CN-122024005-A - River sewage inlet detection and tracing method and system

CN122024005ACN 122024005 ACN122024005 ACN 122024005ACN-122024005-A

Abstract

The application discloses a method and a system for detecting and tracing a river sewage outlet, wherein the method comprises the steps of obtaining remote sensing visual characteristics, pollution source characteristics and environmental semantic characteristics of a river reach to be monitored; the method comprises the steps of constructing a drain detection and tracing integrated model, wherein the model comprises a drain detection network and a multi-mode tracing network, the drain detection network is used for outputting a drain spatial position prediction result, the multi-mode tracing network is used for outputting a pollution source attribute discrimination result, in the model training process, remote sensing visual features, pollution source features and environment semantic features are input into the model, the model is trained by combining a composite objective function, and after model training is completed, a current visible light orthographic image of a region to be predicted is input into the trained model, and the current drain spatial position prediction result and the current pollution source attribute discrimination result are output. The application can improve the detection accuracy of the hidden river sewage outlet and the reliability of pollution source discrimination, and can be widely applied to the technical field of environmental monitoring.

Inventors

ZHANG YINDAN
WANG MIAOMIAO
Hu Naiyue
OUYANG JIE
Lin Xinao

Assignees

西北师范大学

Dates

Publication Date: 20260512
Application Date: 20260106

Claims (10)

