particle size of the cyclone separator 7 is adjustable, and is used to separate particles of different sizes by centrifugal force, wherein microplastics and other particles of suitable sizes are enriched in the collection tank, smaller particles are excluded, and oversized particles are directed to the waste bin.

[0024] The microplastic separation unit includes an electrostatic separation channel 10, a high-voltage electrostatic charging device 9 and a microplastic collection device 11; the electrostatic separation channel 10 includes an electrostatic deflection plate, a collecting electrode and an electric field regulator; the electrostatic separation channel 10 is arranged below a plurality of collecting slots 8, and a feed channel, an electrostatic separation channel and a plurality of collecting chambers are arranged inside the channel from top to bottom, and a plurality of feed ports connected to the feed channel are arranged at the top corresponding to the discharge ports of the plurality of collecting slots 8, and a plurality of discharge ports are opened at the bottom corresponding to the plurality of collecting chambers; the electrostatic deflection plate is installed in the electrostatic separation channel, and the electrostatic deflection plate is used to form a transverse electric field to cause the falling trajectory of the charged sample to deviate under the action of the transverse electric field, and move to the area where different collecting chambers are located along the deviation direction; the collecting electrode is installed in the collecting chamber, and is used to adsorb and The charged sample is introduced into the collection chamber; the electric field regulator is connected to the electrostatic deflection plate to adjust the strength of the transverse electric field; the high-voltage electrostatic charging device 9 includes a high-voltage power supply, a charging electrode and a grounding electrode, and the high-voltage power supply is respectively connected to the charging electrode and the grounding electrode for supplying electricity; the charging electrode and the grounding electrode are relatively arranged on both sides of the feed channel in the electrostatic separation channel 10, so as to charge the sample passing through the feed channel; the microplastic collection device 11 includes a microplastic collection tank and an isolation screen; a plurality of microplastic collection tanks are correspondingly arranged below a plurality of discharge ports of the electrostatic separation channel 10, and the upper end of the microplastic collection tank is an open structure; a plurality of isolation screens are correspondingly assembled at the open ends of a plurality of microplastic collection tanks; by setting the isolation screen, other particles can be prevented from entering the microplastic collection area to ensure the purity of the collected particles; The microplastic identification and aging detection unit includes a high-resolution scanning electron microscope, a Fourier transform infrared spectrometer and an aging characteristic multimodal analysis module; the high-resolution scanning electron microscope is used to bombard the sample surface with an electron beam and generate secondary electrons and backscattered electrons on the sample surface, and at the same time, the secondary electron detector captures the surface secondary electron signal and converts it into image brightness information, thereby generating a high-resolution surface morphology image, and then transmitting the generated surface morphology image to the aging characteristic multimodal analysis module; The Fourier transform infrared spectrometer is used to emit interference light to the sample. The Fourier transform infrared spectrometer can emit light waves in the near infrared to mid-infrared range, and receive light intensity signals with sample information through a detector, and then convert the interference light with sample information into an infrared spectrum through Fourier transform, and then transmit the generated infrared spectrum of the sample to the aging characteristic multimodal analysis module; The aging characteristic multimodal analysis module is respectively connected to a high-resolution scanning electron microscope and a Fourier transform infrared spectrometer, and is used for identifying and tracing microplastics based on surface morphology images and infrared spectra.

[0025] In order to facilitate the centralized collection of waste materials, the particle dispersion and electrostatic dust removal device further comprises a waste box, and a feed port of the waste box is respectively connected to waste ports of a plurality of cyclone separators 7 through a plurality of connecting pipes.

[0026] Preferably, the aging characteristic multimodal analysis module is an industrial computer.

[0027] As a preferred embodiment, the aging characteristic multimodal analysis module includes a data preprocessing module, a multimodal fusion and analysis module, a traceability and path tracking module, a model evaluation and optimization module, and a data storage and database module; the data preprocessing module, the multimodal fusion and analysis module, the traceability and path tracking module, and the model evaluation and optimization module are connected in sequence, and the data storage and database module is respectively connected to the data preprocessing module, the multimodal fusion and analysis module, the traceability and path tracking module, and the model evaluation and optimization module.

[0028] As a preferred embodiment, the multimodal fusion and analysis module includes a feature extraction layer, a feature splicing layer, a feature compression and fusion layer, and a classification prediction layer; the traceability and path tracking module includes a factor tracing unit and a path simulation unit; the model evaluation and optimization module includes an evaluation index calculation unit, an error analysis unit, an optimization algorithm unit and a feedback unit; the model evaluation and optimization module includes an evaluation index calculation unit, an error analysis unit, an optimization algorithm unit and a feedback unit; the data storage and database module includes an original data storage unit, a feature data storage unit, an analysis result storage unit, a model evaluation and performance recording unit, a model version control unit, a historical data management unit and an environment database.

[0029] As a preferred embodiment, the feature extraction layer adopts a multimodal fusion feature extraction architecture combining a ResNeXt structure and a CNN structure, wherein the ResNeXt structure is used to extract image features and the CNN structure is used to extract spectral features; the classification prediction layer includes two-level fully connected layers, dual activation functions and multiple output layers, which are used to accurately identify the main aging factors of microplastics and their combined effects, and at the same time, to realize the quantification of the overall aging process due to different aging factors.

