River pollution traceability analysis system and method based on ultraviolet-visible light image

Abstract

The invention discloses a river pollution traceability analysis system and method based on an ultraviolet-visible light image, and the system comprises a data perception layer, a data preprocessing and transmission layer, a model calculation and traceability engine layer, and an application service layer. According to the method, the existing manual sampling and chemical analysis are replaced or reduced by an ultraviolet-visible light video online monitoring technology, automatic acquisition and nearly real-time analysis of data are realized, the monitoring cost and time delay are greatly reduced, the drainage basin is innovatively abstracted into a digraph structure, and the drainage basin monitoring system has a wide application prospect. Multi-source data such as optical images, hydrology, meteorology and geographic information are comprehensively utilized, a more comprehensive analysis model is constructed, the traceability conclusion is more scientific and reliable, a set of complete system solution from data acquisition, processing and modeling to business application is provided, and the method has good engineering application value and popularization prospects.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of water environment monitoring and information technology, and particularly relates to a river pollution source tracing analysis system and method based on ultraviolet-visible light images. BACKGROUND

[0002] Water pollution control is the core task of ecological environment protection, and quickly and accurately tracing the pollution source is the key to implementing precise control. At present, water pollution source tracing mainly relies on manual field sampling and laboratory chemical analysis, which is time-consuming, labor-intensive, costly and severely lagging, and cannot meet the needs of real-time monitoring and rapid response. Although some online monitoring devices (such as water quality sensors) can provide continuous data, their monitoring range is limited, and they can only reflect the local water quality conditions at the monitoring point, and it is difficult to describe the overall diffusion and migration process and spatio-temporal dynamic distribution of pollutants in the watershed.

[0003] In the prior art, some studies attempt to use mathematical models or existing machine learning algorithms for pollution source tracing, but these methods have obvious deficiencies in multi-source heterogeneous data fusion, and it is usually difficult to effectively couple meteorological, hydrological, topographic, land use and other multi-dimensional information, and the spatio-temporal dynamic process of pollutant diffusion is insufficient, resulting in limited accuracy and reliability of source tracing, which is difficult to support precise source tracing in complex watershed environments.

[0004] Therefore, there is an urgent need for a technical solution that can trace the source of river pollutants in real time, accurately and at low cost, to provide efficient decision support for ecological environmental management departments. SUMMARY

[0005] To solve the above technical problems, the application provides a river pollution source tracing analysis system and method based on ultraviolet-visible light images, which combines the advantages of directed graph convolution network and recurrent neural network to realize fast, accurate and low-cost identification and positioning of watershed pollution sources.

[0006] The technical solution adopted by the application is as follows: a river pollution source tracing analysis system based on ultraviolet-visible light images, comprising: a data perception layer, a data preprocessing and transmission layer, a model calculation and source tracing engine layer, and an application service layer.

[0007] The data perception layer includes ultraviolet-visible light video online monitoring devices and meteorological and hydrological sensors arranged at key monitoring sections of the watershed, which are used to collect water image data and environmental parameter data in real time.

[0008] The data preprocessing and transmission layer is responsible for denoising, correcting and feature extraction of the collected raw image data, and encrypting the data and transmitting it to the cloud platform or central server through the Internet of Things technology.

[0009] Model calculation and tracing engine layer: a trained spatio-temporal convolution tracing model, i.e., a river pollutant tracing convolution network model, is built in; the layer receives real-time data, calls the trained model for inference calculation, completes the tracing analysis task, and outputs the spatial position and probability evaluation of the pollution source;

[0010] Application service layer: a visual interface is provided for users to display the basin water quality map, pollution diffusion simulation animation, tracing result report and early warning information in the form of a Web end or mobile end application, and to provide decision support for management departments.

[0011] Further, in the river pollution tracing analysis system, the ultraviolet-visible light video monitoring device in the data perception layer simultaneously covers the ultraviolet band and the visible light full band.