1. The river sewage inlet detection and tracing method is characterized by comprising the following steps of: acquiring an unmanned aerial vehicle remote sensing image of a river reach to be monitored, and performing first characterization processing on the unmanned aerial vehicle remote sensing image to obtain remote sensing visual characteristics; Acquiring cross section water quality monitoring data synchronized with the unmanned aerial vehicle remote sensing image in time and space, and performing second characterization processing on the cross section water quality monitoring data to obtain pollution source characteristics; acquiring the shore zone environmental text data of the river reach to be monitored, and performing third characterization processing on the shore zone environmental text data to obtain environmental semantic features; The method comprises the steps of constructing a drain outlet detection and tracing integrated model, wherein the drain outlet detection and tracing integrated model comprises a drain outlet detection network and a multi-mode tracing network, the drain outlet detection network is used for outputting a drain outlet spatial position prediction result, and the multi-mode tracing network is used for outputting a pollution source attribute discrimination result; Constructing a composite objective function, and carrying out joint optimization training on the drain detection and tracing integrated model according to the composite objective function, the remote sensing visual characteristics, the pollution source characteristics and the environment semantic characteristics until a preset training requirement is met, so as to obtain a trained drain detection and tracing integrated model; and inputting the current visible light orthographic image of the region to be predicted into the trained drain detection and tracing integrated model, and outputting a current drain prediction result, wherein the current drain prediction result comprises a current drain spatial position prediction result and a current pollution source attribute discrimination result.
2. The method of claim 1, wherein the obtaining the remote sensing image of the unmanned aerial vehicle of the river reach to be monitored and performing a first characterization process on the remote sensing image of the unmanned aerial vehicle to obtain the remote sensing visual feature comprises: The unmanned aerial vehicle remote sensing image of the river reach to be monitored is obtained, wherein the unmanned aerial vehicle remote sensing image comprises a visible light orthographic image, digital surface model data and a thermal infrared image; Performing geometric correction, spatial registration and scale unification on the unmanned aerial vehicle remote sensing image to obtain spatially aligned multi-source remote sensing data; And inputting the spatially aligned multi-source remote sensing data into a feature extraction network to perform joint feature extraction, and generating the remote sensing visual features for representing the spatial form and spatial distribution features of the river sewage outlet.
3. The method of claim 1, wherein the acquiring cross-sectional water quality monitoring data synchronized in time and space with the unmanned aerial vehicle remote sensing image and performing a second characterization process on the cross-sectional water quality monitoring data to obtain a pollution source signature comprises: acquiring the section water quality monitoring data synchronized with the unmanned aerial vehicle remote sensing image in time and space; carrying out data standardization treatment on the section water quality monitoring data to obtain a standardized water quality vector; and inputting the standardized water quality vector into a multi-layer perceptron network for linear transformation and nonlinear activation function processing to obtain the pollution source characteristics in the high-dimensional characteristic space.
4. The method of claim 1, wherein the obtaining the shore zone environmental text data of the river reach to be monitored and performing a third characterization process on the shore zone environmental text data to obtain environmental semantic features comprises: determining the geographic range of the river reach to be monitored based on the spatial metadata of the unmanned aerial vehicle remote sensing image; Based on the geographical range, acquiring the shore zone environment text data corresponding to the river reach to be monitored from a pre-constructed shore zone geographical environment knowledge database, wherein the shore zone geographical environment knowledge database comprises unstructured text data carrying a shore zone geographical tag, and the unstructured text data comprises a shore zone land utilization type description, a historical pollution source record and environment sensitive receptor information; Splicing unstructured text data in the shore zone environment text data to obtain a target text sequence, and adding corresponding classification marks at the head and tail of the target text sequence; mapping the target text sequence into an embedded vector matrix; Performing hierarchical feature transformation on the embedded vector matrix through a bidirectional language representation model to obtain a context hidden layer state matrix; And extracting a global text feature vector corresponding to the first classification mark in the context hidden layer state matrix, and carrying out feature space transformation on the global text feature vector to obtain the environment semantic feature.
5. The method of claim 1, wherein constructing the composite objective function, and performing joint optimization training on the integrated drain detection and tracing model according to the composite objective function, the remote sensing visual feature, the pollution source feature, and the environmental semantic feature until a preset training requirement is met, to obtain a trained integrated drain detection and tracing model, comprises: The composite objective function is constructed, wherein the composite objective function comprises a target detection regression loss, a pollution tracing classification loss and a logic consistency loss, and the logic consistency loss is used for restraining consistency between the pollution source attribute discrimination result and an environment physical rule based on the relationship among water flow direction, topography elevation and space position; Inputting the remote sensing visual features, the pollution source features and the environment semantic features into the drain detection and tracing integrated model, and calculating the target detection regression loss, the pollution tracing classification loss and the logic consistency loss of the drain detection and tracing integrated model in a forward propagation process based on the remote sensing visual features, the pollution source features and the environment semantic features; weighting the target detection regression loss, the pollution tracing classification loss and the logic consistency loss to obtain a total loss value of the composite objective function; and optimizing model parameters of the drain outlet detection and tracing integrated model through back propagation based on the total loss value of the composite objective function until a preset training requirement is met, so as to obtain the trained drain outlet detection and tracing integrated model.
6. The method of claim 5, further comprising the step of calculating the target-detection regression loss, the calculating the target-detection regression loss comprising: Extracting a salient feature map representing abnormal temperature distribution from a thermal infrared image in the remote sensing image of the unmanned aerial vehicle through a shallow convolution network; Based on the saliency feature map, performing spatial weighted fusion on the remote sensing visual features to obtain enhanced visual features; Calculating an elevation gradient field based on digital surface model data in the unmanned aerial vehicle remote sensing image, and generating a binary topology mask for marking an amphibious boundary area according to the elevation gradient field and a preset landform gradient threshold value; Based on a mask attention mechanism, restraining a spatial sampling range of attention calculation by using the binarization topological mask, and carrying out attention calculation on the enhanced visual characteristics to obtain attention enhancement characteristics; outputting the drain spatial position prediction result based on the attention enhancing feature; and calculating the target detection regression loss according to the deviation between the predicted result of the drain outlet space position and the true position mark by adopting an online difficult-case mining strategy.
7. The method of claim 5, further comprising the step of calculating the contamination traceability classification penalty, the calculating the contamination traceability classification penalty comprising: Performing feature transformation on the remote sensing visual features to construct query features; Performing feature stitching on the pollution source features and the environment semantic features, and performing feature transformation on the environment context features obtained after feature stitching to construct key features and value features; Calculating association weights according to the query features and the key features by adopting a scaling dot product attention mechanism, and carrying out weighted summation on the value features according to the association weights to obtain multi-mode fusion features for fusing water quality information and environment semantic information; Inputting the multi-mode fusion features into a classification layer, and outputting the pollution source attribute discrimination result; And calculating the pollution tracing classification loss according to the pollution source attribute discrimination result and the real pollution source label by adopting a cross entropy loss function.
8. The method of claim 5, further comprising the step of calculating the logical consistency loss, the calculating the logical consistency loss comprising: constructing a logic relation matrix between an environment scene and a pollution source; mapping the environmental semantic features into an environmental scene index; extracting a logic constraint mask vector from the logic relation matrix according to the environment scene index; And carrying out consistency check on the pollution source attribute discrimination result based on the logic constraint mask vector to obtain the logic consistency loss.
9. The method according to claim 1, wherein after outputting a current drain prediction result in the drain detection and tracing integrated model in which the current visible light orthographic image of the region to be predicted is input to the training completion, the method further comprises: Performing non-maximum suppression operation on the current drain outlet spatial position prediction result to obtain a target drain outlet spatial position prediction result; Performing confidence screening operation on the current pollution source attribute discrimination result to obtain a target pollution source attribute discrimination result; And carrying out visual processing on the target sewage outlet spatial position prediction result and the target pollution source attribute discrimination result to generate spatial expression data, and displaying the spatial expression data on a visual interface.
10. The river sewage inlet detection and tracing system is characterized by comprising the following modules: The remote sensing visual characteristic generation module is used for acquiring unmanned aerial vehicle remote sensing images of the river reach to be monitored, and performing first characterization processing on the unmanned aerial vehicle remote sensing images to obtain remote sensing visual characteristics; The pollution source characteristic generation module is used for acquiring cross-section water quality monitoring data synchronized with the unmanned aerial vehicle remote sensing image in time and space, and performing second characterization processing on the cross-section water quality monitoring data to obtain pollution source characteristics; The environmental semantic feature generation module is used for acquiring the shore zone environmental text data of the river reach to be monitored, and performing third characterization processing on the shore zone environmental text data to obtain environmental semantic features; The system comprises a model construction module, a pollution discharge outlet detection and tracing integrated model, a pollution discharge outlet detection module and a pollution discharge outlet tracing module, wherein the pollution discharge outlet detection and tracing integrated model comprises a pollution discharge outlet detection network and a multi-mode tracing network, the pollution discharge outlet detection network is used for outputting a predicted result of the space position of the pollution discharge outlet, and the multi-mode tracing network is used for outputting a judging result of the attribute of the pollution source; The model training module is used for constructing a composite objective function, and carrying out combined optimization training on the drain outlet detection and tracing integrated model according to the composite objective function, the remote sensing visual characteristics, the pollution source characteristics and the environment semantic characteristics until the preset training requirements are met, so as to obtain a trained drain outlet detection and tracing integrated model; the drain outlet detection and tracing module is used for inputting the current visible light orthographic image of the area to be predicted into the drain outlet detection and tracing integrated model after training is completed, and outputting a current drain outlet prediction result, wherein the current drain outlet prediction result comprises a current drain outlet space position prediction result and a current pollution source attribute discrimination result.