[0030] In the present invention, a conveyor belt is arranged between the sample receiving tank and the vibrating screen, so that the quantitative unloading of the sample receiving tank can be continuously and uniformly transported to the vibrating screen, thereby facilitating the continuous transport and screening operation. A dry air ejector and an electrostatic eliminator are connected in sequence on the drying pipeline, so that the drying, dust removal and anti-static treatment of the sample can be synchronously achieved in the process of transporting the discharge of the vibrating screen to the dispersing agitator, thereby being conducive to improving the effect and efficiency of the dispersing agitator on the dispersion and multi-stage screening of the sample. Connecting multiple cyclone separators on multiple discharge pipelines of the dispersing agitator can achieve a more refined output of samples in a multi-level particle size range. The charging electrode and the grounding electrode in the high-voltage electrostatic charging device are arranged on both sides of the feed channel, which can facilitate the charging treatment of microplastics and other impurities in the sample, and is conducive to the use of the characteristics that microplastics are charged while other impurities are not easily charged to achieve impurity removal. An electrostatic deflection plate is arranged in the electrostatic separation channel, which can facilitate the use of the formed transverse electric field to achieve the classification of microplastics with different contamination degrees according to the different charges of microplastics. By setting up a collection electrode in the collection chamber, it is possible to further ensure the rapid and accurate classification of microplastics with different degrees of contamination, which is conducive to ensuring the accuracy of subsequent identification and analysis results. The surface morphology image of the sample is obtained using a high-resolution scanning electron microscope, and the infrared spectrum of the sample is obtained using a Fourier transform infrared spectrometer. The aging characteristic multimodal analysis module is then used to identify and analyze microplastics based on the surface morphology image and infrared spectrum. The surface morphology characteristics and chemical functional group characteristics of microplastics can be fully combined to ensure the accuracy of identification and analysis.

[0031] The system has reliable performance and high recognition accuracy. It can realize intelligent identification and analysis of microplastics, helping to quickly and accurately infer the aging sources and possible environmental pathways of microplastics.

