Cooperative supervision method and system for ecological environment of water area

Abstract

The invention discloses a cooperative supervision method and system for a water area ecological environment, and belongs to the technical field of ecological environment supervision, and the method comprises the steps: carrying out the global monitoring of a target water area, obtaining a monitoring image sequence of the target water area, calculating the pollution confidence, recognizing a pollution region, quantifying the area, and generating a diffusion drive signal, obtaining a category label of the target pollutant; obtaining the motion trail of the target pollutant, reversely deducing the local flow velocity and flow direction of the water surface, identifying an abnormal target in the target water area, and generating an abnormal pollution association table in combination with the pollution area; according to the method, local data of a water area are collected in real time, a water area situation base map is generated, a water quality diffusion model is constructed, a multi-time-dimension pollution diffusion simulation result is generated, and plan deduction and risk assessment are performed, so that a supervision decision is upgraded from passive disposal to active prevention and control, and decision requirements of risk pre-judgment and active prevention and control are fully met.

Description

Technical Field

[0001] This invention relates to a collaborative monitoring method and system for aquatic ecological environment, belonging to the field of ecological environment monitoring technology. Background Technology

[0002] As a vital carrier of the Earth's ecosystem, water bodies bear crucial functions such as water quality assurance, biological habitat, and water resource supply. Their ecological environment is directly related to regional ecological security and people's well-being. With the advancement of industrialization and increased human intervention, the pollution risks and ecological imbalances faced by water bodies are becoming increasingly complex. The traditional regulatory model, which relies on manual sampling and monitoring and post-event source tracing, suffers from shortcomings such as delayed data collection, limited coverage, and low source tracing efficiency. It is difficult to meet the current needs for refined and real-time monitoring of the aquatic ecological environment, and there is an urgent need to build an efficient regulatory system using intelligent technologies.

[0003] A Chinese patent application with publication number CN120494735A discloses an artificial intelligence-based aquatic ecological environment monitoring system, including a pollution monitoring module for acquiring geographical information of the water area, establishing multiple water quality monitoring points, and obtaining pollution data; a reverse inference module for acquiring hydrological models, meteorological data, and data from surrounding enterprises to identify suspected pollution sources; an investigation and reporting module for dispatching underwater robots and drones to investigate suspected pollution sources and generating investigation reports to be sent to regulatory authorities; an ecological data collection module for deploying underwater robots and drones for routine patrols of the water area, constructing a biodiversity database, and constructing a hydrological and water quality database; and an ecological visualization module for constructing a twin map of the water area and generating a visualized ecological environment monitoring map.

[0004] While existing technologies have achieved improvements in the efficiency and real-time nature of aquatic ecological environment supervision, they do not consider dynamic simulation and contingency planning, resulting in weak decision support capabilities. In aquatic supervision scenarios, pollution is not static; its diffusion path is dynamically affected by factors such as water flow and weather, and ecological risks also change over time. For example, after an abnormal water quality is detected in a certain area, only identifying the suspected pollution source and presenting the current status data is not enough to predict which direction the pollution will spread or at what speed. Nor can it be used to plan response plans in advance for potential crises such as pollution spreading to drinking water sources and threatening ecological organisms. This means that regulatory decisions can only remain at the level of passive handling after problems are discovered, which is insufficient to meet the decision-making needs for risk prediction and proactive prevention in aquatic ecological environment supervision. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a collaborative monitoring method and system for aquatic ecological environment. This system obtains multi-source data through observation equipment and IoT sensors, combines image analysis to locate pollution, infer the flow field, construct a situational baseline map, uses the baseline map data to drive a coupled model, optimizes parameters using Kalman filtering to adapt to dynamic changes in pollution, infers pollution trends and assesses risks based on the model, matches and simulates the effectiveness of contingency plans, and realizes the transition from passive discovery to proactive prediction and planning, thus solving the technical problem of weak decision support.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] Collaborative regulatory methods for aquatic ecological environment include:

[0008] The target water area is monitored in its entirety, and a sequence of monitoring images of the target water area is acquired. The pollution confidence level is calculated, the pollution area is identified and its area is quantified, a diffusion driving signal is generated, and the category label of the target pollutant is obtained.

[0009] The movement trajectory of the target pollutant is obtained, the local flow velocity and direction of the water surface are deduced, and abnormal targets in the target water area are identified. Combined with the polluted area, an abnormal pollution association table is generated.

[0010] Real-time local water area data is collected to generate a water area status map, construct a water quality diffusion model, and generate multi-time-dimensional pollution diffusion simulation results for contingency plan simulation and risk assessment.

[0011] Specifically, the steps for calculating the pollution confidence level include:

[0012] The monitoring image frame is scanned pixel by pixel to obtain the HSV value of each pixel. Combined with conventional pixel features, the real-time similarity of each pixel is calculated. ;

[0013] Set candidate similarity threshold Determine whether the pixels in the monitored image frame represent normal water bodies. Mark the pixel as a candidate pixel;

[0014] Construct a candidate pixel marker map of the monitored image frame, perform connected component analysis, and generate multiple local candidate regions;

[0015] The total number of pixels in each local candidate region is counted, and the actual area of ​​the local candidate region is calculated by combining the pixel scale.

[0016] If the actual area is greater than the minimum size threshold, it is determined to be a valid candidate region, and a valid candidate region information table containing region IDs is constructed; otherwise, the local candidate region is eliminated.

[0017] Specifically, the steps for calculating the pollution confidence level also include:

[0018] The average HSV value of all pixels within the effective candidate region is calculated, and the average feature of normal water bodies in the corresponding scene is called to calculate the deviation value of a single feature.

[0019] Perform scene adaptation adjustments to obtain the water body deviation value after scene adaptation;

[0020] Obtain the extreme values ​​of HSV features from historical pollution samples, calculate the range of water body differences for scene adaptation, and thus obtain the color difference coefficient;

[0021] Calculate the real-time texture variance of the effective candidate regions, and combine it with the historical maximum texture variance to calculate the texture difference coefficient;

[0022] Extract the scene adaptation target band ratio of pixels within the effective candidate region, call the corresponding band ratio of normal water body, and calculate the spectral difference coefficient;

[0023] Based on the feature importance weights, the pollution confidence level of each valid candidate region is calculated. .

[0024] Specifically, the step of generating the diffusion driving signal includes:

[0025] Set a second-level confidence threshold , ,and ;

[0026] like If it is determined to be a contaminated area, the initial pixel mask is retained; if If the area is determined to be uncontaminated, the initial pixel mask is deleted.

[0027] like If a region is identified as a suspected contaminated area, a secondary verification is triggered. If any one of the three consecutive frames is identified as a non-contaminated area, it is updated to a non-contaminated area; otherwise, it is updated to a contaminated area.

