Water quality monitoring method based on blockchain enabling pollutant fingerprint cooperative authentication

Abstract

The invention discloses a water quality monitoring method based on blockchain enabling pollutant fingerprint cooperative authentication, and relates to the field of water pollution monitoring, and the method comprises the steps: collecting fingerprint data of a water body; preprocessing the fingerprint data into feature data; performing deviation judgment on the Hash value of the fingerprint data and a preorder record, if the deviation exceeds a threshold value, executing calibration, otherwise, writing the feature data into a Hash tree; feature data are loaded into a pre-trained migration model, the input data dimension corresponds to the dimension of the feature data, the migration model is pre-trained corresponding to the target water body type, and the migration model outputs a pollution source probability matrix and a confidence score; according to the method, the data is ensured not to be tampered through the block chain, automatic quality control is realized by using the intelligent contract, and the effect of'one-time calibration and global application 'in complex water body monitoring is achieved.

Description

Technical Field

[0001] The present invention relates to the field of water pollution monitoring, and particularly to a water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints. Background Art

[0002] For complex water bodies, they may include rivers or industrial wastewaters with high turbidity, multiple ions, and rich organic matter; secondly, there are "multiple pollutant" combination types, such as heavy metals + organic matter, nitrogen and phosphorus nutrients + pesticides, etc.; finally, there is the "mutual interference" mechanism, such as electrode pollution, spectral overlap, biological toxicity superposition effect, etc. In the direction of multi-technology integration, it is proposed that a spectro-electrochemical combined instrument can simultaneously avoid turbidity interference and electrode passivation problems; however, physically, there is a spatial conflict between the optical path and the circuit. When the laser needs to penetrate the electrolytic cell to irradiate the electrode surface, it is necessary to ensure the light transmittance of the light window and maintain the active area of the electrode at the same time; the bubbles generated during high-frequency pulse cleaning of the electrode will directly distort the spectral signal. Among them, pulse cleaning can relieve passivation but exacerbates signal fluctuations; the turbidity compensation algorithm has to sacrifice resolution. Therefore, it is urgent to improve the applicability and reliability of monitoring technologies in complex water body environments; to avoid mutual interference between pollutants, it is necessary to comprehensively apply a variety of technical strategies and management means. Summary of the Invention

[0003] The present invention provides a water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints to solve the problems of the prior art.

[0004] In a first aspect, the present invention provides a water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints, including the following steps: Collect fingerprint data of the water body; Preprocess the fingerprint data into feature data; Perform deviation discrimination on the hash value of the fingerprint data and the previous record. If the deviation exceeds the threshold, calibration is performed; otherwise, the feature data is written into the hash tree; Load the feature data into a pre-trained transfer model. The input data dimension corresponds to the dimension of the feature data. The transfer model is pre-trained for the target water body type, and the transfer model outputs a pollution source probability matrix and a confidence score; Among them, the elements of the pollution source probability matrix represent the probability of the pollution source contributing pollutants.

[0005] Further, the collection of fingerprint data of the water body includes: spectral fingerprint data, electrochemical time-series data, and biological response fingerprint data.

[0006] Further, the preprocessing of the fingerprint data into feature data includes: Fourier-transforming the spectral fingerprint data into a 128-dimensional feature vector, implementing wavelet denoising coding on the electrochemical fingerprint data into a 96-dimensional feature matrix, and transforming the biological response fingerprint data into a DNA barcode through PCR amplification technology; among them, the electrochemical time series data generates electrochemical fingerprint data after feature extraction.

[0007] Further, the deviation discrimination of the hash value of the fingerprint data from the previous record is performed. If the deviation exceeds the threshold, calibration is executed; otherwise, the feature data is written into the hash tree, specifically including: If the deviation of the SHA-3 hash value of the fingerprint data from the previous record exceeds the threshold, the calibration instruction is executed. If the deviation is within the threshold range, the feature vector of the fingerprint data is written into the hash tree and a timestamp is generated for evidence storage. The original fingerprint data is stored using the IPFS protocol; Among them, the previous data is the reference fingerprint data of the same monitoring point in the previous calibration period, and the reference fingerprint data depends on the time stability of the water pollutants and is the same as the corresponding previous record.

[0008] Further, the deviation discrimination of the hash value of the fingerprint data from the previous record is performed. If the deviation exceeds the threshold, calibration is executed; otherwise, the feature data is written into the hash tree, where the calibration includes: When the detected turbidity is greater than 200 NTU, the following steps are automatically executed: S1. Switch the sensing mode to anti-interference pulse excitation; The technical effect is to reduce the diffusion layer thickness and inhibit the adsorption of suspended substances; S2. Call the adjacent 3 monitoring nodes for cross-verification; Cross-verification mechanism: The adjacent nodes will synchronously detect the same water body. When the correlation coefficient between the data of the master node and the data of ≥ 2 adjacent nodes is lower than 0.85, calibration is triggered.