[0032] The present invention also provides a method for tracing the source of microplastics and identifying factors during aging, using a system for tracing the source of microplastics and identifying factors intelligently, comprising the following steps: Step 1: Initial rough selection of sample particle size range; The sample to be tested is placed in the sample receiving tank 1 in advance; the conveyor belt 2 and the vibrating screen 3 are started, and the sample is quantitatively lowered from the discharge port of the sample receiving tank 1, and the conveyor belt 2 is used to uniformly transport the lowered sample to the vibrating screen 3, and the vibrating screen 3 uses a screen with a specific aperture to perform a preliminary screening process on the sample and screen out samples with smaller particle sizes; Step 2: Dry, remove dust and remove static electricity from the sample; The receiving funnel 13 is used to receive the sample with smaller particle size screened by the vibrating screen 3, and is transported to the disperser 6 through the drying pipeline 12. During the transportation process, the dry air ejector 4 and the static eliminator 5 are synchronously started, and the dry air ejector 4 is used to eject high-speed dry airflow to the feed section of the drying pipeline 12 to disperse and quickly dry the sample, and the static eliminator 5 is used to eliminate or reduce the accumulation of static electricity in the sample after drying; Step 3: Secondary refinement of the sample particle size range; The sample is stirred, dispersed and sieved by a dispersing stirrer 6, and the samples of multiple particle size ranges obtained after sieving are respectively output to multiple discharge pipes 14; the samples output from the multiple discharge pipes 14 are respectively processed by multiple cyclone separators 7, and the samples of the set particle size range are screened out and output under the centrifugal action of the cyclone separators 7; Step 4: Multi-level sorting of samples; Multiple collecting tanks 8 are used to receive samples of multiple particle size ranges output by multiple cyclone separators 7, and output them to multiple feed ports of an electrostatic separation channel 10. The incoming samples pass through a high-voltage electric field between a charging electrode and a grounding electrode. A high-voltage electrostatic charging device 9 is used to provide a high-voltage electric field to the samples passing through the feed channel. Under the action of the high-voltage electric field, the microplastic particles in the sample are more strongly charged, while other non-microplastic particles are not charged or are less charged. The sample that has completed the charging process falls into the electrostatic separation channel. The electrostatic deflection plate is used to provide a transverse electric field, so that the trajectory of the charged sample falling is deflected under the action of the transverse electric field, and moves along the deflection direction to the area where different collection chambers are located. At the same time, the collection electrode is used to adsorb the charged sample in the collection chamber area and guide it into the corresponding sample collection chamber, wherein other particles that are not charged or have a weak charge fall in a straight line to the waste collection chamber, and each collection chamber discharges the collected sample through the corresponding discharge port; The multiple microplastic collection slots in the microplastic collection device 11 are used to receive samples of multiple particle size ranges discharged from the multiple discharge ports; Step 5: Start the high-resolution scanning electron microscope, place a portion of the sample in the microplastic collection tank on the detection platform of the high-resolution scanning electron microscope, control the high-resolution scanning electron microscope to emit an electron beam and bombard the sample surface, and generate secondary electrons and backscattered electrons on the sample surface. At the same time, use the secondary electron detector to capture the surface secondary electron signal and convert it into image brightness information, thereby generating a high-resolution surface morphology image, and then transmit the generated initial surface morphology image to the aging feature multimodal analysis module and the data storage and database module; Start the Fourier transform infrared spectrometer, transfer the sample on the detection platform to the sample stage of the Fourier transform infrared spectrometer, use the Fourier transform infrared spectrometer to emit interference light to the sample, and at the same time, capture the light intensity signal passing through the sample or reflected by the sample through the detector, and then convert the light intensity signal with sample information into an initial infrared spectrum through Fourier transform, and then transmit the generated initial infrared spectrum of the sample to the aging characteristic multimodal analysis module and the data storage and database module; After receiving the initial surface morphology image and the initial infrared spectrum, the data storage and database module stores them in the original data storage unit and generates a unique sample ID to ensure the identification and consistency of the data; Step 6: Construct a prediction model for tracing the source and identifying factors of microplastic aging process; S61: After receiving the surface morphology image and the infrared spectrum, the aging feature multimodal analysis module uses the data preprocessing module to perform data preprocessing. For the surface morphology image, the data preprocessing module normalizes all the obtained surface morphology images to a uniform size, and applies filtering technology to eliminate random noise in the image, reduce the model error caused by image noise, and make the texture feature clearer. At the same time, through the image enhancement technology, the image is rotated, scaled, cropped, etc. to generate more variable sample data, which is conducive to improving the adaptability of the model to different image features during the training process. The processed surface morphology image is transmitted to the multimodal fusion and analysis module and the data storage and database module. For the infrared spectrum, baseline correction is first performed, and the baseline is automatically adjusted using a multi-point baseline correction algorithm to eliminate background noise and drift. Then, the maximum normalization method is used to normalize the spectral data, and the spectral data of all samples are scaled to the same magnitude. At the same time, the Savitzky-Golay smoothing filter technology is applied to reduce noise and smooth the spectral data to reduce the influence of random noise and smooth the spectral curve. The processed infrared spectral data is stored in the multimodal fusion and analysis module and the data storage and database module; after receiving the processed surface morphology image and the processed infrared spectral data, the data storage and database module stores them in the feature data storage unit; S62: Processing is performed using the multimodal fusion and analysis module; in the feature extraction layer, the ResNeXt structure is used to extract the texture and morphology features in the surface morphology image, and the basic feature extraction is performed on the infrared spectrum through the initial convolution layer of the ResNeXt structure to extract low-level features such as edges, simple textures, and color changes. At the same time, the hierarchical features are improved through multi-layer group convolution. The group convolution divides the input channels into multiple groups, and each group performs convolution operations independently to obtain rich feature representations, such as surface roughness and detailed cracks; residual connections are used between convolution layers to add the input and output to prevent gradient disappearance and improve training stability; after multi-layer convolution, a feature map with high-level features is generated, such as complex cracks and fracture textures, corrosion marks and depressions, as well as non-uniform textures and particle aggregations. Finally, global average pooling is used to compress the feature map into a vector of fixed length to generate an image feature vector, which is passed to the feature splicing layer and data Storage and database module; at the same time, the CNN architecture is used to extract the functional group features of the spectral data; the initial spectral peak features of the spectral data are extracted through the initial convolution layer of CNN, mainly including the basic peak and width information in the spectrum; the key spectral features are further extracted through a multi-layer convolutional network, and each layer of convolution focuses on the changes in different frequency ranges in the spectrum to capture the functional group peak and intensity changes; the maximum pooling or average pooling is used to reduce the data dimension while retaining the key feature information. After the output of the last layer of convolution, the global average pooling is used to generate a fixed-length spectral feature vector, which is passed to the feature concatenation layer and the data storage and database module; after receiving the standardized image feature vector and spectral feature vector, the data storage and database module stores them in the feature data storage unit; as a preferred embodiment, the feature data storage unit forms a complete data record by storing the data of each new sample (original data and its feature data) in the corresponding unit; Next, after receiving the image feature vector and the spectral feature vector, the feature concatenation layer directly connects the image feature vector and the spectral feature vector adjacent to each other through the Concatenation layer to form a joint vector containing all the information, and then inputs the joint vector into the feature compression and fusion layer; Then, after receiving the joint vector, the feature compression and fusion layer performs the first layer of full connection operation, performs preliminary compression on the joint feature vector, reduces the feature dimension from high dimension to a lower dimension, and performs linear transformation through the weight matrix to recombine the image and spectral information and filter out some irrelevant features; after the output of the fully connected layer, a nonlinear activation function is applied to enable the model to learn the nonlinear relationship between the image and spectral features, so that the model can model the complex associations between different modalities to better understand the implicit patterns in multimodal information; after the first layer of full connection operation, the Dropout layer is used to randomly remove some nodes to prevent the model from overfitting in the high-dimensional feature space and improve the generalization ability of the model; repeat the above process, perform multi-layer full connection fusion, further compress the joint feature dimension, gradually reduce the output dimension of each layer, and finally compress the high-dimensional joint feature into a low-dimensional representation; the output of the last layer of the fully connected network is a low-dimensional fused feature vector, which contains the compressed multimodal information, which is the final multimodal representation of the model, and then the low-dimensional fused