[0028] Contaminated region pairs are constructed, boundary distances are calculated, and contaminated region pairs whose boundary distances are less than the adjacent boundary threshold are marked as adjacent contaminated regions.

[0029] Calculate the visual feature similarity between adjacent polluted areas. Once the visual feature similarity is greater than the visual similarity threshold, merge the polluted areas, reassign IDs, and update the area boundary coordinates and actual area.

[0030] Specifically, the step of generating the diffusion driving signal further includes:

[0031] After statistical optimization, the total number of pixels in each polluted area is counted, and combined with the pixel scale, the actual merged area of ​​the polluted area is calculated. ;

[0032] The area growth rate is calculated by calling the actual merged area of ​​the contaminated area over three consecutive frames. Simultaneously set the scene adaptation area threshold Diffusion trend threshold ;

[0033] like and A diffusion-driven signal is generated, and the region ID is associated with the region's geographic coordinates; otherwise, monitoring of the entire target water area continues.

[0034] Construct a preliminary classification model, generate pollutant labels and corresponding bounding box coordinates for polluted areas, associate area IDs, and generate a target pollutant classification table;

[0035] Based on the category labels in the target pollutant classification table, each category is processed, and fine-grained indicators are extracted to determine subcategories by combining industry standards and sample statistical characteristics, and the target pollutant classification table is updated.

[0036] Specifically, the steps for identifying anomalous targets include:

[0037] Based on the bounding box coordinates of the target pollutant, calculate the center point image coordinates and obtain grayscale feature points and texture feature points;

[0038] For target pollutants in two consecutive frames, feature points are extracted and a pixel matching window is set to calculate the pixel displacement vector of each feature point between frames.

[0039] Calculate the average displacement vector between frames and combine it with the frame timestamp to generate the motion trajectory of the target pollutant;

[0040] Calculate the pixel displacement of the target pollutant, and in conjunction with the pixel scale, obtain the actual displacement of the target pollutant, and calculate the initial flow velocity;

[0041] Calculate the flow direction angle of the target pollutant, and combine it with the horizontal rotation angle calibration value to obtain the geographical azimuth angle;

[0042] Calculate the flow velocity confidence of the target pollutant. If the flow velocity confidence is less than the flow velocity confidence threshold, mark it as low confidence and remove the current frame, and generate a verification report.

[0043] Specifically, the steps for identifying anomalous targets also include:

[0044] Obtain an abnormal target sample library, construct an anomaly identification model, and generate an initial list of abnormal targets;

[0045] Perform status label determination for abnormal targets to generate a global list of abnormal targets;

[0046] Based on the calibration parameters of the observation equipment, the geographical coordinates of the polluted area are obtained, and the direction coefficient is obtained by combining the geographical azimuth angle.

[0047] Calculate the actual straight-line distance between the abnormal target and the polluted area, combine it with the effective pollution diffusion distance, calculate the diffusion coefficient, obtain the labeling coefficient, and calculate the pollution source confidence level;

[0048] A secondary judgment threshold is set, and based on the confidence level of the pollution source, the association labels of the abnormal targets are divided, and an abnormal pollution association table is constructed.

[0049] Specifically, the steps for constructing a water quality diffusion model include:

[0050] Configure the specific parameters for the target water area, generate an initial parameter table, and initialize the coupled model;

[0051] Obtain the predicted boundary of the contaminated area, calculate the visual bias, and simultaneously obtain the predicted flow velocity and direction to calculate the dynamic bias;

[0052] Based on the aforementioned visual deviation and the historical maximum visual deviation, an initial visual weight is set, and the flow velocity confidence is used as the initial dynamic weight. After normalization, the visual weight and dynamic weight are obtained.

[0053] Calculate the fusion bias, and use the fusion bias as a feedback signal to adjust the parameters in the initial parameter table to generate a coupled model with parameter assimilation.

[0054] Based on historical pollution events, obtain historical pollution diffusion prediction results, and calculate spatial reproducibility accuracy and concentration deviation rate;

[0055] If the spatial reproduction accuracy is greater than the accuracy threshold and the concentration deviation rate is less than the concentration threshold, the coupling model is deemed to have passed verification, and the water quality diffusion model is output; otherwise, the assimilation process of the fusion deviation is returned.

[0056] Specifically, the steps of the contingency plan simulation include:

[0057] A scenario parameter table is constructed, and combined with the water quality diffusion model, multi-time-dimensional simulations are performed to generate pollution diffusion simulation results.

[0058] Based on the simulation results, the risk levels for people's livelihood and ecological risks are classified to obtain the impact coefficient.

[0059] The time weighting coefficient is calculated, and combined with the influence degree coefficient, the comprehensive risk level is calculated. Then, the risk level is divided through the secondary risk threshold.

[0060] Based on the simulation results, corresponding contingency plans are matched according to the pollutant subclass labels and risk levels, and a candidate contingency plan set is generated based on the matching degree.

[0061] For each candidate plan, key performance indicators are calculated to obtain a comprehensive score, which is then ranked to generate a candidate plan performance evaluation table.

[0062] The collaborative monitoring system for aquatic ecological environment includes: a data acquisition module, an identification module, a prediction module, and a simulation module;

[0063] The data acquisition module is used to monitor the entire target water area and acquire monitoring image sequences, and to collect water area data in real time, and to construct a water area situation map in combination with meteorological data;

[0064] The identification module is used to filter candidate pixels, perform connected region analysis, calculate pollution confidence, generate diffusion driving signals, reverse the flow velocity and direction, identify abnormal targets and determine their status, and generate an abnormal pollution association table.

[0065] The prediction module is used to configure dedicated parameters to generate an initial coupling model, calculate the fusion deviation to dynamically optimize the model parameters, verify the model by combining historical pollution event data, and generate a water quality diffusion model.

[0066] The simulation module is used to perform multi-time-dimensional simulations, combine the list of sensitive targets to determine the degree of pollution impact, use the risk matrix method to classify risk levels to generate early warning reports, match contingency plans according to pollutant type and risk level, simulate the effectiveness of the contingency plans, calculate a comprehensive score and recommend the optimal contingency plan.

