CN-122023892-A - Water quality monitoring method and equipment based on water-on-water-under-water multi-view

CN122023892ACN 122023892 ACN122023892 ACN 122023892ACN-122023892-A

Abstract

A water quality monitoring method based on water and underwater multi-view includes inputting water optical image and underwater acoustic and chemical data, combining historical water quality data with corresponding water quality labels to construct a multi-view water quality matrix with partial marks, denoising the multi-view water quality matrix, extracting noise robust features to obtain water quality feature sharing representation, constructing a graph structure through a multi-view collaborative water quality countermeasure generator based on the water quality feature sharing representation, performing quality assessment, judging based on the water quality relation graph structure, entering the next step if the water quality relation graph structure accords with an ideal graph structure, otherwise feeding back to an optimized multi-view collaborative water quality countermeasure generator to perform node feature reasoning and updating, and outputting water quality pollution level classification and key water quality parameter inversion. The invention can effectively combine the complementary information of the water optical image and the underwater acoustic and chemical data, and can effectively reduce noise on the water quality information, thereby realizing robust and high-precision water quality monitoring.

Inventors

LIANG NAIYAO
Liang Ledao
TANG YONGRONG
HU YUANYUAN
XIANG DAN
XIAO MINGMING
LI JIAHENG
LIU FANG

Assignees

广州航海学院

Dates

Publication Date: 20260512
Application Date: 20260122

Claims (10)