[0009] The cross-verification process includes: S21. Broadcast a verification request to the adjacent nodes within a geographical radius of 1 km; S22. The receiving nodes synchronously collect data and return the hash value; S23. Calculate the consistency index. If the consistency index κ ≥ 0.67, the original data is accepted.

[0010] S3. Send ERC-20 based on the cross-verification result; The essence of ERC-20 is to maintain an incentive mechanism. After calibration, it can be exchanged for cash or the priority right to use equipment, that is, to exchange for equipment maintenance services or offset monitoring service fees; Further, among them, for the migration model setting, it includes: Deploy edge computing nodes around the monitoring points. The nodes run lightweight CNN models to classify pollutant types in real time; Periodically aggregate the gradients of each node on the cloud server to update the global model weights.

[0011] Furthermore, it includes: settings for the training of the transfer model: Preset value. Add conditional normalization to the last layer of the transfer model, and input the learnable water type embedding vector into the normalization process; Pre-training. Set the pre-training objective function as the cross-entropy loss for pollutant classification, and use a multi-basin mixed data training set for training; Fine-tune the model. Collect multiple groups of labeled samples in the target water area, set the objective function of fine-tuning as the contrastive loss with domain adaptation, and perform training and fine-tuning.

[0012] Furthermore, among them, biological data is used as an independent evidence chain to cross-verify with chemical data, that is, DNA barcodes and electrochemical fingerprint data, spectral fingerprint data. The sensor directly uploads the DNA barcode to the blockchain through the IPFS protocol. When the transfer model runs the biotoxicity model and sends the detected toxicity index to the blockchain, when it is queried that the biotoxicity index exceeds the standard, the regulatory agency retrieves the DNA barcode evidence package on the blockchain.

[0013] Furthermore, regarding the loss function used in the training of the transfer model, it includes: Cross-entropy loss function for pollutant classification, including: Multi-label design corresponding to pollutant types; The weights of different categories are positively correlated with the pollutant toxicity level; The sigmoid function is used in the output layer to achieve multi-label prediction; Contrastive loss function with domain adaptation, including: Positive sample pairs: Feature vectors of the same pollutant in different water areas; Negative sample pairs: Feature vectors of different pollutants in the same water area; Use parameter adjustment to measure the distance in the feature space for clustering similar pollutants across domains; Embed the features learned by the contrastive loss function with domain adaptation as the feature input of the cross-entropy loss function for pollutant classification, and jointly optimize through gradient coupling.

[0014] The present invention ensures the immutability of data through the blockchain, uses smart contracts to achieve automated quality control, and achieves the effect of "one calibration, universal application" in complex water body monitoring.

[0015] The transfer model based on federated learning in the present invention breaks through the environmental difference limitations and provides monitoring adaptation accuracy for different water areas. Description of the Drawings

[0016] The accompanying drawings described herein are used to provide a further understanding of the embodiments of the present invention, which form a part of the present invention and do not limit the embodiments of the present invention. In the drawings: Figure 1 It is a flowchart of a water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints provided for an exemplary embodiment of the present invention.

[0017] Figure 2 It is a flowchart of a method automatically executed when the turbidity is monitored to be greater than 200 NTU in a water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints provided for an exemplary embodiment of the present invention. Detailed implementation manners

[0018] Here, the exemplary embodiments will be described in detail, and the examples are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0019] For the present invention, considering various water body environments covering from plateau lakes to industrial estuaries, first, the technical concept is as follows. The spectral and electrochemical raw data formats collected by different devices vary greatly, and there are different systematic errors. Therefore, the present invention considers designing a unified digital fingerprint coding specification. To adapt to the subsequent processing rules, the multi-modal data is fused into a 128-dimensional feature vector, which not only retains key information but also is convenient for storage; Considering processing through a consortium blockchain, which is superior to a public blockchain, taking environmental protection agencies, research institutions, and enterprises as nodes to ensure data authority and improve the efficiency of decentralization. For abnormal data, set it to automatically trigger a confidence-weighted arbitration algorithm; Improve the recognition accuracy by training the knowledge graph of pollutants offline, and adopt a lightweight model for edge processing to diagnose data in real time online.

[0020] The specific application scenario of the present invention is the monitoring of water pollution in complex water bodies. The present invention combines the blockchain-based evidence storage mechanism of pollutant fingerprints, the dynamically calibrated intelligent contract drive, and the cross-water area migration ability of the large model to jointly solve the problem of complex water body monitoring.