[0067] The beneficial effects of this invention are:

[0068] By employing pixel-by-pixel scanning and connected region analysis using observation equipment, combined with HSV spatial similarity calculations, local pollution is accurately located, avoiding the obscuring of pollution characteristics across the entire water body. Simultaneously, IoT sensor data and meteorological data are integrated to construct a water situation map, providing accurate foundational data for dynamic simulation. This addresses the problem of simulation distortion caused by incomplete traditional monitoring data. The situation map data drives the coupled model, and through an ensemble Kalman filter algorithm combined with deviation feedback from anomaly pollution association tables, model parameters are dynamically optimized. This allows the model to adapt in real-time to the dynamic changes in pollution with water flow and weather, overcoming the limitation of static models in tracking pollution diffusion. Based on the optimized model, multi-time-dimensional extrapolation is conducted to predict the concentration distribution of pollution plumes, the arrival time of sensitive targets, and peak concentrations. Combined with the risk matrix method for risk classification, this achieves a shift from passively discovering pollution to proactively predicting risks. Contingency plans are intelligently matched according to pollutant type and risk level. The model simulates the effectiveness of the contingency plans and recommends the optimal solution, solving the problem of the inability to plan countermeasures in advance. This upgrades regulatory decision-making from passive handling to proactive prevention and control, fully meeting the decision-making needs of risk prediction and proactive prevention. Attached Figure Description

[0069] Figure 1A schematic diagram illustrating a collaborative regulatory approach for aquatic ecological environments;

[0070] Figure 2 This is a flowchart of the process for generating the diffusion driving signal in this invention;

[0071] Figure 3 This is a flowchart illustrating the process of identifying anomalous targets in this invention;

[0072] Figure 4 This is a flowchart of the water quality diffusion model constructed in this invention. Detailed Implementation

[0073] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0074] Example 1

[0075] refer to Figures 1 to 4 As shown in the figure, this embodiment introduces a collaborative supervision method for the aquatic ecological environment, including the following steps:

[0076] Step S1: Monitor the entire target water area using observation equipment deployed at water monitoring points, and acquire real-time monitoring image sequences of the target water area, including continuous frame video and timed snapshot images. Optimize image quality using image enhancement algorithms to solve image blurring problems in rainy, foggy, and low-light environments at night. Simultaneously, based on the timestamps of image frames, combine linear interpolation to perform frame synchronization alignment of the monitoring image sequence to ensure uniform time intervals between consecutive frames. Combine image segmentation technology to extract water body boundaries from the monitoring image sequence, calculate pollution confidence, and identify and quantize the area of ​​polluted regions, thereby obtaining pixel-level masks and coordinate ranges of polluted regions to generate diffusion driving signals. The water monitoring points are deployed on towers, bridges, or buildings along the shore, and the observation equipment includes, but is not limited to, high-definition cameras and pan-tilt cameras.

[0077] Step S2: Identify target pollutants to generate category labels for target pollutants, analyze the pixel motion vectors between frames of the monitoring image sequence, track the movement trajectory of target pollutants, infer the local flow velocity and direction of the water surface, and simultaneously combine the target pollutants and the polluted area to identify abnormal targets in the target water area, so as to screen out abnormal targets that cause or aggravate pollution. The coordinates of the identified abnormal targets are spatially superimposed with the coordinates of the polluted area to determine whether the abnormal targets are located upstream of the polluted area or in the polluted area, thereby generating an abnormal pollution association table.

[0078] Step S3: By deploying IoT sensors at water collection points, local monitoring of the target water area is carried out, and water data is collected in real time. At the same time, meteorological data is obtained in real time through the meteorological department's API. Combined with the electronic map of the target water area and the abnormal pollution correlation table, a water situation base map is constructed. The water collection points are deployed downstream of sewage outlets, drinking water source inlets, and river cross-sections. The IoT sensors include, but are not limited to, water quality buoys and underwater detectors. The water data includes, but is not limited to, water temperature, pH value, dissolved oxygen, chemical oxygen demand, ammonia nitrogen, total phosphorus, as well as water flow velocity, flow direction, and water level.

[0079] Step S4: Using water area data and meteorological data from the water area status map as the driving source, import the hydrodynamic-water quality coupling model, such as EFDC, to calculate the water area flow field distribution. Combine the water quality data to construct a water quality diffusion model. Use the abnormal pollution association table as the observed value and the output of the water quality diffusion model as the simulated value to calculate the deviation. Combine the ensemble Kalman filter algorithm and use the deviation value as the feedback signal to automatically adjust the key parameters of the water quality diffusion model.

[0080] Step S5: Once a diffusion-driven signal is detected or a simulation is manually triggered, the water quality diffusion model is used to generate multi-time-dimensional pollution diffusion simulation results for different time dimensions. This outputs the concentration distribution, diffusion boundary, estimated time of arrival at each sensitive target, and peak concentration of the pollution plume at different time periods. Contingency plan simulations are then performed to generate prediction results under differentiated scenarios. The simulation results are overlaid with the attributes of the sensitive targets to determine whether the pollution plume will affect the sensitive targets and the extent of the impact. Based on the toxicity of the pollutants and the simulated concentration, the risk level is classified using the risk matrix method, and a risk warning report is generated.

[0081] Specifically, the steps for generating the diffusion driving signal include:

[0082] Directly calculating features from the global image can easily mask local pollution features due to the high proportion of normal water body pixels. Therefore, a pixel-by-pixel scan is performed based on the latest monitoring image frame in the monitoring image sequence to accurately locate each abnormal water body pixel, obtain the HSV value of each pixel, and simultaneously call upon the corresponding conventional pixel features from the normal water body visual feature library. Based on the HSV spatial Euclidean distance formula, the real-time similarity of each pixel is calculated. To avoid missing local pollution due to the global averaging effect; the normal water body visual feature library is obtained by extracting features from normal water bodies labeled by those skilled in the art.

[0083] Set candidate similarity threshold To determine whether the pixels in the monitored image frame represent normal water bodies; if This indicates that the pixel is an abnormal water pixel and is marked as a candidate pixel; if This indicates that the pixel at this time is a normal water body pixel;

[0084] Based on the form of a binary image, a candidate pixel label map of the monitoring image frame is constructed with 1 as candidate pixels and 0 as normal water body pixels. Connectivity analysis is performed to cluster adjacent candidate pixels into a local candidate region, such as candidate pixels on the top, bottom, left, right and diagonal lines. A unique ID is assigned to each region, and the total number of pixels in each local candidate region is counted. Combined with the pixel scale, the actual area of ​​the local candidate region is calculated.

[0085] Set a minimum size threshold for contaminated areas, and filter out valid candidate areas by combining the actual area of ​​each local candidate area. If the actual area is greater than the minimum size threshold, it is determined to be a valid candidate area, and the initial pixel mask, region boundary image coordinates and actual area of ​​the valid candidate area are obtained. Construct a valid candidate area information table containing region IDs to locate all potential local contaminated areas in the global image. If the actual area is not greater than the minimum size threshold, it is determined to be image noise and removed.

[0086] For each valid candidate region, it is processed one by one by ID. The average HSV of all pixels in the valid candidate region is calculated, including the average values ​​of hue, saturation, and brightness. The average features of normal water bodies in the corresponding scene are called, the deviation value of each feature is calculated, and scene adaptation adjustment is performed. The deviation values ​​of each feature are summed and averaged to obtain the water body deviation value after scene adaptation, which reflects the degree of color difference between the valid candidate region and normal water bodies. For example, the scene adaptation adjustment is performed as follows: since the brightness value of rivers fluctuates greatly due to the influence of silt, it has low reference value. In the river scene, only hue and saturation are counted. The water body deviation value is calculated by the hue deviation value and the saturation deviation value.