feature vector is passed to the classification prediction layer; Finally, in the classification prediction layer, the input low-dimensional fused feature vector is linearly transformed and feature mapped through the first-level fully connected layer, so that the model can identify significant feature patterns related to the classification of aging factors; the first-level fully connected layer transforms the input vector through the weight matrix to extract features with classification directionality, thus laying the foundation for the preliminary classification of aging factors; then, these preliminary extracted features are processed through the Softmax activation function to convert the output of the model into a probability distribution, which is used to identify the dominant aging factor of the sample; the output of this stage provides intuitive labels for the most likely aging factors, so that the model can quickly identify the main aging sources of the samples; after determining the main aging factors, the feature vectors are further passed to the second-level fully connected layer, in which the model performs quantitative analysis on multimodal features to extract the specific contributions of different aging factors to the aging process; at this time, the second-level fully connected layer refines the input features through layer-by-layer processing of neuron nodes and weights; in order to achieve aging The combined quantitative output of the aging factors, the secondary layer uses the Sigmoid activation function to generate probability values ​​for each aging factor one by one, and the probability value represents the possible contribution of each factor to the sample aging; the Sigmoid function converts the output value of each aging factor into an independent probability, thereby quantifying the influence of multiple aging factors in the combined mode; finally, the output of the classification prediction layer includes two parts: one is the dominant aging factor label generated by Softmax, and the other is the combined probability distribution of each aging factor; the dominant aging factor label and the combined probability distribution of each aging factor are sent to the traceability and path tracking module and the data storage and database module; after receiving the dominant aging factor label and the combined probability distribution of each aging factor, the data storage and database module stores them in the analysis result storage unit; the analysis structure storage unit associates the dominant aging factor label and the combined probability distribution of each aging factor and other result information with the original sample data according to the sample ID, and saves the detailed output of each analysis; S63: Processing using the traceability and path tracking modules; first, the factor traceability unit is used to prioritize the dominant aging factor labels and identify the dominant factor labels, which represent the model's judgment of the most likely main aging factor of the sample; at the same time, the factor traceability unit analyzes and processes the combined probability distribution of each aging factor, sorts them according to the probability values ​​of different factors, and confirms the priority of each factor in combination with the typical aging characteristics of the environmental database. For example, if the probability values ​​of UV and mechanical wear are the highest, it means that the sample may have experienced major aging under ultraviolet exposure and physical contact; the factor traceability unit matches the processed dominant labels and probability distributions with the environmental database, which contains typical aging characteristics of microplastics under different environments, such as surface cracks caused by UV aging, functional group changes caused by chemical oxidation, etc.; the factor traceability unit uses the dominant factor labels as preliminary screening conditions to initially narrow the traceability scope. For example, if the dominant aging factor is UV , the possible source environment will be concentrated in outdoor exposure areas, such as beaches or water surfaces, while if the dominant factor is thermal oxidation, the possible source environment is more inclined to industrial areas or high-temperature processing areas; find the possible environment types in the database, and then refine the selection by further comparing the combined probability distribution; after matching the potential environment type, the factor tracing unit further calculates the weights of each aging factor to reflect the specific impact of each factor on the aging process of the current sample; the weight calculation is based on the value of each factor in the combined probability distribution, combined with the typical weight range in the environment in the database; the factor tracing unit adjusts these weights according to the specific values ​​in the combined probability distribution to ensure that the generated weights can reflect the actual aging pattern of the sample in the environment; the factor tracing unit generates one or more possible potential source environments based on the results of feature matching and the output of weight calculation, and sends them to the path simulation unit; the model gives priority to the environment type with the highest matching degree, and provides matching scores for different environment types; When the path simulation unit receives information about the potential source environment, the model loads the corresponding migration conditions and selects a suitable migration model. It then selects different path modeling methods according to the different source environments. In the water environment, the model uses a graph convolutional network (GCN) to simulate the diffusion path of microplastics in flowing water, and combines information such as flow rate to build a migration network. In the soil environment, the model uses a recurrent neural network (RNN) to process the gradual infiltration pattern of samples in the soil medium. During the model construction process, the path simulation unit uses nodes and edges to represent the possible path nodes of microplastics and the relationship between adjacent nodes to form a migration network. After the model is built, the path simulation unit integrates the aging factor weights received from the factor tracing unit into the migration model to adjust the priority of the migration path. The weight integration process is carried out through the edges or nodes in the weighted path model. , so that the model can more accurately reflect the diffusion trend of microplastics under the influence of multiple factors; after integrating the weight of the aging factor, the path simulation unit optimizes the migration path, calculates the total migration probability and matching degree of each path, and dynamically adjusts the path according to the influence of the aging factor weight, recursively calculates the state of the nodes in the path, optimizes the node sequence on the path, so that it is more in line with the actual migration pattern, and re-evaluates the local node selection of the path to ensure that the path meets the environmental conditions and the action mode of the aging factor both globally and locally. The optimized path will be integrated into a path graph data structure, which contains node and edge information. Each node represents a potential position of microplastics in the migration process, and the edge represents the migration direction and probability between adjacent positions; finally, the path simulation unit sends the path graph data structure to the model evaluation and optimization module; S64: When the model evaluation and optimization module receives the output data of the path simulation unit, it first uses the evaluation index calculation unit to compare the received data with the known true labels to obtain the evaluation result data of the model. The evaluation result includes the calculated key performance indicators, which include the accuracy, precision, recall rate, F1 score and other indicators of the model; then the evaluation result is sent to the data storage and database module; after receiving the evaluation result, the data storage and database module stores it in the model evaluation and performance recording unit after adding a timestamp; after each model update or optimization, the model evaluation and performance recording unit will receive the evaluation result of the model, including information such as performance indicators and model running time. The evaluation and performance recording unit will record the specific evaluation data of each model and add a timestamp to facilitate long-term monitoring of the performance changes of the model; After the evaluation index is calculated, the error analysis unit further analyzes the model's error predictions and obtains error analysis results, such as identifying which aging factor combinations have a high classification error rate or a large traceability path deviation; by associating the error with specific environments and factor distribution characteristics, it provides a specific direction for model improvement, and at the same time, sends the error analysis results to the optimization algorithm unit; The optimization algorithm unit selects a suitable optimization strategy based on the error analysis results, adjusts the model parameters, weights and model structure, and uses an automated hyperparameter tuning process to improve the prediction accuracy of the model under a specific combination of aging factors to obtain the optimized model, and sends the optimized model information to the data storage and database module and the feedback unit; after receiving the optimized model information, the data storage and database module stores it in the model version control unit; after each model update, the model version control unit records the new model version information, including the model structure, hyperparameters, optimization algorithm, timestamp, etc.; the feedback unit feeds the optimized model back to the multimodal fusion and analysis module and the traceability and path tracking module, and performs data analysis and traceability testing again until the error analysis result obtained by the error analysis unit is less than the set threshold, and a prediction model for tracing the microplastic aging process and identifying factors is obtained; Step 7: Conduct real-time analysis and prediction of microplastics, and output the analysis and prediction results; According to the process in step 1 to step 5, the initial surface morphology image and the initial infrared spectrum are obtained, and the initial surface morphology image and the initial infrared spectrum are input as input data to the prediction model for tracing the source and identifying the factors of the microplastic aging process. The prediction model for tracing the source and identifying the factors of the microplastic aging process is used to analyze and predict, and the analysis and prediction results are obtained. At the same time, the analysis and prediction results are output and sent to the data storage and database module. After receiving the analysis and prediction results, the data storage and database module stores them in the historical data management unit. Storing a snapshot of each analysis and prediction result in the historical data management unit can facilitate the formation of a time series record, which is conducive to the system regularly archiving the analysis results for long-term trend analysis and historical comparison.