[0012] Further, the application further provides a river pollution tracing analysis method based on ultraviolet-visible light images, which is applied to the river pollution tracing analysis system, and the specific steps are as follows:

[0013] S1, based on the data perception layer, water body image data and environmental parameters are collected and input into the data preprocessing and transmission layer to construct a data set;

[0014] First, a monitoring section is selected in the target river, and an ultraviolet-visible light video online monitoring device is deployed, ultraviolet-visible light image data of the water surface are continuously collected, and water sample samples are synchronously collected, and key pollutant concentration values are obtained through laboratory testing.

[0015] The key pollutant concentration values include the concentrations of total phosphorus TP and total nitrogen TN.

[0016] Then, the image data is preprocessed, ultraviolet-visible light image features are extracted through the BLS, high-dimensional joint feature representation is constructed through the feature nodes and the enhancement nodes, multi-source feature adaptive fusion and key feature screening are realized in combination with the analysis results of the main pollutant concentrations of the river, the screened feature data are associated with the corresponding monitoring section labels, the data are standardized to construct a comprehensive feature data set for model training.

[0017] The preprocessing includes radiation correction, noise suppression and image enhancement; the extracted ultraviolet-visible light image features include spectral features, texture features and color features of the ultraviolet-visible light images.

[0018] S2, a directed graph structure of the target river is constructed;

[0019] Collect multi-source background data of the target river, construct a directed graph structure that can accurately characterize the hydrological connectivity of the target river, calculate the spatial distance and topographic relationship between adjacent stations based on the geographical location of monitoring stations and digital elevation model data, and abstract the target watershed into a directed graph structure G = (V, E, A) according to the direction of water flow and the confluence relationship.

[0020] Where V represents the set of nodes, with each node corresponding to a monitoring section; E represents the set of directed edges, with each edge representing a monitoring section. An edge represents a water flow path from node i to node j, with the direction determined by the actual water flow direction. A represents the adjacency matrix of the graph, used to describe the spatial connection relationships and weights between nodes. The weights are determined based on the geographical location information of the two nodes and one or more factors from the collected multi-source background data. The multi-source background data includes: digital elevation model (DEM) data, meteorological data, hydrological data, potential pollution source distribution data, and land use data.

[0021] S3. Input the standardized dataset obtained in step S1 and the directed graph structure obtained in step S2 into the model computation and source tracing engine layer, and train and validate the model through the built-in river pollutant source tracing convolutional network model.

[0022] The river pollutant source tracing convolutional network model is a spatiotemporal source tracing model that integrates a directed graph convolutional network (GCN) and a gated recurrent unit (GRU), including: a spatial feature extraction module and a temporal feature extraction module.

[0023] The spatial feature extraction module utilizes the directed graph structure of the watershed constructed in step S2 of the directed graph convolutional network (GCN). The GCN aggregates information from upstream nodes along the water flow direction to capture the spatial diffusion characteristics of pollutants in the river network. Its graph convolution operation follows the water flow direction defined by the directed graph structure. The temporal feature extraction module inputs the time series of feature data obtained from video monitoring images at each node into the gated recurrent unit (GRU) to capture the time dependence and dynamic change law of pollutant diffusion, and outputs a feature representation for predicting the probability distribution of pollutant sources.

[0024] Then, the spatial features extracted by GCN are fused with the temporal features extracted by GRU. The dataset constructed in step S1 is then divided into training and validation sets according to the actual situation. The model is trained and validated, the model parameters are optimized through backpropagation, and the model is evaluated by the classification accuracy. The training and validation are completed, and the trained model is finally obtained.

[0025] S4. Based on step S3, deploy the trained river pollutant source tracing convolutional network model in the online application system, input real-time data to complete the river pollutant source tracing analysis task, output the spatial location and probability assessment of pollution sources, and display them through the application service layer.

[0026] The trained spatiotemporal convolutional source tracing model is deployed in an online application system. The system receives ultraviolet-visible image data of each monitoring section of the target in real time and inputs it into the model. When an abnormal increase in concentration is detected at a certain node, the system analyzes the upstream path pointing to that node in the directed graph and calculates the probability that different upstream paths are pollution sources by combining the spatiotemporal diffusion patterns output by the model. Finally, the upstream area with the highest probability is determined as the most likely source of pollution, thus achieving real-time and accurate location of the pollution source.