[0087] Obtain the extreme values ​​of HSV features of historical pollution samples, such as hue maxima and minima. Also consider scene adaptation. Calculate the range of water body differences for scene adaptation by summing the extreme value differences of each feature. Calculate the color difference coefficient based on the ratio of the water body deviation value to the water body difference range.

[0088] For pixels within the valid candidate region, local binary mode is used to extract texture features. The real-time texture variance of the LBP values ​​of all pixels within the region is calculated to reflect the non-uniformity of the texture within the region. The maximum texture variance of historical contaminated samples is called. Based on the ratio of the real-time texture variance to the maximum texture variance, the texture difference coefficient is calculated.

[0089] Extract the scene adaptation target band ratio of pixels within the effective candidate region, call the corresponding band ratio of normal water body, and calculate the spectral difference coefficient through absolute difference operation;

[0090] Based on feature importance weights and combined with color, texture, and spectral difference coefficients, the contamination confidence level of each valid candidate region is calculated. And set a second confidence threshold, including the first confidence threshold. Second confidence threshold , To effectively distinguish between truly contaminated areas and suspected areas;

[0091] like If it is determined to be a contaminated area, the initial pixel mask is retained, and the initial list of contaminated areas is entered; if If it is determined to be a non-contaminated area, the initial pixel mask is deleted, and a list of non-contaminated areas to be removed is entered; if If a region is identified as a suspected contaminated area, a secondary verification is immediately triggered for the next three frames to avoid missed contamination due to the global averaging effect. If any one of the three consecutive frames is identified as a non-contaminated area, it is updated to a non-contaminated area to avoid incorrect judgments caused by transient interference; otherwise, it is updated to a contaminated area.

[0092] Morphological closing operations are performed on the initial pixel mask of each contaminated area to fill the tiny holes in the mask whose number of pixels is less than the minimum hole pixel threshold. At the same time, the mask edges are smoothed to avoid area calculation errors caused by jagged edges.

[0093] During the process of identifying contaminated areas, various factors such as image noise, blurred edges of contaminants, and limited segmentation accuracy of algorithms can cause the same physical contaminant to be misidentified as multiple independent contaminated areas. Based on any two contaminated areas, multiple pairs of contaminated areas are constructed. The boundary distance between the pairs of contaminated areas is calculated. Contaminated area pairs with a boundary distance less than the adjacent boundary threshold are marked as adjacent contaminated areas. The visual feature similarity of adjacent contaminated areas is calculated. Once the visual feature similarity is greater than the visual similarity threshold, they are determined to be different parts of the same contaminant. The initial pixel mask is merged into a complete contaminated area, the ID is reassigned, and the area boundary coordinates and actual area are updated.

[0094] After statistical optimization, the total number of pixels in each complete contaminated area is counted. Combined with the pixel scale, the actual merged area of ​​the contaminated area is calculated. The actual merged area of ​​the contaminated region for three consecutive frames is called. , , ,by , The difference is the numerator. and , The product of the time intervals is used as the denominator to calculate the area growth rate. Simultaneously set the scene adaptation area threshold Diffusion trend threshold ;

[0095] like and Generate a diffusion-driven signal and associate the region ID with the region's geographic coordinates; otherwise, continue monitoring the entire target water area.

[0096] Obtain a historical pollution sample dataset, including pollutant-annotated images and corresponding masks. Combine transfer learning and scene fine-tuning to train a YOLOv8 model and generate a preliminary classification model. Pollutant categories include oil, biological, chemical, and solid floating pollutants, which are then arranged according to the priority of pollutant disposal.

[0097] Input the pixel mask of the polluted area into the preliminary classification model to perform coarse classification of pollutants within the mask, generate pollutant labels and corresponding bounding box coordinates for the polluted area, associate the area ID, and generate a target pollutant classification table.

[0098] Based on the major category labels in the target pollutant classification table, each category is processed individually. Combining industry standards and sample statistical characteristics, fine-grained indicators are extracted to determine subcategories, and the target pollutant classification table is updated. Specifically, for oil pollutants, gradient direction histograms are extracted. If the reflectance intensity gradient is unidirectional, it is marked as a light oil film; if it is multidirectional, it is marked as a heavy oil film. For biological pollutants, the proportion of green pixels in the polluted area is statistically analyzed to classify them into cyanobacterial blooms and aquatic plant decay. For chemical pollutants, a mapping table containing visual features and water quality indicators is used, such as dark brown patches corresponding to high COD and grayish-white flocculent matter corresponding to high ammonia nitrogen. Sensor data or pre-trained regression models are combined to associate water quality indicators. For solid floating pollutants, the roundness is calculated based on the pollutant outline and they are classified into plastic fragments and plastic bottles and cans.

[0099] Specifically, the steps for identifying anomalous targets include:

[0100] Since a single feature point is prone to trajectory breakage due to local occlusion, such as the shadow of a bridge pier covering the center point or floating objects obscuring the edge point, the combination of basic anchor points and local detail points can maintain trajectory continuity through the other two points when a certain feature point is occluded, balancing tracking stability and computational efficiency. Based on the bounding box coordinates of the target pollutant, the image coordinates of the center point are calculated as the basic anchor point of the trajectory. The reflective areas of oil pollutants and the bright edges of solid pollutants are often the points with the maximum gray value. All pixels within the bounding box are traversed, and the gray value of each pixel is calculated. The pixel with the maximum gray value is determined as the gray feature point to capture local motion details. The flocculent texture of chemical pollutants and the irregular edges of biological pollutants often correspond to high texture variance. The pixels within the bounding box are divided into sub-regions according to preset pixel blocks, and the LBP texture variance of each sub-region is calculated. The center point of the sub-region with the maximum texture variance is determined as the texture feature point to supplement edge motion deviation. The feature points of the target pollutant are saved to the target pollutant classification table.

[0101] For target pollutants in two consecutive frames, inter-frame feature point matching is performed based on the Lucas-Kanade optical flow method. Feature points are extracted as seed points, a pixel matching window is set, and the pixel displacement vector of each feature point between frames is calculated, including the displacement of the geometric center point, the displacement of grayscale feature points, and the displacement of texture feature points. If a feature point is not extracted due to occlusion, Kalman filtering is started. Based on the feature point displacement trend of the previous 3 frames, the coordinates of the occluded feature points in the current frame are predicted, and the displacement vector is completed.