[0033] As a preferred embodiment, in step four, an electric field regulator is used to adjust the electric field strength of the electrostatic deflection plate to meet the needs of separating microplastic particles with different charges and effectively ensure the separation accuracy.

[0034] The present invention adopts a multimodal fusion analysis method, in conjunction with a deep learning model, through a comprehensive analysis of the surface morphology and chemical characteristics of microplastics, effectively combines the morphology and chemical functional groups of microplastics, and quantifies the contribution of different environmental factors to the aging process, so that the different factors in the aging process of microplastics can be more comprehensively identified, and the aging sources and possible environmental paths of microplastics can be quickly and accurately inferred. Compared with the traditional technology that relies only on a single data (such as an image or spectrum) for aging discrimination, the present invention achieves more accurate aging feature extraction through the data fusion of a high-resolution scanning electron microscope and a Fourier transform infrared spectrometer, greatly improving the recognition accuracy of complex aging processes. At the same time, the present invention is equipped with a special traceability and path tracking module, which can effectively narrow the traceability range by matching the features of the aging factor label and the combined probability distribution, and identify the potential source environment of the sample by combining the aging feature templates of different environmental conditions in the environmental database. This traceability analysis based on multi-factor feature comparison greatly improves the accuracy of identifying the source of microplastic pollution, so that the method can effectively distinguish the aging paths of microplastics in different environments (such as oceans, beaches, and industrial areas). At the same time, through the automated sample pretreatment unit, the present invention realizes automated operation in the links of sample transportation, electrostatic dust removal and particle enrichment. The coordination of devices such as vibrating screens, electrostatic eliminators and multi-stage cyclone separators can not only effectively remove unnecessary particles, but also enrich microplastic particles of a specific particle size, reducing the links of manual intervention and reducing the labor intensity of manual labor. At the same time, this highly automated pretreatment process greatly shortens the sample processing time and improves the overall processing efficiency, and is suitable for the rapid identification and analysis of batch samples.

[0035] The present invention adopts the joint architecture of ResNeXt and CNN for multimodal feature extraction, and uses fully connected networks, Softmax and Sigmoid activation functions and other technologies in the subsequent feature splicing, compression and classification process to build a powerful multimodal analysis capability. Compared with the traditional aging recognition method based on rules or simple machine learning, the present invention can automatically learn complex nonlinear feature relationships through deep learning, ensuring higher adaptability and generalization ability under different samples and aging conditions, and effectively improving the recognition and analysis capabilities of various microplastic aging characteristics. The method uses multimodal fusion and analysis methods to effectively distinguish the intensity of the effects of various aging factors in the microplastic aging process, significantly enhances the recognition and monitoring capabilities of the dynamic distribution of microplastics in the environment, provides a scientific basis for the source analysis of microplastic pollution and the identification of pollution diffusion paths, and can provide strong technical support for the governance of microplastic pollution.

Claims

1. A microplastic aging process traceability and factor intelligent identification system, including a sample pretreatment unit, characterized in that: It also includes a microplastic separation unit, a microplastic identification and aging detection unit; The sample pretreatment unit comprises a sample conveying and primary screening device, a particle dispersion and electrostatic dust removal device, and a microplastic enrichment device; the sample conveying and primary screening device comprises a sample receiving slot (1), a vibrating screen (3), and a conveyor belt (2); the sample receiving slot (1) has a feed port at its upper end and a discharge port at its lower end for receiving samples and quantitatively outputting the samples; the vibrating screen (3) and the sample receiving slot (1) are spaced apart and are used for screening the samples; the conveyor belt (2) is arranged between the sample receiving slot (1) and the vibrating screen (3); The feed end is located below the discharge port of the sample receiving tank (1), and the discharge end is located above the feed port of the vibrating screen (3), and is used to uniformly transport the sample output from the sample receiving tank (1) to the vibrating screen (3); the particle dispersion and electrostatic dust removal device comprises a dispersion stirrer (6), a drying pipeline (12), a dry air ejector (4) and a static eliminator (5); a stirring mechanism and a multi-stage screening mechanism are arranged inside the dispersion stirrer (6), and are used to stir and disperse the sample and perform multi-stage screening processing, and realize multi-stage output of samples with multi-stage particle size ranges; The bottom of the disperser (6) is provided with multiple discharge ports, which are respectively connected to multiple discharge pipelines (14); the discharge port of the drying pipeline (12) is connected to the feed port at the top of the disperser (6), and the feed port is connected to a receiving funnel (13), and the receiving funnel (13) is arranged below the discharge port of the vibrating screen (3); the air outlet pipeline of the dry air ejector (4) is sleeved on the outside of the feed section of the drying pipeline (12), and the two are connected by multiple connecting pipelines evenly distributed in the circumferential direction; the static eliminator (5) is sleeved on the drying pipeline (12) ) outside the discharge section, used to eliminate or reduce the accumulation of static electricity in the sample in the drying pipeline (12); the microplastics enrichment device includes a cyclone separator (7) and a collecting tank (8); the plurality of cyclone separators (7) are respectively arranged corresponding to the plurality of discharge pipelines (14), and are respectively connected to the outlet ends of the plurality of discharge pipelines (14), and at the same time, the divided particle sizes of the plurality of cyclone separators (7) are arranged in a graded manner; the plurality of collecting tanks (8) are correspondingly arranged below the outlet ends of the plurality of discharge pipelines (14), the upper end of the collecting tank (8) is an open structure, and a discharge port is arranged at the bottom thereof; The microplastic separation unit comprises an electrostatic separation channel (10), a high-voltage electrostatic charging device (9) and a microplastic collection device (11); the electrostatic separation channel (10) comprises an electrostatic deflection plate, a collection electrode and an electric field regulator; The electrostatic separation channel (10) is arranged below the plurality of collecting slots (8), and is provided with a feed channel, an electrostatic separation channel and a plurality of collecting chambers from top to bottom. The top of the channel is provided with a plurality of feed ports connected to the feed channel at positions corresponding to the discharge ports of the plurality of collecting slots (8), and the bottom of the channel is provided with a plurality of discharge ports at positions corresponding to the plurality of collecting chambers. The electrostatic deflection plate is installed in the electrostatic separation channel, and is used to form a transverse electric field to cause the falling trajectory of the charged sample to deviate under the action of the transverse electric field and move along the deflection direction to the area where different collecting chambers are located. The collecting electrode is installed in the collecting chamber, and is used to adsorb and guide the charged sample into the collecting chamber. The electric field regulator is connected to the electrostatic deflection plate, and is used to adjust the strength of the transverse electric field. The high-voltage electrostatic charging device (9) includes a high-voltage power supply, a charging electrode and a grounding electrode, and the high-voltage power supply is respectively connected to the charging electrode and the grounding electrode for supplying electricity. The charging electrode and the grounding electrode are relatively arranged on both sides of the feed channel in the electrostatic separation channel (10), and are used to charge the sample passing through the feed channel. The microplastic collection device (11) comprises a microplastic collection tank and an isolation screen; The plurality of microplastic collection tanks are correspondingly arranged below the plurality of discharge ports of the electrostatic separation channel (10), and the upper ends of the microplastic collection tanks are open structures; and the plurality of isolation screens are correspondingly mounted on the open ends of the plurality of microplastic collection tanks; The microplastic identification and aging detection unit includes a high-resolution scanning electron microscope, a Fourier transform infrared spectrometer and an aging characteristic multimodal analysis module; the high-resolution scanning electron microscope is used to bombard the sample surface with an electron beam and generate secondary electrons and backscattered electrons on the sample surface, and at the same time, the secondary electron detector captures the surface secondary electron signal and converts it into image brightness information, thereby generating a high-resolution surface morphology image, and then transmitting the generated surface morphology image to the aging characteristic multimodal analysis module; The Fourier transform infrared spectrometer is used to emit interference light to the sample, and receive the light intensity signal with sample information through the detector, and then convert the interference light with sample information into an infrared spectrum through Fourier transform, and then transmit the generated infrared spectrum of the sample to the aging characteristic multimodal analysis module; The aging characteristic multimodal analysis module is respectively connected to a high-resolution scanning electron microscope and a Fourier transform infrared spectrometer, and is used for identifying and tracing microplastics based on surface morphology images and infrared spectra.