1. The water quality monitoring method based on the water-on-water-under-water multi-view is characterized by comprising the following steps of: S1, inputting an on-water optical image and underwater acoustic and chemical data, and constructing a partially marked multi-view water quality matrix by combining historical water quality data and corresponding water quality labels; s2, denoising the multi-view water quality matrix by a non-local mean denoising technology; s3, extracting noise robust features through a sharing contrast learning technology to obtain water quality feature sharing representation; s4, constructing a graph structure through a multi-view collaborative water quality countermeasure generator based on the water quality characteristic sharing representation; s5, carrying out quality evaluation on the generated graph structure through a dynamic perception water quality countermeasure discriminator; Step S6, judging based on the water quality relation diagram structure, if the water quality relation diagram structure accords with the ideal diagram structure, entering step S7, otherwise, feeding back to the step S4 to optimize the multi-view collaborative water quality countermeasure generator; step S7, based on the optimized graph structure, performing node characteristic reasoning and updating through a noise reduction graph convolution network; and S8, outputting water pollution level classification and key water quality parameter inversion.
2. The water quality monitoring method based on the water-on-water-under-water multi-view according to claim 1, wherein in step S2, the formula for denoising the multi-view water quality matrix by the non-local mean denoising technique is as follows: ; Wherein, the A multi-view water quality matrix for the v-th view, Is the denoising water quality matrix of the v-th view, A search window representing a pixel i is shown, Representing the weight between pixel i and pixel j, Representing the square of the euclidean distance between pixel i and pixel j, H is a smoothing parameter and is a normalization factor.
3. The water quality monitoring method based on multiple views of water and water according to claim 1, wherein the optimization objective function of the sharing contrast learning in step S3 is: ; Wherein Z is a water quality characteristic sharing representation, , Is the denoising water quality matrix of the v-th view, For the projection matrix of the v-th view, For the Frobenius norm, Ω is the labeled sample set, the function f (·) represents the difference between the label and the actual value of the sample, Representing a graph convolution network, representing a multi-view collaborative water quality countermeasure generator, Y representing a label matrix set corresponding to the denoising water quality matrix, For the sum of the exponential similarity of all positive sample pairs (i, j), For the sum of the exponential similarity of all positive and negative pairs of samples, Is a vector And Is used for the internal product of (a), The comparison learning vector is shared for the water quality characteristics of samples i and j respectively, For the temperature parameter, P is the positive sample pair set, N is the negative sample pair set, For superparameters for the importance of the reconstruction and classification tasks, β is a canonical parameter, the objective function is minimized by gradient descent, and the shared representation Z and related model parameters are updated.
4. The method for monitoring water quality based on multiple views above and below water according to claim 1, wherein in step S4, a formula for constructing a graph structure by the multiple view collaborative water quality countermeasure generator based on the water quality feature sharing representation is: ; Wherein, the Is the shared feature splice vector for samples i and j, Is a multi-layer perceptron, the Sigmoid function maps the output to a (0, 1) interval, representing the probability or intensity that there is an association between samples.
5. The method for monitoring water quality based on multiple views of water according to claim 1, wherein the optimizing objective function of the dynamic sensing water quality countermeasure discriminator in step S5 is: ; Wherein, the For the arbiter score, a is a graph structure generated based on the water quality feature sharing representation, Is the adjacency matrix of the v-th view constructed by k-nearest neighbor algorithm.
6. The water quality monitoring method based on the above-water and underwater multi-view according to claim 1, wherein in step S6, the specific flow of the selection judgment based on the judgment score is as follows: calculating the average score of the discriminant pair to generate graph structure a If (1) , If the quality of the graph is qualified, the flow proceeds to step S7, otherwise, the judging result is that As the loss signal counter-propagates, the parameters of the multi-view collaborative water quality countermeasure generator are updated, and then the graph structure a is regenerated back to S4.
7. The water quality monitoring method based on the water-borne and underwater multiview according to claim 1, wherein in step S7, a formula for performing noise reduction graph convolution network reasoning by using the optimized graph structure is as follows: ; Wherein, the Is a node characteristic of the first layer, , Is to add a adjacency matrix for the self-connection, Is its degree matrix. Is a noise reduction function specially designed for water quality data, the function performs sparsification and noise reduction on the boundary weights through thresholding and entropy constraint, Is a matrix of trainable weights that, To activate the function, output features are computed by forward propagation And updating network weights by back propagation using tag data 。
8. The method for monitoring water quality based on multiple views of water according to claim 1, wherein the step S8 of classifying the output water quality pollution level and inverting the key water quality parameters comprises: the water pollution level classification is that the probability distribution of the pollution level is calculated by a classifier based on node characteristics output by a graph convolution network, and the formula is as follows: ; Wherein the method comprises the steps of As a pollution level probability matrix, For the weights of the classifier(s), For node characteristics generated through the noise reduction graph convolution network, the final classification result is that ; And inverting the key water quality parameters, namely inverting and outputting the key water quality parameters by constructing a multi-task regression output layer, wherein the key water quality parameters comprise chemical oxygen demand, ammonia nitrogen, total phosphorus and dissolved oxygen.
9. A water quality monitoring device for implementing the water quality monitoring method based on multiple views of water on water and under water according to any one of claims 1-8, characterized by comprising a multifunctional unmanned ship, said multifunctional unmanned ship comprising a camera sensor mounted at a preset position above a ship body for omnibearing monitoring of the water environment; the sonar sensor is arranged at the center position below the ship body and is used for ensuring the stability of sound wave emission and receiving and effectively detecting the underwater condition; The chemical sensor is arranged on the side surface of the ship body and close to the water surface and is used for contacting with water in real time and accurately monitoring chemical substances in the water; The multifunctional unmanned ship is also integrated with a preprocessing unit, a noise reduction unit, a prediction unit and an evaluation unit, wherein the preprocessing unit is used for inputting an optical image on water, underwater acoustic and chemical data, and combining historical water quality data and corresponding water quality labels to construct a partially marked multi-view water quality matrix; the noise reduction unit is used for reducing noise of the multi-view water quality matrix through a non-local mean value noise reduction technology; The prediction unit is used for carrying out node characteristic reasoning and updating through a noise reduction graph convolution network; the evaluation unit is used for outputting the water quality pollution level classification and the key water quality parameter inversion.
10. A water quality monitoring and assessment device, the device comprising a processor and a memory; The memory is used for storing program codes and transmitting the program codes to the processor; The processor is configured to execute the above-water-under-water multi-view based water quality monitoring method according to any one of claims 1-8 according to instructions in the program code.