[0102] Different weights are assigned to each feature point. Combined with the pixel displacement vector, the average displacement vector between frames is calculated. Combined with the frame timestamp, the average displacement coordinates of each frame are recorded in chronological order to generate the timestamped motion trajectory of the target pollutant. At the same time, the trajectory is filtered by a 5-frame moving average to eliminate single-frame displacement jumps caused by sudden gusts of wind and image noise.

[0103] In the motion trajectory, the average displacement vector of 10 consecutive frames is extracted. Based on the Euclidean distance formula, the pixel displacement of the target pollutant is calculated and converted into the actual displacement of the target pollutant by combining the pixel scale. The initial flow velocity of the target pollutant is calculated with the total time of 10 consecutive frames as the denominator.

[0104] Based on the latest average displacement vector in the trajectory, the flow direction angle of the target pollutant is calculated using the arctangent formula, and the geographical azimuth is obtained by combining the horizontal rotation calibration value of the observation equipment.

[0105] The number of unmatched deviation feature points in each frame is obtained. The deviation coefficient is calculated by multiplying the preset matching coefficient by the number of deviation feature points. The flow velocity confidence of the target pollutant is calculated by multiplying the preset base coefficient by the deviation coefficient. Once the flow velocity confidence is less than the flow velocity confidence threshold, it is marked as low confidence and the current frame is removed. Finally, a verification report is generated, including the region ID, the category of the target pollutant, the initial flow velocity (including confidence), and the geographical azimuth. The base coefficient and the matching coefficient are set by those skilled in the art. In this embodiment, the base coefficient is 1 and the matching coefficient is 0.2.

[0106] A scenario-based abnormal target sample library is obtained and divided into sub-libraries according to scenario labels. For example, the river scenario sub-library contains illegal industrial sewage outlets and small cargo boats illegally dumping waste, while the lake scenario sub-library contains sewage outlets at dams and illegal dumping by sightseeing boats. Each sub-library sample contains labeled images, bounding boxes, and functional subtype labels. Combined with the Faster R-CNN model, the model is trained using transfer learning and scenario fine-tuning. For the river scenario, the recognition weights of industrial sewage outlets and small cargo boats are optimized. For the lake scenario, the weights of sewage outlets at dams and sightseeing boats are optimized to improve scenario adaptability. Finally, an anomaly recognition model is generated. The monitoring image sequence is input into the anomaly recognition model to generate an initial list of abnormal targets, including the target temporary ID, abnormal target type, and bounding box coordinates.

[0107] Anomaly target status label determination is performed. For sewage outlet targets, color features of abnormal targets are continuously extracted from multiple frames of images. If abnormal water bodies are identified in multiple consecutive frames, the status label is marked as continuous discharge; otherwise, it is marked as intermittent discharge. For vessel targets, the motion trajectory within multiple frames is tracked. If the displacement distance is less than a preset distance threshold, the status label is marked as dwell timeout; otherwise, it is marked as normal navigation. For identified drainage outlets and vessels, if the color features and motion status match the visual feature library of normal water bodies, they are determined to be normal facilities and removed from the list. Finally, a global abnormal target list is generated, including target temporary ID, functional subtype, bounding box coordinates, and status label. The status labels include continuous discharge, intermittent discharge, dwell timeout, and normal navigation.

[0108] Based on the calibration parameters of the observation equipment, the center point image coordinates are converted into geographic coordinates using the perspective projection formula to ensure that they are consistent with the coordinate format of the polluted area. The geographic coordinates of the polluted area are used as the origin, and the direction of the incoming flow, i.e. the upstream direction, is determined based on the geographic azimuth. If the abnormal target is located on the side of the incoming flow direction, it is marked as upstream; otherwise, it is marked as downstream. The Haversine formula is used to calculate the actual straight-line distance between the abnormal target and the polluted area to reflect the spatial correlation between the two.

[0109] Based on the actual straight-line distance, whether the abnormal target is located upstream of the target pollutant, and the status label, an effective pollution diffusion distance is set. The diffusion coefficient is calculated by subtracting the ratio of the actual straight-line distance to the effective pollution diffusion distance from the base coefficient. When the actual straight-line distance is less than the effective pollution diffusion distance, whether the abnormal target is located upstream of the target pollutant is quantified into specific data: 1 for upstream and 0 for downstream, to obtain the direction coefficient. Similarly, the status label is quantified into specific data to obtain the label coefficient. The pollution source confidence is calculated by weighted summation of the diffusion coefficient, direction coefficient, and label coefficient.

[0110] A two-tiered judgment threshold is set, including a first judgment threshold and a second judgment threshold, wherein the first judgment threshold is greater than the second judgment threshold; if the confidence level of the pollution source is greater than the first judgment threshold, the associated label of the abnormal target is marked as a suspected pollution source; if the confidence level of the pollution source is less than the second judgment threshold, the abnormal target is marked as an irrelevant anomaly and removed; if the confidence level of the pollution source is within the range of the second judgment threshold and the first judgment threshold, the abnormal target is marked as a pollution aggravating factor.

[0111] Finally, an abnormal pollution association table is constructed, which includes target temporary ID, pollution area ID, functional subtype, geographic coordinates, pollution source confidence level, and association label.

[0112] Specifically, the steps for constructing a water quality diffusion model include:

[0113] By calling known hydrodynamic-water quality coupling models, such as EFDC, and configuring specific parameters for the target water pollution scenario based on multi-dimensional data collected / computed, the general model can be adapted to specific regulatory needs. For hydrodynamic parameter configuration, the initial flow velocity and geographical azimuth are used as the initial flow velocity field, and the target water topographic data, such as water depth and shoreline boundary, are used to generate the model calculation grid. Hydrological monitoring data is used to set the upstream and downstream boundary conditions of the model. For water quality parameter configuration, the pollutant subclass labels are used to match differentiated water quality parameters, such as diffusion coefficient and degradation rate, and the associated labels are used as the initial pollution concentration field. The parameters are bound through a shared calculation grid to ensure that the hydrodynamic parameters drive the pollutant diffusion calculation of the water quality module in real time, thereby generating the initial parameter table and initializing the coupling model.

[0114] The system obtains the latest geographic coordinates, initial flow velocity, and geographic azimuth of the pollution area boundary in the target water area. Based on the initial coupling model, it calculates the predicted pollution area boundary, calculates the visual bias using the Euclidean distance formula, and obtains the predicted flow velocity and flow direction. It calculates the absolute deviation of flow velocity and absolute deviation of flow direction, assigns a weighting coefficient to the absolute deviation of flow direction, and calculates the dynamic bias.

[0115] The initial visual weight is set by the ratio of visual deviation to the historical maximum visual deviation, and the initial dynamic weight is set by the flow velocity confidence. The initial visual weight and the initial dynamic weight are normalized to obtain the visual weight and dynamic weight, and the fusion deviation of the polluted area is calculated.