[0027] The beneficial effects of this invention are as follows: The system of this invention includes: a data perception layer, a data preprocessing and transmission layer, a model calculation and source tracing engine layer, and an application service layer. This invention replaces or reduces existing manual sampling and chemical analysis through ultraviolet-visible light video online monitoring technology, achieving automatic data acquisition and near real-time analysis, significantly reducing monitoring costs and time delays. It innovatively abstracts the watershed into a directed graph structure, employing a directed graph convolutional network (GCN) that fully considers water flow direction and river network topology, enabling a more accurate description of the spatial diffusion process of pollutants. Combined with GRU time series prediction, it achieves a high degree of simulation of the spatiotemporal dynamic evolution of pollutants, significantly improving the accuracy of source tracing results. It comprehensively utilizes multi-source data such as optical images, hydrology, meteorology, and geographic information to construct a more comprehensive analysis model, making source tracing conclusions more scientific and reliable. Therefore, it provides a complete system solution from data acquisition, processing, modeling to business applications, possessing good engineering application value and promising prospects for promotion. Attached Figure Description

[0028] Figure 1 This is a structural diagram of a river pollution source tracing and analysis system based on ultraviolet-visible light images according to the present invention.

[0029] Figure 2 This is a flowchart of a river pollution source tracing analysis method based on ultraviolet-visible light images according to the present invention.

[0030] Figure 3 This is a schematic diagram of the spatiotemporal graph convolutional tracing model (GCN-GRU) in an embodiment of the present invention. Detailed Implementation

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0032] like Figure 1As shown, the present invention provides a river pollution source tracing analysis system based on ultraviolet-visible light images, comprising: a data perception layer, a data preprocessing and transmission layer, a model calculation and source tracing engine layer, and an application service layer.

[0033] Data sensing layer: includes ultraviolet-visible light video online monitoring equipment (ultraviolet-visible light cameras) and meteorological and hydrological sensors deployed at key monitoring sections of the watershed, used to collect water body image data and environmental parameter data in real time;

[0034] Data preprocessing and transmission layer: responsible for denoising, correcting and extracting features from the acquired raw image data, and transmitting the data to the cloud platform or central server in encryption through Internet of Things technology;

[0035] Model computation and source tracing engine layer: This is the core of the system. It has a built-in spatiotemporal convolutional source tracing model that has been trained, namely the river pollutant source tracing convolutional network model. This layer receives real-time data, calls the trained model to perform inference calculations, completes the source tracing analysis task, and outputs the spatial location and probability assessment of the pollution source.

[0036] Application Service Layer: Provides users with a visual interface to display watershed water quality maps, pollution diffusion simulation animations, source tracing results reports, and early warning information in the form of web or mobile applications, providing decision support for management departments.

[0037] In this embodiment, in the river pollution source tracing and analysis system, the ultraviolet-visible light video monitoring equipment in the data sensing layer simultaneously covers the ultraviolet band and the entire visible light band.

[0038] like Figure 2 As shown, in this embodiment, a method for tracing the source of river pollution based on ultraviolet-visible light images is also provided, which is applied to the river pollution source tracing analysis system. The specific steps are as follows:

[0039] S1. Collect water body image data and environmental parameters based on the data sensing layer and input them into the data preprocessing and transmission layer to construct a dataset;

[0040] First, a monitoring section is selected in the target river, and then ultraviolet-visible light video online monitoring equipment is deployed to continuously collect ultraviolet-visible light image data of the water surface and collect water samples simultaneously. The concentration values ​​of key pollutants are obtained through laboratory testing.

[0041] The concentration values ​​of the key pollutants include the concentrations of total phosphorus (TP) and total nitrogen (TN).