Description

River sewage inlet detection and tracing method and system Technical Field The application relates to the technical field of environmental monitoring, in particular to a river sewage inlet detection and tracing method and system. Background The river inlet drain outlet is a key node for pollutants to enter the river water body, and the management and control effect directly influences the quality and ecological safety of the river basin water environment. Currently, drain inspection is still mainly performed on manual site inspection, and when challenges such as complex landline topography and hidden drain distribution are faced, the traditional manual net-pulling type inspection mode is low in efficiency and high in labor intensity, is easily subjected to influence of environmental and subjective factors, is easy to generate inspection omission, and is difficult to meet the requirement of fine supervision. With the development of remote sensing and intelligent recognition technologies, automatic drain inspection based on remote sensing images has been widely adopted. However, the related art relies on visual features in the remote sensing image to locate and identify the drain. When the sewage outlet is shielded by vegetation, is positioned under water or is discharged through a concealed pipe, the method based on the visual features is difficult to effectively identify. Meanwhile, the related model usually detects a sewage outlet as a single target, mainly focuses on whether the sewage outlet exists or not and the space position thereof, and limits the application of the sewage outlet in the aspects of pollution source attribute discrimination and accurate tracing. In summary, the related drain detection technology still has shortcomings, and improvement is needed. Disclosure of Invention Embodiments of the present application aim to solve at least one of the technical problems in the related art to some extent. Therefore, the main purpose of the embodiment of the application is to provide a method and a system for detecting and tracing a river sewage outlet, which can improve the speed and accuracy for judging the pollution source attribute of the river sewage outlet and effectively improve the identification capability of a hidden sewage outlet. In order to achieve the above purpose, an aspect of the embodiments of the present application provides a method for detecting and tracing a sewage outlet, the method including the following steps: acquiring an unmanned aerial vehicle remote sensing image of a river reach to be monitored, and performing first characterization processing on the unmanned aerial vehicle remote sensing image to obtain remote sensing visual characteristics; Acquiring cross section water quality monitoring data synchronized with the unmanned aerial vehicle remote sensing image in time and space, and performing second characterization processing on the cross section water quality monitoring data to obtain pollution source characteristics; acquiring the shore zone environmental text data of the river reach to be monitored, and performing third characterization processing on the shore zone environmental text data to obtain environmental semantic features; The method comprises the steps of constructing a drain outlet detection and tracing integrated model, wherein the drain outlet detection and tracing integrated model comprises a drain outlet detection network and a multi-mode tracing network, the drain outlet detection network is used for outputting a drain outlet spatial position prediction result, and the multi-mode tracing network is used for outputting a pollution source attribute discrimination result; Constructing a composite objective function, and carrying out joint optimization training on the drain detection and tracing integrated model according to the composite objective function, the remote sensing visual characteristics, the pollution source characteristics and the environment semantic characteristics until a preset training requirement is met, so as to obtain a trained drain detection and tracing integrated model; and inputting the current visible light orthographic image of the region to be predicted into the trained drain detection and tracing integrated model, and outputting a current drain prediction result, wherein the current drain prediction result comprises a current drain spatial position prediction result and a current pollution source attribute discrimination result. In order to achieve the above purpose, another aspect of the embodiments of the present application provides a system for detecting and tracing a sewage outlet, the system comprising: The remote sensing visual characteristic generation module is used for acquiring unmanned aerial vehicle remote sensing images of the river reach to be monitored, and performing first characterization processing on the unmanned aerial vehicle remote sensing images to obtain remote sensing visual characteristics; The p