[0021] A water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints provided by the present invention aims to solve the above technical problems of the prior art.

[0022] The technical solution of the present invention and how the technical solution of the present invention solves the above technical problems will be described in detail below with specific embodiments. These several specific embodiments below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present invention will be described below in conjunction with the accompanying drawings.

[0023] Embodiment 1: This embodiment provides a water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints, as Figure 1 、 Figure 2 shown, including the following steps: A1. Collect fingerprint data of the water body; Including: spectral fingerprint data, electrochemical time series data, biological response fingerprint data; A2. Preprocess the fingerprint data into feature data; Fourier transform the spectral fingerprint data into a 128-dimensional feature vector, perform wavelet noise reduction coding on the electrochemical fingerprint data into a 96-dimensional feature matrix, and convert the biological response fingerprint data into a DNA barcode through PCR amplification technology; among them, the electrochemical time series data generates electrochemical fingerprint data after feature extraction. Specifically: the electrochemical time series data is processed by wavelet noise reduction, and parameters such as peak current (i_p), peak potential (E_p), and peak separation (ΔE_p) are extracted to form an electrochemical feature matrix, that is, electrochemical fingerprint data.

[0024] Regarding the selection of feature dimensions, the two values of 128 and 96 come from the balance of actual sensor resolution and information entropy: a UV-VIS spectrometer usually outputs 2048 original data points. Based on the standard resolution of the UV-VIS spectrometer (2nm interval, a total of 301 original data points in the 200-800nm range), after Fourier transform compression, the first 128 low-frequency components are retained, which can cover more than 95% of the feature information; and the typical sampling points of the cyclic voltammetry data collected by an electrochemical workstation are 1024. After threshold filtering by wavelet transform, 96 dimensions can optimally retain the redox characteristic peaks. Among them, 32 characteristic peaks × 3 parameters = 96 dimensions, and when it exceeds 96 dimensions, the risk of model overfitting increases (the cross-validation accuracy rate drops by ≥ 3%).

[0025] A3. Judge the deviation of the hash value of the fingerprint data from the previous record. If the deviation exceeds the threshold, perform calibration; otherwise, write the feature data into the hash tree; Specifically including: If the deviation of the SHA-3 hash value of the fingerprint data from the previous record exceeds the threshold, execute the calibration instruction. If the deviation is within the threshold range, write the feature vector of the fingerprint data into the hash tree and generate a timestamp for certification; Use the IPFS protocol to store the original fingerprint data; Among them, the pre-order data is the reference fingerprint data of the same monitoring point in the previous calibration period, and the reference fingerprint data is the same as the corresponding previous record depending on the time stability of the water pollutants.

[0026] Previous record: The median of the fingerprint hash values of the sliding window of the previous 7 days at the same monitoring point; That is, relative to the duration of the calibration period, the water pollutants can be regarded as stationary. In the two adjacent previous and subsequent periods, at the same time, the regularization term of the objective function under the time stability assumption rule of the water pollutants constitutes the core of transfer learning. This term restricts the new water area model parameters not to deviate too far from the basic model to avoid overfitting.

[0027] Calibration includes: Deploy a dynamic calibration sub-contract on the Ethereum virtual machine and set it in the smart contract unit. The measurement source is a 90° scattering turbidimeter (ISO 7027 standard) integrated in the sensor module. Trigger logic: Activate the contract when the turbidity value of 5 consecutive samples > 200 NTU; When the turbidity greater than 200 NTU is detected, the following steps are automatically executed: S1. Switch the sensing mode to anti-interference pulse excitation; Switch the potentiostatic mode to: Pulse width: 50 ms; Interval time: 20 ms; Potential step: +0.2V → -0.3V (vs. Ag / AgCl); The technical effect is to reduce the diffusion layer thickness and inhibit the adsorption of suspended matter; S2. Call 3 adjacent monitoring nodes for cross-validation; Cross-validation mechanism: Adjacent nodes will synchronously detect the same water body. When the correlation coefficient between the data of the master node and the data of ≥ 2 adjacent nodes is lower than 0.85, calibration is triggered.

[0028] The cross-validation process includes: S21. Broadcast a verification request to adjacent nodes within a geographical radius of 1 km; S22. The receiving node synchronously collects data and returns the hash value; S23. Calculate the consistency index. If the consistency index κ ≥ 0.67, accept the original data.

[0029] S3. Send ERC-20 based on the cross-validation result.

[0030] The essence of ERC-20 is to maintain an incentive mechanism. After calibration, it can be exchanged for cash or the priority right to use equipment, that is, to exchange for equipment maintenance services or offset monitoring service fees.