[0116] An ensemble Kalman filter algorithm is adopted, with the fusion bias as the feedback signal. The parameters are corrected for the current real-time pollution scenario. The parameters in the initial parameter table are adjusted and iterated until the bias meets the threshold judgment rule. The optimized parameters are output for real-time parameter calibration. The optimized parameters are substituted into the initial coupled model to make the model prediction more in line with the current reality, thereby generating a coupled model after parameter assimilation.

[0117] Archived historical pollution event data with known diffusion results are obtained and input into the coupled model to obtain historical pollution diffusion prediction results for historical scenario verification and optimization. The predicted pollution area and the actual pollution area in the historical event are extracted. The spatial reproduction accuracy is calculated by using the intersection area of ​​the predicted pollution area and the actual pollution area as the numerator and the union area of ​​the predicted pollution area and the actual pollution area as the denominator. Simultaneously, the predicted pollutant concentration and the archived measured concentration are extracted. The concentration deviation rate is calculated by using the absolute difference between the predicted pollutant concentration and the archived measured concentration as the numerator and the archived measured concentration as the denominator.

[0118] If the spatial reproduction accuracy is greater than the accuracy threshold and the concentration deviation rate is less than the concentration threshold, the coupled model is deemed to have passed the verification and the water quality diffusion model is output; otherwise, the assimilation process of the fusion deviation is traced back, the weight of the observations is adjusted, such as increasing the weight of high-confidence visual data in historical events, and the assimilation process is run again until the verification is passed.

[0119] Specifically, the triggering plan simulation includes automatic triggering and manual triggering. Automatic triggering occurs when a diffusion-driven signal is detected, while manual triggering occurs when a simulation scenario is manually set based on pollutant subclass labels. For example, if a chemical plant upstream leaks benzene-containing wastewater, the system automatically matches the historical best parameters corresponding to the benzene-containing wastewater, such as diffusion coefficient and degradation rate, without requiring manual parameter input, thus improving efficiency.

[0120] The steps involved in contingency plan simulation include:

[0121] A scenario parameter table is constructed to input the water quality diffusion model for multi-time-dimensional simulation. For each time point, the concentration distribution of the pollution plume, the diffusion boundary, and the arrival time and peak concentration of sensitive targets are output. The diffusion boundary of the pollution plume is extracted using the concentration contour method, that is, the contour lines formed by the excess concentration threshold of the corresponding pollutant are selected as the boundary and then converted into geographic coordinates. The arrival time of sensitive targets is calculated by the straight-line distance between the boundary of the pollution plume and the geographic coordinates of the sensitive targets at different time points. When the distance is zero, the arrival time and the peak concentration at that time are recorded, thereby generating the pollution diffusion simulation results. Based on the electronic map of the target water area, a list of sensitive targets is constructed. Sensitive targets include the names of drinking water intakes, ecological protection red lines, aquaculture areas, and swimming areas, as well as their geographic coordinates and water quality thresholds / protection standards.

[0122] The simulation results were stratified and sorted. Based on whether the pollution plume would affect sensitive targets, the simulation results were divided into two groups. Predictions with sensitive targets arriving at a time less than the sensitive time threshold and peak concentrations greater than the sensitive concentration threshold were selected and classified into the sensitive target group according to the correspondence between target name, arrival time, and peak concentration. Other prediction results were directly classified into the non-sensitive target group.

[0123] By comparing the peak concentration of pollutants at sensitive targets with the corresponding water quality thresholds in the list of sensitive targets, the level of public welfare risk is determined. If the peak concentration does not exceed the threshold, there is no public welfare risk; if it exceeds the threshold but is less than twice the threshold, it is a mild public welfare risk; if it exceeds twice the threshold, it is a severe public welfare risk. The biodiversity database is then used to compare the peak concentration with the pollutant tolerance thresholds of aquatic organisms to determine the level of ecological risk. If the peak concentration does not exceed half of the tolerance threshold, there is no ecological risk; if it exceeds half but is less than the tolerance threshold, it is a mild ecological risk; if it exceeds the tolerance threshold, it is a severe ecological risk.

[0124] Different values ​​are assigned to different livelihood risks and ecological risks. For each prediction result, the maximum value of the livelihood risk and ecological risk is selected and defined as the impact degree coefficient. The time weight coefficient is calculated based on the ratio of the arrival time of the sensitive target to the preset sensitive time threshold. Based on the weighted superposition, the comprehensive risk degree is calculated. The risk level of the prediction result is divided through the preset secondary risk threshold, thereby generating a risk warning report, including the name and type of the affected target, the arrival time of the sensitive target, the peak concentration, the comprehensive risk degree, and the risk level.

[0125] Intelligent contingency plan matching is performed. For each prediction result, corresponding contingency plans are matched based on the pollutant subclass label. For example, oil pollutants are matched with contingency plans related to oil booms and oil-absorbing pads, and chemical wastewater is matched with contingency plans related to reagent dosing and activated carbon adsorption. At the same time, contingency plans are matched based on the risk level. High-risk contingency plans are matched with emergency response plans to ensure the deployment of pollution prevention facilities in a short time, and medium-risk contingency plans are matched with conventional response plans to arrange the response work according to the conventional process. After the matching is completed, multiple candidate contingency plans are generated and sorted by matching degree to generate a candidate contingency plan set. The matching degree is calculated by superimposing the pollutant type adaptation score and the risk level adaptation score with a certain weight. The risk level adaptation score is determined by the risk level, and different risk levels correspond to different adaptation scores. The pollutant type adaptation score is called from a preset adaptation score library, which stores the correspondence between pollutant type contingency plans and adaptation scores.

[0126] For each candidate contingency plan, the implementation parameters are input into the water quality diffusion model, and the corresponding parameters of the model are adjusted. For example, when implementing the oil boom contingency plan, the oil boom boundary is added to the model and constraints that prevent pollutants from crossing are set. When implementing the chemical dosing contingency plan, the degradation rate of pollutants in the model is adjusted according to historical verification data. The key performance indicators after the implementation of the contingency plan are calculated, including the pollution control rate, the ecological restoration cycle, and the disposal cost. Among them, the pollution control rate is the percentage change in peak concentration at sensitive targets before and after the implementation of the contingency plan, the ecological restoration cycle is the time required for the pollutant concentration to drop to a certain percentage of the tolerance threshold of aquatic organisms, and the disposal cost is the total cost calculated based on the contingency plan material list and market unit price.

[0127] The comprehensive score of each candidate plan is calculated by taking into account the pollution control rate, ecological restoration cycle and disposal cost. The candidate plans are ranked according to the comprehensive score, the plan with the highest score is recommended, and a candidate plan effectiveness evaluation table is generated, which includes the candidate plan name, implementation parameters, various effectiveness indicators and comprehensive score.