[0042] Then, after preprocessing the image data, the ultraviolet-visible image features are extracted by the width learning system (BLS). A high-dimensional joint feature representation is then constructed by feature nodes and enhancement nodes. Combined with the analysis results of the main pollutant concentrations in the river (key pollutant concentration values), the adaptive fusion of multi-source features and key feature selection are achieved. The selected feature data are associated with the corresponding monitoring section labels. After data standardization, a comprehensive feature dataset that can be used for model training is constructed.

[0043] The preprocessing includes radiometric correction, noise suppression, and image enhancement; the extracted ultraviolet-visible image features include spectral features, texture features, and color features of the ultraviolet-visible image.

[0044] In this embodiment, six monitoring sections were set up on the main stream and major tributaries of the river. Each section was equipped with a dual-spectrum ultraviolet-visible camera, and the sampling frequency of the equipment was set to once every 15 minutes. Water samples were collected weekly for TP and TN laboratory analysis, and this process was continued for one quarter to ensure that the collected image data covered different time periods and lighting conditions of the river. Based on this, key image features were extracted using a width learning system.

[0045] S2. Construct the directed graph structure of the target river;

[0046] Collect multi-source background data of the target river, construct a directed graph structure that can accurately characterize the hydrological connectivity of the target river, calculate the spatial distance and topographic relationship between adjacent stations based on the geographical location of monitoring stations and digital elevation model data, and abstract the target watershed into a directed graph structure G = (V, E, A) according to the direction of water flow and the confluence relationship.

[0047] Where V represents the set of nodes, with each node corresponding to a monitoring section; E represents the set of directed edges, with each edge representing a monitoring section. An edge represents a water flow path from node i to node j, with the direction determined by the actual water flow direction. The direction of flow is used to determine the orientation between connected stations. Nodes upstream of the river are considered source nodes, and downstream nodes are sink nodes. Directed edges from source nodes to sink nodes are used to simplify the pollutant transport process. The weights of the directed edges are calculated using selected influencing factor data and the geographical location information of the two nodes, quantifying the influence of the source node on the sink node. A represents the adjacency matrix of the graph, used to describe the spatial connection relationships and weights between nodes. The weights are determined based on the geographical location information of the two nodes and one or more factors from the collected multi-source background data. The multi-source background data includes: digital elevation model (DEM) data, meteorological data (rainfall, wind speed, wind direction), hydrological data (flow velocity, flow rate), potential pollution source distribution data (location and type of industrial areas, agricultural areas, and residential areas), and land use data.

[0048] This embodiment establishes a directed graph structure based on the above rules, which contains 6 nodes and 10 directed edges.

[0049] S3. Input the standardized dataset obtained in step S1 and the directed graph structure obtained in step S2 into the model computation and source tracing engine layer, and train and validate the model through the built-in river pollutant source tracing convolutional network model.

[0050] like Figure 3 As shown, the river pollutant source tracing convolutional network model is a spatiotemporal source tracing model that integrates a directed graph convolutional network (GCN) and a gated recurrent unit (GRU), including a spatial feature extraction module and a temporal feature extraction module. In this embodiment, the model includes a two-layer directed graph convolutional network and a gated recurrent unit with 32 hidden units. The input is the monitoring data of the first 10 time steps, and the final output is the probability distribution of different upstream areas as pollution sources.

[0051] in, Figure 3 ZhongX t Let t represent the input at time t, Y represent the final output, i.e., the probability distribution of different upstream regions as pollution sources, and n represent the time step of the input sequence. The spatial feature extraction module uses a directed graph convolutional network (GCN) to process the watershed directed graph structure constructed in step S2. This network can aggregate information from upstream nodes along the water flow direction, effectively capturing the spatial diffusion characteristics of pollutants in the river network. Its graph convolution operation follows the water flow direction defined by the directed graph structure. The temporal feature extraction module inputs the time series of feature data obtained from video monitoring images at each node into a gated recurrent unit (GRU) to capture the time dependence and dynamic change law of pollutant diffusion, and outputs a feature representation for predicting the probability distribution of pollutant sources to support the determination of the most probable source of pollution.