[0031] A4. Load the feature data into the pre-trained transfer model. The input data dimension corresponds to the dimension of the feature data. The transfer model is pre-trained for the target water body type, and the transfer model outputs a pollution source probability matrix and a confidence score. The transfer learning model adapts the domain adaptation layer according to the target water body type and dynamically adjusts the model parameters. The objective function of the transfer model is set based on the federated learning loss and the domain adaptation regularization term.

[0032] Among them, the elements of the pollution source probability matrix represent the probability of the pollution source contributing pollutants.

[0033] Among them, for the transfer model settings, including: Deploy edge computing nodes around the monitoring points. The nodes run a lightweight CNN model to classify the pollutant types in real time. Periodically aggregate the gradients of each node on the cloud server to update the global model weights.

[0034] For the training settings of the transfer model: Preset value. Add conditional normalization to the last layer of the transfer model, and input the learnable water body type embedding vector into the normalization process. Pre-training. Set the pre-training objective function as the cross-entropy loss for pollutant classification, and use a multi-basin mixed data training set for training. Fine-tune the model. Collect multiple groups of labeled samples in the target water area, set the objective function of fine-tuning as the contrastive loss with domain adaptation, and perform training fine-tuning.

[0035] Regarding the loss function used in the training of the transfer model, including: The cross-entropy loss function for pollutant classification, including: Multi-label design corresponding to pollutant types; The weights of different categories are positively correlated with the pollutant toxicity level; The output layer uses sigmoid to achieve multi-label prediction; The contrastive loss function with domain adaptation, including: Positive sample pairs: Feature vectors of the same pollutant in different water areas; Negative sample pairs: Feature vectors of different pollutants in the same water area; Use parameter adjustment to measure the distance in the feature space for clustering similar pollutants across domains; Take the feature embedding learned by the contrastive loss function with domain adaptation as the feature input of the cross-entropy loss function for pollutant classification, and jointly optimize through gradient coupling.

[0036] Preferably, when biological data serves as an independent evidence chain, it can be cross-verified with chemical data, that is, DNA barcodes are cross-verified with electrochemical fingerprint data and spectral fingerprint data. The sensor directly uploads the DNA barcode to the blockchain through the IPFS protocol. When the migration model runs the biotoxicity model and sends the detected toxicity index to the blockchain, when it is queried that the biotoxicity index exceeds the standard, the regulatory agency retrieves the DNA barcode evidence package on the blockchain.

[0037] Example 2: Based on Example 1, a specific instance of step A3 is that if the SHA-3 hash value of the fingerprint data deviates from the previous record by more than the threshold, the calibration instruction is executed, that is, when the smart contract detects that |H_{new} - H_{baseline}|>5%: the threshold is 5%.

[0038] Send a calibration instruction to the device and update the new reference hash value to the hash standard response curve; Input data: Digital fingerprint after feature extraction (128-dimensional vector + 96-dimensional matrix); Hash algorithm: SHA3-256; Tree structure construction: def build_merkle_tree(hashes): if len(hashes) == 1: return hashes[0] new_level = [] for i in range(0, len(hashes), 2): left = hashes[i] right = hashes[i+1] if i+1<len(hashes) else left new_level.append(sha3(left + right)) # Node merging hash return build_merkle_tree(new_level) Among them, the content uploaded to the chain: only the root hash value (fixed 256 bits) and the timestamp; Realize double anti-tampering: At the data level: Modify the original file → The CID changes → It does not match the record on the chain; At the structural level: Modify any feature → The root hash changes → The blockchain consensus rejects; To verify data integrity, only the hash path needs to be provided, and there is no need to download the complete original data (only about 1KB of proof data is required).

[0039] In several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there can be other division methods. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0040] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical modules, that is, they can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0041] In addition, in each embodiment of the present invention, the functional modules can be integrated in a processing module, or each module can exist physically alone, or two or more modules can be integrated in one module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of a combination of hardware and software functional modules.

[0042] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods or systems. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects.

[0043] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, commodity or device. Without further limitation, the element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, commodity or device including the element.

[0044] The above are only the embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the scope of the claims of the present invention.

[0045] Other embodiments of the present invention will be readily apparent to those skilled in the art upon consideration of the specification and examples disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include known common general knowledge or conventional technical means in the technical field not disclosed by the present invention. The specification and examples are only to be considered as exemplary, and the true scope and spirit of the present invention are pointed out by the appended claims above.

