Water quality monitoring method and equipment based on overwater and underwater multiple views
By constructing a multi-view matrix above and below water and combining it with noise reduction and contrastive learning techniques, the problem of noise interference in water quality monitoring was solved, achieving high-precision water quality assessment in complex environments and improving the accuracy and stability of water quality monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU MARITIME INST
- Filing Date
- 2026-01-22
- Publication Date
- 2026-05-12
AI Technical Summary
Existing water quality monitoring technologies suffer from severe noise interference and low signal-to-noise ratio in field environments, making sensor data susceptible to interference. Multi-view learning technology suffers from unstable shared feature representations in situations with strong heterogeneity and high noise levels above and below water, making it impossible to dynamically construct a graph structure that reflects the intrinsic correlation of water quality, resulting in inaccurate assessment results.
A monitoring method based on multiple views above and below water is adopted. By constructing a partially labeled multi-view water quality matrix, robust features are extracted by combining nonlocal mean denoising and shared contrastive learning. A graph structure is constructed using a multi-view collaborative water quality adversarial generator, and quality assessment is performed through a dynamic perception adversarial discriminator. Finally, a denoised graph convolutional network is used for node feature inference and updating.
It achieves highly accurate monitoring of water quality information in complex field environments, with significant noise reduction effect, improving the accuracy and stability of water quality assessment.
Smart Images
Figure CN122023892A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of water quality monitoring technology, and more specifically, relates to a water quality monitoring method and equipment based on multiple views above and below water, especially a method and multi-functional unmanned surface vessel equipped with multiple sensors to realize environmental monitoring above and below water. Background Technology
[0002] Water quality monitoring refers to the systematic monitoring and analysis of physical, chemical, and biological indicators in water bodies using scientific methods to assess water quality status, pollution levels, and trends. The monitoring scope is very broad, including unpolluted and polluted natural water bodies such as rivers, lakes, seas, groundwater, and various industrial wastewater. The main monitoring items can be divided into two categories: one is comprehensive indicators reflecting water quality status, such as temperature, color, turbidity, conductivity, suspended solids, dissolved oxygen, chemical oxygen demand (COD), and biochemical oxygen demand (BOD); the other is toxic substances, such as phenols, cyanides, arsenic, lead, chromium, cadmium, mercury, and organic pesticides. Real-time monitoring of river water quality using sensors is an effective means of protecting water resources and has been increasingly applied. However, the environment of open water bodies is complex and variable, and sensor data is easily interfered with by water flow, biological activity, and equipment noise, resulting in a low signal-to-noise ratio. Most existing water quality assessment models lack targeted noise reduction designs, have poor robustness to noise, leading to inaccurate extraction of key model features and large fluctuations and low accuracy in assessment results.
[0003] In addition, existing multi-view learning techniques are widely used in fields such as image classification and target recognition, but when directly applied to water quality monitoring, especially when processing multi-view data with strong heterogeneity and high noise levels, such as above-water and underwater data, there are often problems such as the sensitivity of shared feature representations to noise and the inability to dynamically construct graph structures that reflect the intrinsic relationships of water quality. Summary of the Invention
[0004] In view of the above-mentioned deficiencies of the prior art, the present invention provides a water quality monitoring method and equipment based on multiple views above and below water to overcome the shortcomings of shared feature representation being sensitive to noise and unable to dynamically construct graph structures reflecting the intrinsic correlation of water quality. Specifically, it includes the following technical solutions: The first aspect provides a water quality monitoring method based on multiple views above and below water, with the following specific steps: Step S1: Input surface optical images and underwater acoustic and chemical data, combine them with historical water quality data and corresponding water quality labels, and construct a partially labeled multi-view water quality matrix; Step S2: Denoise the multi-view water quality matrix using nonlocal mean denoising technology; Step S3: Extract noise-robust features using shared contrastive learning techniques to obtain a shared representation of water quality features; Step S4: Based on the water quality feature sharing representation, construct a graph structure through a multi-view collaborative water quality adversarial generator; Step S5: Evaluate the quality of the generated graph structure using a dynamic water quality discrimination analyzer; Step S6: Based on the structure of the water quality relationship diagram, make a judgment. If the structure of the water quality relationship diagram conforms to the structure of the ideal diagram, proceed to step S7; otherwise, feed back to step S4 to optimize the multi-view collaborative water quality adversarial generator. Step S7: Based on the optimized graph structure, perform node feature inference and update using a denoised graph convolutional network; Step S8: Output water pollution level classification and key water quality parameter inversion.