2. According to claim 1, a microplastic aging process traceability and factor intelligent identification system is characterized in that: The particle dispersion and electrostatic dust removal device also includes a waste box, the feed port of the waste box being connected to the waste ports of the plurality of cyclone separators (7) respectively through a plurality of connecting pipes.

3. A microplastic aging process traceability and factor intelligent identification system according to claim 2, characterized in that: The aging characteristic multimodal analysis module is an industrial computer.

4. A microplastic aging process tracing and factor intelligent identification system according to claim 3, characterized in that: The aging characteristic multimodal analysis module includes a data preprocessing module, a multimodal fusion and analysis module, a traceability and path tracking module, a model evaluation and optimization module, and a data storage and database module; the data preprocessing module, the multimodal fusion and analysis module, the traceability and path tracking module, and the model evaluation and optimization module are connected in sequence, and the data storage and database module is respectively connected to the data preprocessing module, the multimodal fusion and analysis module, the traceability and path tracking module, and the model evaluation and optimization module.

5. A microplastic aging process tracing and factor intelligent identification system according to claim 4, characterized in that: The multimodal fusion and analysis module includes a feature extraction layer, a feature splicing layer, a feature compression and fusion layer, and a classification prediction layer; the traceability and path tracking module includes a factor traceability unit and a path simulation unit; the model evaluation and optimization module includes an evaluation index calculation unit, an error analysis unit, an optimization algorithm unit, and a feedback unit; the model evaluation and optimization module includes an evaluation index calculation unit, an error analysis unit, an optimization algorithm unit, and a feedback unit; the data storage and database module includes an original data storage unit, a feature data storage unit, an analysis result storage unit, a model evaluation and performance recording unit, a model version control unit, a historical data management unit, and an environment database.

6. A microplastic aging process tracing and factor intelligent identification system according to claim 5, characterized in that: The feature extraction layer adopts a multimodal fusion feature extraction architecture that combines the ResNeXt structure and the CNN structure, wherein the ResNeXt structure is used to extract image features and the CNN structure is used to extract spectral features; the classification prediction layer includes two-level fully connected layers, dual activation functions and multiple output layers, which are used to accurately identify the main aging factors of microplastics and their combined effects, and at the same time, to realize the quantification of the overall aging process by different aging factors.