[0128] Example 2

[0129] Another embodiment of the present invention provides a collaborative monitoring system for aquatic ecological environment, comprising: a data acquisition module, an identification module, a prediction module, and a deduction module;

[0130] The data acquisition module is used to monitor the entire target water area and acquire monitoring image sequences through observation equipment deployed at water monitoring points. The image quality is optimized and the frame interval is unified through image enhancement and frame synchronization processing. At the same time, the target water area is monitored locally through IoT sensors deployed at water collection points, and water area data is collected in real time. Meteorological data is obtained through meteorological department API. The water area status map is constructed by combining the electronic map of the target water area and the abnormal pollution association table.

[0131] The identification module is used to scan the monitoring image sequence pixel by pixel, calculate the real-time similarity by combining HSV spatial Euclidean distance, and select candidate pixels to construct a candidate pixel label map of the monitoring image frame. Based on connected component analysis and pollution confidence judgment, it identifies the pollution area and generates diffusion driving signals to avoid the global water body from masking local pollution features. At the same time, it trains the YOLOv8 model to coarsely classify pollutants in the pollution area and then divides them into subclasses by combining fine-grained indicators to generate a target pollutant classification table. It also tracks the movement trajectory of pollutants by combining multiple feature points to infer the flow velocity and direction. Combined with the Faster R-CNN model, it identifies abnormal targets and determines their status. Through spatial overlay analysis, it generates an abnormal pollution association table, which solves the problems of easy omission and trajectory breakage in traditional identification and accurately locates the pollution source and pollutant type.

[0132] The prediction module is used to call the coupled model, configure exclusive parameters to generate an initial coupled model, use the abnormal pollution association table as the observation value, calculate the fusion deviation with the simulated value of the initial coupled model, dynamically optimize the model parameters through the ensemble Kalman filter algorithm, verify the model by combining historical pollution event data, and output a water quality diffusion model. This overcomes the limitation of traditional static models that cannot adapt to the dynamic changes of pollution with water flow and meteorology, and ensures the accuracy of pollution diffusion prediction.

[0133] The simulation module is used to construct a simulation scenario parameter table based on diffusion-driven signals or manual triggering, and to conduct multi-time-dimensional simulations. It outputs the concentration distribution of pollution plumes, diffusion boundaries, and arrival times and peak concentrations of sensitive targets. Combined with a list of sensitive targets, it judges the degree of pollution impact, uses the risk matrix method to classify risk levels and generate early warning reports, matches contingency plans according to pollutant type and risk level, simulates the effectiveness of contingency plans, calculates a comprehensive score, and recommends the optimal contingency plan. It realizes full-process support from pollution discovery to risk prediction and contingency plan planning, solves the problem of passive handling in traditional supervision, and meets the needs of aquatic ecological environment supervision for risk prediction and proactive prevention and control.

[0134] Working principle and effects:

[0135] The monitoring image sequence of the target water area is obtained by the observation equipment deployed on the shore. After image enhancement and frame synchronization processing, the abnormal water body pixels are located by scanning pixel by pixel and calculating HSV spatial similarity. Then, the effective pollution area is screened by connected component analysis. At the same time, the target pollutant category is identified, its movement trajectory is tracked to infer the flow velocity and direction, and an association table is generated by identifying abnormal targets in the pollution area.

[0136] Simultaneously relying on IoT sensors to collect water area data and access meteorological data, and integrating electronic maps to construct a water area situation base map, it solves the problems of traditional monitoring that easily misses local pollution and has incomplete data, providing comprehensive data support for subsequent analysis. The situation base map data drives the coupled model, and the abnormal pollution correlation table is used to calculate the deviation between the observed value and the model simulation value. The model parameters are dynamically optimized by ensemble Kalman filtering, so that the model can adapt to the dynamic changes of pollution under the influence of water flow and meteorology, breaking through the limitation of traditional static model simulation distortion.

[0137] When pollution signals are detected or simulations are manually triggered, multi-time-dimensional pollution diffusion results are generated. The impact level is judged and the risk level is classified by combining the attributes of sensitive targets. At the same time, contingency plans are matched according to pollutant type and risk level, and the effectiveness of the contingency plans is simulated to recommend the optimal solution. Ultimately, the entire process from pollution discovery to risk prediction and contingency plan planning is supported, enabling regulatory decision-making to be upgraded from passive disposal to proactive prevention and control, and meeting the needs of aquatic ecological environment supervision for risk prediction and proactive response.

[0138] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A collaborative regulatory method for aquatic ecological environment, characterized in that, include: The target water area is monitored in its entirety, and a sequence of monitoring images of the target water area is acquired. The pollution confidence level is calculated, the pollution area is identified and its area is quantified, a diffusion driving signal is generated, and the category label of the target pollutant is obtained. The movement trajectory of the target pollutant is obtained, the local flow velocity and direction of the water surface are deduced, and abnormal targets in the target water area are identified. Combined with the polluted area, an abnormal pollution association table is generated. Real-time local water area data is collected to generate a water area status map, construct a water quality diffusion model, and generate multi-time-dimensional pollution diffusion simulation results for contingency plan simulation and risk assessment.

2. The collaborative supervision method for aquatic ecological environment according to claim 1, characterized in that, The steps for calculating the pollution confidence level include: The monitoring image frame is scanned pixel by pixel to obtain the HSV value of each pixel. Combined with conventional pixel features, the real-time similarity of each pixel is calculated. ; Set candidate similarity threshold Determine whether the pixels in the monitored image frame represent normal water bodies. Mark the pixel as a candidate pixel; Construct a candidate pixel marker map of the monitored image frame, perform connected component analysis, and generate multiple local candidate regions; The total number of pixels in each local candidate region is counted, and the actual area of ​​the local candidate region is calculated by combining the pixel scale. If the actual area is greater than the minimum size threshold, it is determined to be a valid candidate region, and a valid candidate region information table containing region IDs is constructed; otherwise, the local candidate region is eliminated.

3. The collaborative supervision method for aquatic ecological environment according to claim 2, characterized in that, The steps for calculating the pollution confidence level also include: The average HSV value of all pixels within the effective candidate region is calculated, and the average feature of normal water bodies in the corresponding scene is called to calculate the deviation value of a single feature. Perform scene adaptation adjustments to obtain the water body deviation value after scene adaptation; Obtain the extreme values ​​of HSV features from historical pollution samples, calculate the range of water body differences for scene adaptation, and thus obtain the color difference coefficient; Calculate the real-time texture variance of the effective candidate regions, and combine it with the historical maximum texture variance to calculate the texture difference coefficient; Extract the scene adaptation target band ratio of pixels within the effective candidate region, call the corresponding band ratio of normal water body, and calculate the spectral difference coefficient; Based on the feature importance weights, the pollution confidence level of each valid candidate region is calculated. .