[0005] The second aspect provides a water quality monitoring equipment based on multiple views above and below water, namely a multi-functional unmanned surface vessel: The aforementioned multi-functional unmanned surface vessel (USV) includes a camera sensor mounted at a preset position on top of the hull, providing wide-angle shooting and high-definition imaging capabilities for comprehensive monitoring of the aquatic environment; a sonar sensor mounted at the center below the hull to ensure stable sound wave transmission and reception, effectively detecting underwater conditions; and a chemical sensor mounted on the side of the hull near the water surface, enabling real-time contact with water and accurate monitoring of chemical substances in the water. During actual monitoring, the USV navigates the waterway according to a preset route.
[0006] Preferably, the multifunctional unmanned surface vessel also integrates a preprocessing unit, a noise reduction unit, a prediction unit, and an evaluation unit. The preprocessing unit is used to input surface optical images and underwater acoustic and chemical data, and combine them with historical water quality data and corresponding water quality labels to construct a partially labeled multi-view water quality matrix. The noise reduction unit is used to reduce noise in the multi-view water quality matrix using nonlocal mean noise reduction technology; The prediction unit is used for node feature inference and updating through a denoised graph convolutional network; The assessment unit is used to output water pollution level classification and key water quality parameters.
[0007] A third aspect of the present invention provides a water quality monitoring and assessment device, the device comprising a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute any of the water quality monitoring methods based on above-water and underwater multi-view according to the instructions in the program code; A fourth aspect of the present invention provides a computer-readable storage medium for storing program code, which, when executed by a processor, implements the water quality monitoring method based on multiple views above and below water as described in any of the first aspects. Compared with existing technologies, the water quality monitoring method and equipment based on multiple views above and below water provided by this invention can effectively combine the complementary information of optical images above water and acoustic and chemical data below water, and can effectively reduce noise in water quality information, thereby achieving robust and highly accurate water quality monitoring. Attached Figure Description
[0008] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Appendix Figure 1 This invention provides a structural schematic diagram of a multi-functional unmanned surface vessel (USV) for water quality monitoring based on multiple views above and below water, with the positions of camera sensors, sonar sensors, and chemical sensors marked. Appendix Figure 2 A schematic diagram of a water quality monitoring method based on multiple views above and below water provided by the present invention; Appendix Figure 3 This is a schematic diagram of a water quality monitoring and assessment device based on multiple views above and below water, provided by the present invention. Detailed Implementation
[0009] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
[0010] Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments; based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0011] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It is understandable that certain well-known structures and descriptions in the accompanying drawings may be omitted.
[0012] The following describes in detail, with reference to the accompanying drawings, a specific implementation scheme of a water quality monitoring method and equipment based on multiple views above and below water provided by the present invention.
[0013] Appendix Figure 2 This invention demonstrates a water quality monitoring method based on multiple views above and below water, specifically including the following steps: Step S1: Input surface optical images and underwater acoustic and chemical data, combine them with historical water quality data and corresponding water quality labels, and construct a partially labeled multi-view water quality matrix. .
[0014] Where n represents the total number of sample points collected. Let R be the feature dimension, and let R represent the real number field.