7. A method for tracing the source and identifying factors of microplastic aging process, using a system for tracing the source and identifying factors of microplastic aging process as claimed in any one of claims 1 to 6, characterized in that: The following steps are involved: Step 1: Initial rough selection of sample particle size range; The sample to be tested is placed in the sample receiving slot (1) in advance; The conveyor belt (2) and the vibrating screen (3) are started, and the sample receiving slot (1) is controlled to quantitatively lower the sample through the discharge port, and the lowered sample is uniformly transported to the vibrating screen (3) by using the conveyor belt (2). The vibrating screen (3) uses a screen with a specific aperture to perform a preliminary screening process on the sample, and screens out samples with smaller particle sizes; Step 2: Dry, remove dust and remove static electricity from the sample; A receiving funnel (13) is used to receive the sample with a smaller particle size screened by the vibrating screen (3), and the sample is transported to the disperser (6) through the drying pipeline (12). During the transportation process, the dry air ejector (4) and the static eliminator (5) are synchronously started. The dry air ejector (4) is used to eject a high-speed dry airflow to the feed section of the drying pipeline (12). The sample is blown away by the airflow and the sample is dried and dusted simultaneously. The static eliminator (5) is used to eliminate or reduce the accumulation of static electricity in the sample after drying. Step 3: Secondary refinement of the sample particle size range; The sample is stirred, dispersed and sieved by using a dispersing stirrer (6), and the samples in multiple particle size ranges obtained after sieving are respectively output to multiple discharge pipes (14); the samples output from the multiple discharge pipes (14) are respectively processed by using multiple cyclone separators (7), and the samples in a set particle size range are screened out and output under the centrifugal action of the cyclone separators (7); Step 4: Multi-level sorting of samples; A plurality of collecting tanks (8) are used to respectively receive samples of multiple particle size ranges output by a plurality of cyclone separators (7), and output the samples to a plurality of feed ports of an electrostatic separation channel (10); a high-voltage electrostatic charging device (9) is used to provide a high-voltage electric field to the samples passing through the feed channel; under the action of the high-voltage electric field, the microplastic particles in the samples are more strongly charged, while other non-microplastic particles are not charged or are less charged; the samples that have completed the charging process fall into the electrostatic separation channel; The electrostatic deflection plate is used to provide a transverse electric field, so that the trajectory of the charged sample falling is deflected under the action of the transverse electric field, and moves along the deflection direction to the area where different collection chambers are located. At the same time, the collection electrode is used to adsorb the charged sample in the collection chamber area and guide it into the corresponding sample collection chamber, wherein other particles that are not charged or have a weak charge fall in a straight line to the waste collection chamber, and each collection chamber discharges the collected sample through the corresponding discharge port; Using a plurality of microplastic collection slots in a microplastic collection device (11) to correspondingly receive samples of multiple particle size ranges discharged from a plurality of discharge ports; Step 5: Obtain the surface morphology image and infrared spectrum of the sample; Start a high-resolution scanning electron microscope, place a portion of the sample in the microplastic collection tank on the detection platform of the high-resolution scanning electron microscope, control the high-resolution scanning electron microscope to emit an electron beam and bombard the sample surface, and generate secondary electrons and backscattered electrons on the sample surface. At the same time, use a secondary electron detector to capture the surface secondary electron signal and convert it into image brightness information, thereby generating a high-resolution surface morphology image, and then transmit the generated initial surface morphology image to the aging feature multimodal analysis module and the data storage and database module; Start the Fourier transform infrared spectrometer, transfer the sample on the detection platform to the sample stage of the Fourier transform infrared spectrometer, use the Fourier transform infrared spectrometer to emit interference light to the sample, and at the same time, capture the light intensity signal passing through the sample or reflected by the sample through the detector, and then convert the light intensity signal with sample information into an initial infrared spectrum through Fourier transform, and then transmit the generated initial infrared spectrum of the sample to the aging characteristic multimodal analysis module and the data storage and database module; After receiving the initial surface morphology image and the initial infrared spectrum, the data storage and database module stores them in the original data storage unit and generates a unique sample ID to ensure the identification and consistency of the data; Step 6: Construct a prediction model for tracing the source and identifying factors of microplastic aging process; S61: After receiving the surface morphology image and infrared spectrum, the aging feature multimodal analysis module uses the data preprocessing module to perform data preprocessing. For the surface morphology image, the data preprocessing module normalizes all the obtained surface morphology images to a uniform size, and applies filtering technology to eliminate random noise in the image, reduce the model error caused by image noise, and make the texture feature clearer. At the same time, the image is rotated, scaled, and cropped through image enhancement technology to generate more variable sample data. The processed surface morphology image is transmitted to the multimodal fusion and analysis module and the data storage and database module. For the infrared spectrum, a baseline calibration is first performed. Positive, use the multi-point baseline correction algorithm to automatically adjust the baseline, eliminate background noise and drift, and then use the maximum normalization method to normalize the spectral data, scale the spectral data of all samples to the same level, and at the same time, apply the Savitzky-Golay smoothing filter technology to reduce the noise and smooth the spectral data to reduce the influence of random noise and smooth the spectral curve. The processed infrared spectral data is transmitted to the multimodal fusion and analysis module and the data storage and database module; after receiving the processed surface morphology image and the processed infrared spectral data, the data storage and database module stores them in the feature data storage unit; S62: Processing is performed using a multimodal fusion and analysis module; in the feature extraction layer, the ResNeXt structure is used to extract texture and morphological features from the surface morphology image, and the basic feature extraction of the infrared spectrum image is performed through the initial convolution layer of the ResNeXt structure to extract low-level features such as edges, simple textures, and color changes. At the same time, multi-layer group convolutions are used to improve hierarchical features. Group convolutions divide the input channels into multiple groups, and each group performs convolution operations independently to obtain rich feature representations; residual connections are used between convolution layers to add input and output to prevent gradient vanishing and improve training stability; after multi-layer convolutions, a feature map with high-level features is generated, and finally global average pooling is used to compress the feature map into a vector of fixed length to generate image features. vector, and pass it to the feature splicing layer and data storage and database module; at the same time, use the CNN architecture to extract the functional group features of the spectral data; extract the preliminary spectral peak features of the spectral data through the initial convolution layer of CNN; further extract the key spectral features through a multi-layer convolutional network, each layer of convolution focuses on the changes in different frequency ranges in the spectrum, and at the same time, captures the functional group peak and intensity changes; use maximum pooling or average pooling to reduce the data dimension while retaining key feature information. After the last layer of convolution output, use global average pooling to generate a fixed-length spectral feature vector and pass it to the feature splicing layer and data storage and database module; after receiving the image feature vector and spectral feature vector, the data storage and database module stores them in the feature data storage unit; Next, after receiving the image feature vector and the spectral feature vector, the feature concatenation layer directly connects the image feature vector and the spectral feature vector adjacent to each other through the Concatenation layer to form a joint vector containing all the information, and then inputs the joint vector into the feature compression and fusion layer; Then, after receiving the joint vector, the feature compression and fusion layer performs the first layer of full connection operation, performs preliminary compression on the joint feature vector, reduces the feature dimension from high dimension to a lower dimension, and performs linear transformation through the weight matrix to recombine the image and spectral information and filter out some irrelevant features; after the output of the fully connected layer, a nonlinear activation function is applied to enable the model to learn the nonlinear relationship between the image and spectral features, so that the model can model the complex associations between different modalities to better understand the implicit patterns in multimodal information; after the first layer