[0046] It should be understood that the present invention is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A water quality monitoring method based on blockchain-enabled pollutant fingerprint collaborative authentication, characterized in that: It includes the following steps: Collect the fingerprint data of the water body; Preprocess the fingerprint data into feature data; Perform deviation discrimination on the hash value of the fingerprint data and the previous record. If the deviation exceeds the threshold, calibration is executed. Otherwise, the feature data is written into the hash tree; Load the feature data into the pre-trained transfer model. The input data dimension corresponds to the dimension of the feature data. The transfer model is pre-trained for the target water body type, and the transfer model outputs the pollution source probability matrix and the confidence score; Among them, the elements of the pollution source probability matrix represent the probability of the pollution source contributing pollutants.

2. The water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints according to claim 1, wherein The collection of the fingerprint data of the water body includes: spectral fingerprint data, electrochemical time series data, and biological response fingerprint data.

3. The water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints according to claim 2, wherein The preprocessing of the fingerprint data into feature data includes: Fourier transforming the spectral fingerprint data into a 128-dimensional feature vector, performing wavelet denoising encoding on the electrochemical fingerprint data into a 96-dimensional feature matrix, and converting the biological response fingerprint data into a DNA barcode through PCR amplification technology; among them, the electrochemical time series data generates electrochemical fingerprint data after feature extraction.

4. The water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints according to claim 3, characterized in that, The deviation discrimination of the hash value of the fingerprint data and the previous record. If the deviation exceeds the threshold, calibration is executed. Otherwise, the feature data is written into the hash tree, specifically including: If the deviation between the SHA-3 hash value of the fingerprint data and the previous record exceeds the threshold, the calibration instruction is executed. If the deviation is within the threshold range, the feature vector of the fingerprint data is written into the hash tree and a timestamp is generated for evidence storage; Use the IPFS protocol to store the original fingerprint data; Among them, the previous data is the reference fingerprint data of the same monitoring point in the previous calibration cycle, and the reference fingerprint data depends on the time stability of the water body pollutants and is the same as the corresponding previous record.

5. The water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints according to claim 4, wherein, The deviation discrimination of the hash value of the fingerprint data and the previous record. If the deviation exceeds the threshold, calibration is executed. Otherwise, the feature data is written into the hash tree, where the calibration includes: When the measured turbidity is greater than 200 NTU, the following steps are automatically executed: S1. Switch the sensing mode to anti-interference pulse excitation; S2. Call the adjacent 3 monitoring nodes for cross-validation; S3. Send ERC-20 based on the cross-validation results.

6. The water quality monitoring method based on blockchain empowered collaborative authentication of pollutant fingerprints according to claim 5, wherein, Among them, For the transfer model settings, it includes: Deploy edge computing nodes around the monitoring point. The nodes run a lightweight CNN model to classify the pollutant types in real time; Periodically aggregate the gradients of each node on the cloud server to update the global model weights.

7. The water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints according to claim 6, wherein It includes: For the training settings of the transfer model: Preset value, add conditional normalization to the last layer of the transfer model, and input the learnable water body type embedding vector into the normalization process; Pre-training, set the pre-training objective function as the cross-entropy loss of pollutant classification, and use the multi-basin mixed data training set for training; Fine-tune the model, collect multiple groups of labeled samples of the target water area, set the objective function of fine-tuning as the contrast loss with domain adaptation, and perform training fine-tuning.

8. The water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints according to claim 3, wherein, Among them, Biological data is cross-verified with chemical data as an independent evidence chain, i.e., DNA barcodes are cross-verified with electrochemical fingerprint data and spectral fingerprint data. The sensor directly uploads the DNA barcode to the blockchain through the IPFS protocol. When the migration model runs the biotoxicity model and sends the detected toxicity index to the blockchain, when it is queried that the biotoxicity index exceeds the standard, the regulatory agency retrieves the DNA barcode evidence package from the blockchain.

9. The water quality monitoring method based on blockchain-enabled collaborative authentication of pollutant fingerprints according to claim 7, characterized in that, Regarding the loss function adopted in the training of the migration model, it includes: The pollutant classification cross-entropy loss function, including: The multi-label design corresponding to the pollutant type; The weights of different categories are positively correlated with the toxicity level of the pollutant; The sigmoid function is used in the output layer to achieve multi-label prediction; The contrastive loss function with domain adaptation, including: Positive sample pairs: The feature vectors of the same pollutant in different water areas; Negative sample pairs: The feature vectors of different pollutants in the same water area; The distance metric of the feature space is adjusted by parameters, which is used to cluster similar pollutants during cross-domain; The features embedded by the contrastive loss function with domain adaptation are used as the feature input of the pollutant classification cross-entropy loss function, and are jointly optimized through gradient coupling.

Images