[0015] Step S2: Denoise the multi-view water quality matrix using nonlocal mean denoising technology. The formula for nonlocal mean denoising is: ; in, For the multi-view water quality matrix of the v-th view, The denoised water quality matrix for the v-th view. The search window represents pixel i. This represents the weight between pixel i and pixel j. Let represent the square of the Euclidean distance between pixel i and pixel j. is the normalization factor, and h is the smoothing parameter.
[0016] Step S3: Extract noise-robust features using shared contrastive learning technique to obtain a shared representation of water quality features. The optimization objective function of the shared contrastive learning is: ; Where Z represents the shared water quality characteristics. , The denoised water quality matrix for the v-th view. Let v be the projection matrix of the v-th view. Let Frobenius norm be Ω, Ω be the set of labeled samples, and the function f(·) represent the difference between the labels and the actual values of the samples. Let represent a graph convolutional network, represent a multi-view collaborative water quality adversarial generator, and Y represent the set of label matrices corresponding to the denoised water quality matrix. This is the sum of the exponential similarities of all positive sample pairs (i,j). This is the sum of the exponential similarities of all positive and negative sample pairs. It is a vector and The inner product (dot product) of . These are the water quality feature sharing and comparison learning vectors for samples i and j, respectively. Let P be the temperature parameter, P be the set of positive sample pairs, and N be the set of negative sample pairs. β is a hyperparameter used to balance the importance of reconstruction and classification tasks. β is a regularization parameter. The objective function is minimized by gradient descent to update the shared representation Z and related model parameters.
[0017] Step S4: Based on the shared representation of water quality features, a graph structure is constructed by a multi-view collaborative water quality adversarial generator. The formula can be expressed as: ; in, It is a concatenated vector of shared features from samples i and j. It is a multilayer perceptron. The sigmoid function maps the output to the (0,1) interval, representing the probability or strength of the correlation between samples.
[0018] Step S5: The dynamic water quality adversarial discriminator determines and generates a high-quality water quality relationship graph structure. The optimization objective function of this process can be expressed as: ; in, For the discriminator scoring, A is a graph structure generated based on the shared representation of water quality features. Let be the adjacency matrix of the v-th view constructed using the k-nearest neighbor algorithm.
[0019] Step S6: Based on the structure of the water quality relationship diagram, a judgment is made. If the structure of the water quality relationship diagram matches the structure of the ideal diagram, proceed to step S7; otherwise, feedback is given to step S4 to optimize the multi-view collaborative water quality adversarial generator. The logic for the selection judgment is as follows: Calculate the discriminator's average score D(A) for the generated graph structure A; if D(A) > ( If a preset threshold (e.g., 0.7) is set, the image quality is considered acceptable, and the process proceeds to step S7; otherwise, the result (i.e., ...) is evaluated. The loss signal is backpropagated to update the parameters of the multi-view collaborative water quality adversarial generator, and then the process returns to step S4 to regenerate the graph structure A.
[0020] Step S7: Perform inference using the optimized graph structure A for the denoised graph convolutional network. The inference formula is as follows:
[0021] Among them, among them, These are the node features of the l-th layer. , It is an adjacency matrix with self-connections. It is its degree matrix. It is a noise reduction function specifically designed for water quality data. This function sparsifies and reduces noise by thresholding and entropy constraints on edge weights. It is a trainable weight matrix. As the activation function, the output features are calculated through forward propagation. And update network weights through backpropagation using label data. .
[0022] Step S8: Based on the node features output by the graph convolutional network, the probability distribution of pollution levels is calculated through a classifier and a parallel multi-task regression output layer. The output includes pollution level classification and key water quality parameters. The water pollution level classification formula is as follows: ; in This is a probability matrix for pollution levels. For classifier weights, The node features generated by the denoising graph convolutional network result in the final classification. The classification levels include Class I (excellent), Class II (good), Class III (lightly polluted), Class IV (moderately polluted), and Class V (heavily polluted); Key water quality parameter inversion values: The values obtained from the fully connected layer in the convolutional network are input into a multi-task regression output layer. The output of this layer is the predicted concentration value of each water quality parameter, including chemical oxygen demand (COD), ammonia nitrogen (NH3-N), total phosphorus (TP), and dissolved oxygen (DO).