of full connection operation, the Dropout layer is used to randomly remove some nodes to prevent the model from overfitting in the high-dimensional feature space and improve the generalization ability of the model; repeat the above process, perform multi-layer full connection fusion, further compress the joint feature dimension, gradually reduce the output dimension of each layer, and finally compress the high-dimensional joint feature into a low-dimensional representation; the output of the last layer of the fully connected network is a low-dimensional fused feature vector, which contains the compressed multimodal information, which is the final multimodal representation of the model, and then the low-dimensional fused feature vector is passed to the classification prediction layer; Finally, in the classification prediction layer, the input low-dimensional fused feature vector is linearly transformed and feature mapped through the first-level fully connected layer, so that the model can identify significant feature patterns related to the classification of aging factors; the first-level fully connected layer transforms the input vector through the weight matrix to extract features with classification directionality, thus laying the foundation for the preliminary classification of aging factors; then, these preliminary extracted features are processed through the Softmax activation function to convert the output of the model into a probability distribution, which is used to identify the dominant aging factor of the sample; the output of this stage provides intuitive labels for the most likely aging factors, so that the model can quickly identify the main aging sources of the samples; after determining the main aging factors, the feature vectors are further passed to the second-level fully connected layer, in which the model performs quantitative analysis on multimodal features to extract the specific contributions of different aging factors to the aging process; at this time, the second-level fully connected layer processes the neuron nodes and weights layer by layer, The input features are refined; the secondary layer uses the Sigmoid activation function to generate probability values ​​for each aging factor one by one, and the probability value represents the possible contribution of each factor to the aging of the sample; the Sigmoid function converts the output value of each aging factor into an independent probability, thereby quantifying the influence of multiple aging factors in the combination mode; finally, the output of the classification prediction layer includes two parts: one is the dominant aging factor label generated by Softmax, and the other is the combined probability distribution of each aging factor; the dominant aging factor label and the combined probability distribution of each aging factor are sent to the traceability and path tracking module and the data storage and database module; after receiving the dominant aging factor label and the combined probability distribution of each aging factor, the data storage and database module stores them in the analysis result storage unit; the analysis structure storage unit associates the dominant aging factor label and the combined probability distribution of each aging factor with the original sample data according to the sample ID; S63: Processing using the traceability and path tracking modules; first, the factor traceability unit is used to prioritize the dominant aging factor labels and identify the dominant factor labels. At the same time, the factor traceability unit analyzes and processes the combined probability distribution of each aging factor, sorts them according to the probability values ​​of different factors, and confirms the priority of each factor in combination with the typical aging characteristics of the environmental database; the factor traceability unit matches the processed dominant labels and probability distributions with the environmental database, which contains the typical aging characteristics of microplastics in different environments; the factor traceability unit uses the dominant factor labels as preliminary screening conditions to initially narrow the traceability scope and find possible environmental types in the database, and then further compares the group Combined probability distribution is used to refine the selection; after matching the potential environment type, the factor tracing unit further calculates the weights of each aging factor to reflect the specific impact of each factor on the aging process of the current sample; the weight calculation is based on the value of each factor in the combined probability distribution, combined with the typical weight range in the environment in the database; the factor tracing unit adjusts these weights according to the specific values ​​in the combined probability distribution to ensure that the generated weights can reflect the actual aging pattern of the sample in the environment; the factor tracing unit generates one or more possible potential source environments based on the results of feature matching and the output of weight calculation, and sends them to the path simulation unit; the model gives priority to the environment type with the highest matching degree, and provides matching scores for different environment types; When the path simulation unit receives information about the potential source environment, the model loads the corresponding migration conditions and selects a suitable migration model, and then selects different path modeling methods according to the different source environments. During the model construction process, the path simulation unit uses nodes and edges to represent the possible path nodes of microplastics and the relationship between adjacent nodes to form a migration network; after the model is built, the path simulation unit integrates the aging factor weights received from the factor tracing unit into the migration model to adjust the priority of the migration path. The weight integration process is carried out through the edges or nodes in the weighted path model, so that the model can more accurately reflect the diffusion trend of microplastics under the influence of multiple factors; after integrating the weights of the aging factors, the path simulation unit The migration path is optimized, the total migration probability and matching degree of each path are calculated, and the path is dynamically adjusted according to the influence of the aging factor weight. The state of the nodes in the path is recursively calculated, and the node sequence on the path is optimized to make it more consistent with the actual migration pattern. The local node selection of the path is re-evaluated to ensure that the path meets the environmental conditions and the action mode of aging factors both globally and locally. The optimized path will be integrated into a path map data structure, which contains node and edge information. Each node represents a potential position of microplastics in the migration process, and the edge represents the migration direction and probability between adjacent positions. Finally, the path simulation unit sends the path map data structure to the model evaluation and optimization module. S64: When the model evaluation and optimization module receives the output data of the path simulation unit, it first uses the evaluation index calculation unit to compare the received data with the known true label to obtain the evaluation result data of the model, the evaluation result includes the calculated key performance indicators, and then sends the evaluation result to the data storage and database module; after receiving the evaluation result, the data storage and database module stores it in the model evaluation and performance recording unit after adding a timestamp; The error analysis unit further analyzes the model's error prediction and obtains the error analysis results. By associating the errors with specific environments and factor distribution characteristics, a specific direction is provided for the model's improvement. At the same time, the error analysis results are sent to the optimization algorithm unit. The optimization algorithm unit selects a suitable optimization strategy based on the error analysis results, adjusts the model parameters, weights and model structure, and uses an automated hyperparameter tuning process to improve the prediction accuracy of the model under a specific combination of aging factors to obtain an optimized model, and sends the optimized model information to the data storage and database module and the feedback unit; after receiving the optimized model information, the data storage and database module stores it in the model version control unit; the feedback unit feeds the optimized model back to the multimodal fusion and analysis module and the traceability and path tracking module, and performs data analysis and traceability tests again until the error analysis result obtained by the error analysis unit is less than the set threshold, and a prediction model for tracing the microplastic aging process and identifying factors is obtained; Step 7: Conduct real-time analysis and prediction of microplastics, and output the analysis and prediction results; According to the process in step one to step five, an initial surface morphology image and an initial infrared spectrum are obtained, and the initial surface morphology image and the initial infrared spectrum are input as input data to a prediction model for tracing the source and identifying factors of the microplastic aging process. The prediction model for tracing the source and identifying factors of the microplastic aging process is used to perform analysis and prediction to obtain analysis and prediction results. At the same time, the analysis and prediction results are output and sent to the data storage and database module. After receiving the analysis and prediction results, the data storage and database module stores them in the historical data management unit.

8. The monitoring method for predicting the magnitude of anchoring force based on surface deformation according to claim 7 is characterized in that: In step four, the electric field strength of the electrostatic deflection plate is adjusted using an electric field regulator to meet the needs of separating microplastic particles with different charges and effectively ensure the separation accuracy.

Images