Marine pollution tracing and monitoring method based on satellite remote sensing and unmanned aerial vehicle

Abstract

The invention provides a marine pollution tracing and monitoring method based on satellite remote sensing and an unmanned aerial vehicle, and relates to the technical field of marine pollution tracing and monitoring, and the method comprises the steps: configuring a pollution diffusion path prediction model and a data driving model based on range pollution data, precise pollution data and deepwater data after multi-modal data fusion, and obtaining a pollution prediction result and pollution error data, carrying out hybrid drive fusion based on the pollution prediction result and the pollution error data, and tracking an initial source of pollutants to obtain pollution source tracing data. Through preprocessing and fusion of multi-modal data, in combination with range pollution data, precise pollution data and deep water data, a pollution diffusion path prediction model and a data driving model are dynamically adjusted, so that a more precise pollution prediction result and pollution error data are generated, and meanwhile, through reverse diffusion simulation and pollution traceability analysis, a more precise pollution prediction result is obtained. The high efficiency of pollution source positioning and decision support is improved, and an accurate and decision basis is provided for coping with marine pollution events.

Description

Technical Field

[0001] The present invention relates to the technical field of marine pollution source tracing and monitoring, and more particularly to a marine pollution source tracing and monitoring method based on satellite remote sensing and unmanned aerial vehicles. Background Art

[0002] As the problem of marine pollution becomes increasingly serious, tracking and monitoring pollution sources has become particularly important. Existing technologies mostly use a single monitoring method, such as satellite remote sensing or drone data collection methods. Although they can cover large-area monitoring needs, they have limitations in data accuracy, deep pollution detection and real-time performance. The spread of marine pollution is affected by the combined effects of fluid dynamics, pollution source intensity and environmental parameters. Therefore, it is necessary to combine multiple data sources and analysis methods to build an accurate pollution tracing and prediction system.

[0003] In practical applications, existing methods have independent and scattered data processing processes and fail to fully integrate multimodal data, resulting in low accuracy and reliability of prediction models. Finally, traditional decision-making systems fail to dynamically adjust strategies based on historical risk data and real-time monitoring results, making it difficult to adapt to rapid changes in pollution situations, resulting in delayed governance effects. Summary of the invention

[0004] The purpose of the present invention is to provide a method for tracing and monitoring marine pollution based on satellite remote sensing and drones to address the deficiencies of the existing technology. The method aims to build a comprehensive multimodal pollution monitoring system by integrating satellite remote sensing, drone data and deep-water monitoring data, and use a hybrid drive model to achieve high-precision pollution spread prediction and source tracking. Combined with dynamic evaluation and real-time optimization decision-making mechanisms, this method significantly improves the comprehensiveness of monitoring, the accuracy of prediction and the flexibility of decision-making, and provides a scientific basis for rapid and effective marine pollution control. To achieve the above object, the present invention provides the following technical solutions: Marine pollution source tracing and monitoring methods based on satellite remote sensing and drones include the following: Step S100: Obtain range pollution data, precise pollution data and deep water data.

[0005] Step S200: pre-processing and multi-modal data fusion are performed based on the obtained range pollution data, precise pollution data and deep water data.

[0006] Step S300: Based on the range pollution data, precise pollution data and deep-water data after multimodal data fusion, a pollution diffusion path prediction model and a data-driven model are configured to obtain pollution prediction results and pollution error data, and hybrid-driven fusion is performed based on the pollution prediction results and pollution error data, and the initial source of the pollutants is tracked to obtain pollution source tracing data.

[0007] Step S400: Generate three-dimensional volume data based on the obtained pollution source tracing data and the pollution prediction results and pollution error data after hybrid drive fusion, and obtain a three-dimensional visual model through time series processing. Based on the obtained pollution source tracing data, mark the location of the pollution source through the three-dimensional visual model.

[0008] Step S500: Preset classification thresholds and risk index calculations are performed based on the pollution prediction results and pollution error data after hybrid drive fusion to obtain risk index data and potential impact area data, and decision presets and decision preset optimizations are performed based on the obtained risk index data and potential impact area data, and visualization processing is performed and report content is generated.

[0009] Step S600, obtain the latest risk index data, compare the historical risk index data with the latest risk index data to obtain the comparison result, analyze the current execution decision preset to obtain the analysis result, and generate the final report result based on the obtained report content, report results, comparison results and analysis results.

[0010] As a preferred solution of the present invention, obtaining range pollution data, precise pollution data and deep water data specifically includes: Step S100.1, obtaining range pollution data, precise pollution data and deep water data.

[0011] Through satellite remote sensing, we conduct large-scale coverage monitoring of pollutants in the ocean surface and obtain range pollution data, which include: pollutant distribution range, pollutant concentration, pollutant type, location of potential pollution sources, ocean current speed and direction, surface water temperature and surface salinity.