[0023] like Figure 1 As shown, a multi-view water quality monitoring device based on above-water and underwater views, namely a multi-functional unmanned surface vessel (USV), includes: a camera sensor 1 installed at a preset position on the upper part of the hull, equipped with wide-angle shooting and high-definition imaging capabilities, enabling all-round monitoring of the aquatic environment; a sonar sensor 3 installed at the center position below the hull, ensuring the stability of sound wave transmission and reception, and effectively detecting underwater conditions; and a chemical sensor 2 installed on the side of the hull near the water surface, capable of real-time contact with water and accurate monitoring of chemical substances in the water. During actual monitoring, the USV navigates in the water according to a preset route.
[0024] like Figure 3 As shown, the multifunctional unmanned surface vessel of the present invention also integrates a preprocessing unit, a noise reduction unit, a prediction unit, and an evaluation unit; wherein, the preprocessing unit is used to input surface optical images and underwater acoustic and chemical data, and combine them with historical water quality data and corresponding water quality labels to construct a partially labeled multi-view water quality matrix. The noise reduction unit is used to reduce noise in the multi-view water quality matrix using nonlocal mean noise reduction technology; The prediction unit is used for node feature inference and updating through a denoised graph convolutional network; The assessment unit is used to output water pollution level classification and key water quality parameters.
[0025] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.
Claims
1. A water quality monitoring method based on multiple views above and below water, characterized in that, Includes the following steps: Step S1: Input surface optical images and underwater acoustic and chemical data, combine them with historical water quality data and corresponding water quality labels, and construct a partially labeled multi-view water quality matrix; Step S2: Denoise the multi-view water quality matrix using nonlocal mean denoising technology; Step S3: Extract noise-robust features using shared contrastive learning techniques to obtain a shared representation of water quality features; Step S4: Based on the water quality feature sharing representation, construct a graph structure through a multi-view collaborative water quality adversarial generator; Step S5: Evaluate the quality of the generated graph structure using a dynamic water quality discrimination analyzer; Step S6: Based on the structure of the water quality relationship diagram, make a judgment. If the structure of the water quality relationship diagram conforms to the structure of the ideal diagram, proceed to step S7; otherwise, feed back to step S4 to optimize the multi-view collaborative water quality adversarial generator. Step S7: Based on the optimized graph structure, perform node feature inference and update using a denoised graph convolutional network; Step S8: Output water pollution level classification and key water quality parameter inversion.
2. The water quality monitoring method based on multiple views above and below water as described in claim 1, characterized in that, In step S2, the formula for denoising the multi-view water quality matrix using nonlocal mean denoising technology is as follows: ； in, For the multi-view water quality matrix of the v-th view, The denoised water quality matrix for the v-th view. The search window represents pixel i. This represents the weight between pixel i and pixel j. Let represent the square of the Euclidean distance between pixel i and pixel j. is the normalization factor, and h is the smoothing parameter.
3. The water quality monitoring method based on multiple views above and below water as described in claim 1, characterized in that, The optimization objective function for the shared contrastive learning described in step S3 is: ； Where Z represents the shared water quality characteristics. , The denoised water quality matrix for the v-th view. Let v be the projection matrix of the v-th view. Let Frobenius norm be Ω, Ω be the set of labeled samples, and the function f(·) represent the difference between the labels and the actual values of the samples. Let represent a graph convolutional network, represent a multi-view collaborative water quality adversarial generator, and Y represent the set of label matrices corresponding to the denoised water quality matrix. This is the sum of the exponential similarities of all positive sample pairs (i,j). This is the sum of the exponential similarities of all positive and negative sample pairs. It is a vector and The inner product, These are the water quality feature sharing and comparison learning vectors for samples i and j, respectively. Let P be the temperature parameter, P be the set of positive sample pairs, and N be the set of negative sample pairs. β is a hyperparameter used to balance the importance of reconstruction and classification tasks. β is a regularization parameter. The objective function is minimized by gradient descent to update the shared representation Z and related model parameters.