Description

Water quality monitoring method and equipment based on water-on-water-under-water multi-view Technical Field The invention belongs to the technical field of water quality monitoring, and particularly relates to a water quality monitoring method and equipment based on water-on-water-under-water multi-view, in particular to a method capable of realizing water-on-water-under-water environment monitoring by being provided with various sensors and multifunctional unmanned ship equipment. Background The water quality monitoring refers to the process of systematically monitoring and analyzing physical, chemical and biological indexes in a water body by a scientific method so as to evaluate the water quality condition, pollution degree and change trend. The monitoring range is very wide, and the monitoring range comprises uncontaminated and polluted natural water rivers, lakes, seas, underground water, various industrial drainage and the like. The main monitoring items can be divided into two major categories, one is comprehensive index reflecting water quality condition, such as temperature, chromaticity, turbidity, value, conductivity, suspended matter, dissolved oxygen, chemical oxygen demand, biochemical oxygen demand, etc., and the other is toxic matter, such as phenol, cyanogen, arsenic, lead, chromium, cadmium, mercury, organic pesticide, etc. The real-time monitoring of river water quality through the sensor is an effective means for protecting water resources, and is increasingly applied, however, the field water area environment is complex and changeable, sensor data is easily interfered by water flow, biological activity, equipment noise and the like, and the signal to noise ratio is low. Most of existing water quality assessment models lack a targeted noise reduction design, have poor robustness to noise, cause inaccurate extraction of key features of the models, and have large fluctuation and low accuracy of assessment results. In addition, the existing multi-view learning technology is widely applied to the fields of image classification, target recognition and the like, but is directly applied to water quality monitoring, and particularly when multi-view data with strong isomerism and prominent noise such as water and water are processed, the problems that the shared feature representation is sensitive to noise, a graph structure reflecting the inherent association of water quality cannot be dynamically constructed and the like often exist. Disclosure of Invention In view of the above-mentioned drawbacks of the prior art, the present invention provides a water quality monitoring method and apparatus based on water-on-water-under-water multi-view, which overcomes the drawbacks that the shared feature representation is sensitive to noise and the graph structure reflecting the internal association of water quality cannot be dynamically constructed, and specifically includes the following technical scheme: The first aspect provides a water quality monitoring method based on water on-water and underwater multi-view, which comprises the following specific steps: S1, inputting an on-water optical image and underwater acoustic and chemical data, and constructing a partially marked multi-view water quality matrix by combining historical water quality data and corresponding water quality labels; s2, denoising the multi-view water quality matrix by a non-local mean denoising technology; s3, extracting noise robust features through a sharing contrast learning technology to obtain water quality feature sharing representation; s4, constructing a graph structure through a multi-view collaborative water quality countermeasure generator based on the water quality characteristic sharing representation; s5, carrying out quality evaluation on the generated graph structure through a dynamic perception water quality countermeasure discriminator; Step S6, judging based on the water quality relation diagram structure, if the water quality relation diagram structure accords with the ideal diagram structure, entering step S7, otherwise, feeding back to the step S4 to optimize the multi-view collaborative water quality countermeasure generator; step S7, based on the optimized graph structure, performing node characteristic reasoning and updating through a noise reduction graph convolution network; and S8, outputting water pollution level classification and key water quality parameter inversion. The second aspect provides water quality monitoring equipment based on water-borne and underwater multi-view, namely a multifunctional unmanned ship: the multifunctional unmanned ship comprises a camera sensor which is arranged at a preset position above a ship body, has wide-angle shooting and high-definition imaging functions, can monitor the water environment in all directions, a sonar sensor which is arranged at a central position below the ship body, ensures the stability of sound wave emission and receiving, effectively detects th