[0012] The range pollution data is obtained by preliminarily identifying the distribution and concentration of pollutants through satellite remote sensing high-resolution multispectral images, extracting the spectral characteristics of pollutants using hyperspectral images to classify substances, and capturing areas of abnormal temperature using thermal infrared to locate pollution sources.

[0013] Through the use of drones, high-precision monitoring of hot spots and areas that are difficult to cover by satellites can be carried out to obtain accurate pollution data, which includes: refined pollutant distribution, pollutant concentration, pollutant characteristics, specific pollution source types, biological pollution sources, local water temperature, local water flow speed and local water transparency.

[0014] Accurate pollution data is obtained by using drones to carry spectral sensors to identify the optical properties of different pollutants, using high-resolution cameras to obtain clear pollutant distribution maps, and using environmental DNA sampling devices to collect water samples for biological traceability analysis.

[0015] The deepwater data are obtained by collecting dissolved organic matter, heavy metal content and pH value of columnar water through buoy monitoring. The deepwater data include: deep pollutant concentration, vertical distribution of pollutants, type of deepwater pollution source, intensity of pollution source, deep water temperature, deep salinity, deep water flow speed and direction and water density.

[0016] Deepwater data are obtained by moving underwater robots in the target area and collecting deepwater samples of pollutants.

[0017] As a preferred solution of the present invention, preprocessing and multimodal data fusion are performed based on the obtained range pollution data, precise pollution data and deep water data, specifically: Step S200.1: pre-processing is performed based on the obtained range pollution data, precise pollution data and deep water data.

[0018] Preprocessing includes: image denoising, numerical data cleaning and spectral data normalization.

[0019] Image denoising uses convolutional neural networks to denoise range pollution data, identify and repair noise areas, and perform geometric distortion correction on precise pollution data. Numerical data cleaning uses an RNN-based time series anomaly detection model to identify and remove sudden error data points. Spectral data standardization uses normalization to uniformly adjust spectral data collected by multiple sensors.

[0020] It should be noted that, since preprocessing is a well-known technology in the art, it will not be described in detail here.

[0021] Step S200.2: Perform multimodal data fusion based on the pre-processed range pollution data, precise pollution data and deep water data.

[0022] Multimodal data fusion includes: spatiotemporal alignment, feature association enhancement and weighted fusion.

[0023] Spatiotemporal alignment is based on the dynamic time warping algorithm to align the time series of multi-source data, use the geospatial interpolation algorithm to fill in the data of uncovered areas, strengthen the feature correlation, and map the spectral data, thermal infrared data and deep-water physical parameters to a unified feature space through multimodal feature maps. The graph neural network is used to establish a correlation model between data to improve the accuracy and robustness of data fusion. Weighted fusion combines spatiotemporal alignment and feature correlation enhancement to dynamically assign weights to the range pollution data, precision pollution data and deep-water data quality. Specifically: Where: is the initial weight of the data, is the resolution factor, adjusted according to the relative quality of resolution. is the timeliness factor, adjusted according to the impact of data delay, is the data integrity factor, adjusted for the missing rate, It is the weighted influencing factor corresponding to resolution, timeliness and completeness.

[0024] It should be noted here that since multimodal data fusion is a well-known technology in the art, it will not be elaborated here.

[0025] The experimental settings are: Because the resolution weight has a lower impact, The timeliness impact is medium. It has a great impact on data integrity.

[0026] The weight factor is set as: Resolution factor , Timeliness Factor and integrity factor .

[0027] Resolution is less than 1 meter, , resolution 1m to 10m, , with a resolution better than 10 meters, .

[0028] Data delay is less than 30 minutes, , data delay 30 to 60 minutes, , data delay exceeds 60 minutes, .

[0029] The data missing rate is lower than , data missing rate to , the data missing rate is higher than .

[0030] As a preferred solution of the present invention, a pollution diffusion path prediction model and a data-driven model are configured based on the range pollution data, precise pollution data and deep-water data after multimodal data fusion to obtain pollution prediction results and pollution error data. Hybrid driven fusion is performed based on the pollution prediction results and pollution error data, and the initial source of the pollutants is tracked to obtain pollution source tracing data, specifically: Step S300.1: Based on the range pollution data, precise pollution data and deep water data after multimodal data fusion, a pollution diffusion path prediction model is configured, specifically: Where: is the pollutant concentration, is the fluid velocity field, is the diffusion coefficient, which is dynamically adjusted to accommodate differences in water quality, is the pollutant source term, representing the intensity of the new pollutant.

[0031] The finite volume method is used to discretize the partial differential equations, and the parallel computing is used to accelerate the solution to obtain the pollution prediction results.