[0052] Then, the spatial features extracted by GCN are fused with the temporal features extracted by GRU. The dataset constructed in step S1 is then divided into training and validation sets according to the actual situation. The model is trained and validated, the model parameters are optimized through backpropagation, and the model is evaluated by the classification accuracy. The training and validation are completed, and the trained model is finally obtained.

[0053] To evaluate the model's feature learning and source tracing analysis capabilities, the t-SNE nonlinear dimensionality reduction algorithm was used to map high-dimensional features to a two-dimensional plane. The model's learning ability was characterized by visualizing the clustering distribution characteristics of different pollution source samples in the low-dimensional space. Simultaneously, the confusion matrix, along with multiple evaluation metrics such as precision and recall, were combined to comprehensively and quantitatively assess the effectiveness of river pollution source tracing.

[0054] S4. Based on step S3, deploy the trained river pollutant source tracing convolutional network model in the online application system, input real-time data to complete the river pollutant source tracing analysis task, output the spatial location and probability assessment of pollution sources, and display them through the application service layer.

[0055] The trained spatiotemporal convolutional source tracing model is deployed in an online application system. The system receives ultraviolet-visible image data from each monitoring section of the target river in real time and inputs it into the model. When an abnormal increase in concentration is detected at a certain node, the system analyzes the upstream path pointing to that node in the directed graph and calculates the probability that different upstream paths are pollution sources by combining the spatiotemporal diffusion patterns output by the model. Finally, the upstream area with the highest probability is identified as the most likely source of pollution, thus achieving real-time and accurate location of the pollution source.

[0056] In summary, the system can accurately identify pollution events and achieve rapid and precise tracing of pollution sources. This invention replaces or reduces existing manual sampling and chemical analysis through ultraviolet-visible light video online monitoring technology, achieving automatic data acquisition and near real-time analysis, significantly reducing monitoring costs and time delays. It innovatively abstracts the watershed into a directed graph structure using a directed graph convolutional network (GCN), fully considering water flow direction and river network topology, enabling a more accurate description of the spatial diffusion process of pollutants. Combined with GRU time series prediction, it achieves a high degree of simulation of the spatiotemporal dynamic evolution of pollutants, significantly improving the accuracy of source tracing results. By comprehensively utilizing multi-source data such as optical images, hydrology, meteorology, and geographic information, a more comprehensive analysis model is constructed, making source tracing conclusions more scientific and reliable. Therefore, it provides a complete system solution from data acquisition, processing, modeling to business applications, possessing good engineering application value and promising prospects for widespread adoption.

[0057] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this invention, and these modifications and substitutions should all be covered within the scope of protection of this invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A river pollution source tracing and analysis system based on ultraviolet-visible light images, comprising: Data perception layer, data preprocessing and transmission layer, model calculation and tracing engine layer, application service layer; Data sensing layer: includes ultraviolet-visible light video online monitoring equipment and meteorological and hydrological sensors deployed at key monitoring sections of the watershed, used to collect water body image data and environmental parameter data in real time; Data preprocessing and transmission layer: responsible for denoising, correcting and extracting features from the acquired raw image data, and transmitting the data to the cloud platform or central server in encryption through Internet of Things technology; Model computation and source tracing engine layer: Built-in spatiotemporal convolutional source tracing model, namely river pollutant source tracing convolutional network model; This layer receives real-time data, calls the trained model to perform inference computation, completes the source tracing analysis task, and outputs the spatial location and probability assessment of the pollution source; Application Service Layer: Provides users with a visual interface to display watershed water quality maps, pollution diffusion simulation animations, source tracing results reports, and early warning information in the form of web or mobile applications, providing decision support for management departments.

2. The river pollution source tracing and analysis system based on ultraviolet-visible light images according to claim 1, characterized in that, The ultraviolet-visible video monitoring equipment in the data sensing layer covers both the ultraviolet band and the entire visible light band.