4. The water quality monitoring method based on multiple views above and below water as described in claim 1, characterized in that, In step S4, based on the shared representation of water quality features, the formula for constructing the graph structure by the multi-view collaborative water quality adversarial generator is as follows: ； in, It is a concatenated vector of shared features from samples i and j. It is a multilayer perceptron. The sigmoid function maps the output to the (0,1) interval, representing the probability or strength of the correlation between samples.
5. A water quality monitoring method based on multiple views above and below water as described in claim 1, characterized in that, The optimization objective function of the dynamic water quality discrimination device described in step S5 is: ； in, For the discriminator scoring, A is a graph structure generated based on the shared representation of water quality features. Let be the adjacency matrix of the v-th view constructed using the k-nearest neighbor algorithm.
6. A water quality monitoring method based on multiple views above and below water as described in claim 1, characterized in that, In step S6, the specific process for making a selection judgment based on the discrimination score is as follows: Calculate the average score of the discriminator for the generated graph structure A. ;like , If the preset threshold is met, the image quality is considered acceptable, and the process proceeds to step S7; otherwise, the judgment result is... As a loss signal, backpropagation updates the parameters of the multi-view collaborative water quality adversarial generator, and then returns to S4 to regenerate the graph structure A.
7. A water quality monitoring method based on multiple views above and below water as described in claim 1, characterized in that, In step S7, the formula for inference using the optimized graph structure in the denoised graph convolutional network is as follows: ； in, These are the node features of the l-th layer. , It is an adjacency matrix with self-connections. It is its degree matrix. This is a noise reduction function specifically designed for water quality data. It uses thresholding and entropy constraints to sparsify and reduce noise in the edge weights. It is a trainable weight matrix. As the activation function, the output features are calculated through forward propagation. And update network weights through backpropagation using label data. .
8. A water quality monitoring method based on multiple views above and below water as described in claim 1, characterized in that, The output water pollution level classification and key water quality parameter inversion in step S8 include: Water pollution level classification: Based on the node features output by a graph convolutional network, a classifier calculates the probability distribution of pollution levels, using the following formula: ； in This is a probability matrix for pollution levels. For classifier weights, The node features generated by the denoising graph convolutional network result in the final classification. ; Key water quality parameters inversion: Key water quality parameters are inverted and output by constructing a multi-task regression output layer. The key water quality parameters include chemical oxygen demand, ammonia nitrogen, total phosphorus and dissolved oxygen.
9. A water quality monitoring equipment for implementing the water quality monitoring method based on multiple views above and below water as described in any one of claims 1-8, characterized in that... The invention includes a multi-functional unmanned surface vessel (USV), which includes a camera sensor installed at a preset position on the hull for all-round monitoring of the aquatic environment. The sonar sensor, installed at the center of the hull, is used to ensure the stability of sound wave transmission and reception and to effectively detect underwater conditions. Chemical sensors installed on the side of the hull near the water surface are used to monitor chemical substances in the water in real time. The multi-functional unmanned surface vessel also integrates a preprocessing unit, a noise reduction unit, a prediction unit, and an evaluation unit; among which, the preprocessing unit is used to input surface optical images and underwater acoustic and chemical data, and combine them with historical water quality data and corresponding water quality labels to construct a partially labeled multi-view water quality matrix. The noise reduction unit is used to reduce noise in the multi-view water quality matrix using nonlocal mean noise reduction technology; The prediction unit is used for node feature inference and updating through a denoised graph convolutional network; The assessment unit is used to output water pollution level classification and key water quality parameters.
10. A water quality monitoring and assessment device, the device comprising a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the water quality monitoring method based on multiple views above and below water as described in any one of claims 1-8, according to the instructions in the program code.