4. The collaborative supervision method for aquatic ecological environment according to claim 3, characterized in that, The steps for generating the diffusion drive signal include: Set a second-level confidence threshold , ,and ; like If it is determined to be a contaminated area, the initial pixel mask is retained; if If the area is determined to be uncontaminated, the initial pixel mask is deleted. like If a region is identified as a suspected contaminated area, a secondary verification is triggered. If any one of the three consecutive frames is identified as a non-contaminated area, it is updated to a non-contaminated area; otherwise, it is updated to a contaminated area. Contaminated region pairs are constructed, boundary distances are calculated, and contaminated region pairs whose boundary distances are less than the adjacent boundary threshold are marked as adjacent contaminated regions. Calculate the visual feature similarity between adjacent polluted areas. Once the visual feature similarity is greater than the visual similarity threshold, merge the polluted areas, reassign IDs, and update the area boundary coordinates and actual area.

5. The collaborative supervision method for aquatic ecological environment according to claim 4, characterized in that, The step of generating the diffusion drive signal further includes: After statistical optimization, the total number of pixels in each polluted area is counted, and combined with the pixel scale, the actual merged area of ​​the polluted area is calculated. ; The area growth rate is calculated by calling the actual merged area of ​​the contaminated area over three consecutive frames. Simultaneously set the scene adaptation area threshold Diffusion trend threshold ; like and A diffusion-driven signal is generated, and the region ID is associated with the region's geographic coordinates; otherwise, monitoring of the entire target water area continues. Construct a preliminary classification model, generate pollutant labels and corresponding bounding box coordinates for polluted areas, associate area IDs, and generate a target pollutant classification table; Based on the category labels in the target pollutant classification table, each category is processed, and fine-grained indicators are extracted to determine subcategories by combining industry standards and sample statistical characteristics, and the target pollutant classification table is updated.

6. The collaborative supervision method for aquatic ecological environment according to claim 5, characterized in that, The steps for identifying anomalous targets include: Based on the bounding box coordinates of the target pollutant, calculate the center point image coordinates and obtain grayscale feature points and texture feature points; For target pollutants in two consecutive frames, feature points are extracted and a pixel matching window is set to calculate the pixel displacement vector of each feature point between frames. Calculate the average displacement vector between frames and combine it with the frame timestamp to generate the motion trajectory of the target pollutant; Calculate the pixel displacement of the target pollutant, and in conjunction with the pixel scale, obtain the actual displacement of the target pollutant, and calculate the initial flow velocity; Calculate the flow direction angle of the target pollutant, and combine it with the horizontal rotation angle calibration value to obtain the geographical azimuth angle; Calculate the flow velocity confidence of the target pollutant. If the flow velocity confidence is less than the flow velocity confidence threshold, mark it as low confidence and remove the current frame, and generate a verification report.

7. The collaborative supervision method for aquatic ecological environment according to claim 6, characterized in that, The steps for identifying anomalous targets also include: Obtain an abnormal target sample library, construct an anomaly identification model, and generate an initial list of abnormal targets; Perform status label determination for abnormal targets to generate a global list of abnormal targets; Based on the calibration parameters of the observation equipment, the geographical coordinates of the polluted area are obtained, and the direction coefficient is obtained by combining the geographical azimuth angle. Calculate the actual straight-line distance between the abnormal target and the polluted area, combine it with the effective pollution diffusion distance, calculate the diffusion coefficient, obtain the labeling coefficient, and calculate the pollution source confidence level; A secondary judgment threshold is set, and based on the confidence level of the pollution source, the association labels of the abnormal targets are divided, and an abnormal pollution association table is constructed.

8. The collaborative supervision method for aquatic ecological environment according to claim 7, characterized in that, The steps to construct a water quality diffusion model include: Configure the specific parameters for the target water area, generate an initial parameter table, and initialize the coupled model; Obtain the predicted boundary of the contaminated area, calculate the visual bias, and simultaneously obtain the predicted flow velocity and direction to calculate the dynamic bias; Based on the aforementioned visual deviation and the historical maximum visual deviation, an initial visual weight is set, and the flow velocity confidence is used as the initial dynamic weight. After normalization, the visual weight and dynamic weight are obtained. Calculate the fusion bias, and use the fusion bias as a feedback signal to adjust the parameters in the initial parameter table to generate a coupled model with parameter assimilation. Based on historical pollution events, obtain historical pollution diffusion prediction results, and calculate spatial reproducibility accuracy and concentration deviation rate; If the spatial reproduction accuracy is greater than the accuracy threshold and the concentration deviation rate is less than the concentration threshold, the coupling model is deemed to have passed verification, and the water quality diffusion model is output; otherwise, the assimilation process of the fusion deviation is returned.

9. The collaborative supervision method for aquatic ecological environment according to claim 8, characterized in that, The steps involved in contingency plan simulation include: A scenario parameter table is constructed, and combined with the water quality diffusion model, multi-time-dimensional simulations are performed to generate pollution diffusion simulation results. Based on the simulation results, the risk levels for people's livelihood and ecological risks are classified to obtain the impact coefficient. The time weighting coefficient is calculated, and combined with the influence degree coefficient, the comprehensive risk level is calculated. Then, the risk level is divided through the secondary risk threshold. Based on the simulation results, corresponding contingency plans are matched according to the pollutant subclass labels and risk levels, and a candidate contingency plan set is generated based on the matching degree. For each candidate plan, key performance indicators are calculated to obtain a comprehensive score, which is then ranked to generate a candidate plan performance evaluation table.

10. A collaborative monitoring system for aquatic ecological environment, used to implement the collaborative monitoring method for aquatic ecological environment as described in any one of claims 1-9, characterized in that, include: Data acquisition module, identification module, prediction module, and inference module; The data acquisition module is used to monitor the entire target water area and acquire monitoring image sequences, and to collect water area data in real time, and to construct a water area situation map in combination with meteorological data; The identification module is used to filter candidate pixels, perform connected region analysis, calculate pollution confidence, generate diffusion driving signals, reverse the flow velocity and direction, identify abnormal targets and determine their status, and generate an abnormal pollution association table. The prediction module is used to configure dedicated parameters to generate an initial coupling model, calculate the fusion deviation to dynamically optimize the model parameters, verify the model by combining historical pollution event data, and generate a water quality diffusion model. The simulation module is used to perform multi-time-dimensional simulations, combine the list of sensitive targets to determine the degree of pollution impact, use the risk matrix method to classify risk levels to generate early warning reports, match contingency plans according to pollutant type and risk level, simulate the effectiveness of the contingency plans, calculate a comprehensive score and recommend the optimal contingency plan.

Images