3. A method for tracing the source of river pollution based on ultraviolet-visible light images, applied to the river pollution source tracing analysis system described in claim 1, comprising the following specific steps: S1. Collect water body image data and environmental parameters based on the data sensing layer and input them into the data preprocessing and transmission layer to construct a dataset; First, a monitoring section is selected in the target river, and then an ultraviolet-visible light video online monitoring device is deployed to continuously collect ultraviolet-visible light image data of the water surface and collect water samples at the same time. The concentration values ​​of key pollutants are obtained through laboratory testing. in, The concentration values ​​of the key pollutants include: the concentrations of total phosphorus (TP) and total nitrogen (TN); Then, after preprocessing the image data, the ultraviolet-visible image features are extracted by the width learning system BLS. A high-dimensional joint feature representation is then constructed by feature nodes and enhancement nodes. Combined with the analysis results of the concentration of major pollutants in the river, the adaptive fusion of multi-source features and the selection of key features are realized. The selected feature data are associated with the corresponding monitoring section labels. After data standardization, a comprehensive feature dataset for model training is constructed. The preprocessing includes radiometric correction, noise suppression, and image enhancement; the extracted ultraviolet-visible image features include spectral features, texture features, and color features of the ultraviolet-visible image. S2. Construct the directed graph structure of the target river; Collect multi-source background data of the target river, construct a directed graph structure that can accurately characterize the hydrological connectivity of the target river, calculate the spatial distance and topographic relationship between adjacent stations based on the geographical location and digital elevation model data of the monitoring stations, and abstract the target watershed into a directed graph structure G=(V,E,A) according to the direction of water flow and the confluence relationship. Where V represents the set of nodes, with each node corresponding to a monitoring section; E represents the set of directed edges, with each edge representing a monitoring section. An edge represents a water flow path from node i to node j, and its direction is determined by the actual water flow direction; A represents the adjacency matrix of the graph, which describes the spatial connection relationship and weight between nodes. The weight is determined based on the geographical location information of the two nodes and one or more factors from the collected multi-source background data; the multi-source background data includes: digital elevation model (DEM) data, meteorological data, hydrological data, potential pollution source distribution data, and land use data; S3. Input the standardized dataset obtained in step S1 and the directed graph structure obtained in step S2 into the model computation and source tracing engine layer, and train and validate the model through the built-in river pollutant source tracing convolutional network model. The river pollutant source tracing convolutional network model is a spatiotemporal source tracing model that integrates a directed graph convolutional network (GCN) and a gated recurrent unit (GRU), including: a spatial feature extraction module and a temporal feature extraction module; The spatial feature extraction module utilizes the directed graph structure of the watershed constructed in step S2 of the directed graph convolutional network (GCN). The GCN aggregates information from upstream nodes along the water flow direction to capture the spatial diffusion characteristics of pollutants in the river network. Its graph convolution operation follows the water flow direction defined by the directed graph structure. The temporal feature extraction module inputs the time series of feature data obtained from video monitoring images at each node into the gated recurrent unit (GRU) to capture the time dependence and dynamic change law of pollutant diffusion, and outputs a feature representation for predicting the probability distribution of pollutant sources. Then, the spatial features extracted by GCN are fused with the temporal features extracted by GRU. The dataset constructed in step S1 is then divided into training and validation sets according to the actual situation. The model is trained and validated, the model parameters are optimized through backpropagation, and the model is evaluated by the classification accuracy. The training and validation are completed, and the trained model is finally obtained. S4. Based on step S3, deploy the trained river pollutant source tracing convolutional network model in the online application system, input real-time data to complete the river pollutant source tracing analysis task, output the spatial location and probability assessment of pollution sources, and display them through the application service layer. The trained spatiotemporal convolutional source tracing model is deployed in an online application system. The system receives ultraviolet-visible image data of each monitoring section of the target in real time and inputs it into the model. When an abnormal increase in concentration is detected at a certain node, the system analyzes the upstream path pointing to that node in the directed graph and calculates the probability that different upstream paths are pollution sources by combining the spatiotemporal diffusion patterns output by the model. Finally, the upstream area with the highest probability is determined as the most likely source of pollution, thus achieving real-time and accurate location of the pollution source.

Images