[0032] Step S300.2: configuring a data-driven model based on the range pollution data, precision pollution data and deepwater data after multimodal data fusion.

[0033] Based on deep learning, a physically enhanced neural network is adopted, and the law of conservation of diffusion is added in the training process. The physical error and data error are combined through the loss function to obtain the pollution error data.

[0034] Step S300.3: Perform hybrid drive fusion based on the obtained pollution prediction results and pollution error data, specifically: Where: is the final fused pollutant concentration prediction value, is the pollutant concentration predicted by the physical driving model, is the pollutant concentration predicted by the data-driven model and are the weight factors of the physics-driven model and the data-driven model respectively.

[0035] Step S300.4: Based on the pollution prediction results and pollution error data after hybrid drive fusion, the initial source of the pollutants is tracked through reverse diffusion simulation, specifically: Where: is the geographical location of the pollution source, is the current distribution location of pollutants, is the fluid velocity field, describing the path of the pollutants moving with the environmental dynamics is the current time, is the initial time when pollution occurs.

[0036] Extract the spectral features of range pollution data and precise pollution data, match them with the known pollutant feature library, extract the water sample DNA information in the precise pollution data, compare it with the biological database of polluted areas, identify possible biological pollution sources, and obtain pollution source tracing data.

[0037] As a preferred solution of the present invention, three-dimensional volume data is generated based on the pollution source tracing data obtained, the pollution prediction results after hybrid drive fusion, and the pollution error data, and a three-dimensional visual model is obtained through time series processing. Based on the pollution source tracing data obtained, the pollution source position is marked through the three-dimensional visual model, specifically: Step S400.1: Generate three-dimensional data based on the obtained pollution source tracing data, the pollution prediction results after hybrid drive fusion, and the pollution error data.

[0038] The three-dimensional volume data is based on the grid unit of 10m×10m×1m, and the three-dimensional grid is obtained. The pollutant concentration values ​​are directly mapped to the 3D grid, and a color gradient is used to represent the concentration range.

[0039] Color gradients include: blue, yellow, and red.

[0040] Blue represents low concentration 0-10 mg / L, yellow represents medium concentration 10-50 mg / L, and red represents high concentration greater than 50 mg / L.

[0041] Arrange the blue below, the yellow in the middle, and the red above, and generate a frame-by-frame animation of pollutant diffusion through time series, combined with the fluid velocity field , vector arrows are superimposed on the three-dimensional grid to intuitively display the diffusion direction and obtain a three-dimensional visual model.

[0042] In the three-dimensional grid, the length of the vector arrow represents the flow velocity, and the direction indicates the direction of fluid movement. It is dynamically superimposed on the three-dimensional visual model to show the trend and direction of pollutant diffusion.

[0043] Step S400.2: Based on the obtained pollution source tracing data, the pollution source locations are marked through a three-dimensional visual model to obtain the pollution source marking results.

[0044] The labels include: location of pollution source, intensity of pollution source and type of pollution source.

[0045] When the user clicks on the pollution source label, detailed information is provided, including: spectral characteristics, environmental DNA and pollution prediction results.

[0046] As a preferred solution of the present invention, based on the pollution prediction results and pollution error data after hybrid drive fusion, preset classification thresholds and risk index calculations are performed to obtain risk index data and potential impact area data, and decision presets and decision preset optimizations are performed based on the obtained risk index data and potential impact area data, and visualization processing and report content are generated, specifically including: Step S500.1: Preset a classification threshold based on the pollution prediction result and pollution error data after hybrid drive fusion.

[0047] Classification thresholds include: low pollution area, medium pollution area and high pollution area.

[0048] when , then it is a low pollution area. , then it is a medium pollution area. , then it is: highly polluted area.

[0049] Highly polluted areas represent a serious threat to the ecosystem, moderately polluted areas require monitoring, and low polluted areas are basically harmless.

[0050] Based on the three-dimensional visual model with the location of the pollution source marked, the pollution area is divided by preset classification thresholds to generate a pollution distribution zoning map, and the pollutant types are marked and classified in combination with the pollution source tracing data.

[0051] Step S500.2: Calculate the risk index based on the pollution prediction result after the preset classification threshold and the pollution error data to obtain the risk index data, specifically: Where: It is a risk index that quantifies the threat posed by pollutants to the environment. is the pollutant concentration, is the environmental sensitivity factor, is the intensity of the pollution source.

[0052] Priority areas are set based on the resulting risk index data.

[0053] The priority area is set as follows: , then it is a high priority area. , then it is a medium priority area. , it is a low priority area, and combined with the pollution distribution zoning map, a priority map is generated to mark the high-risk areas.

[0054] Based on the obtained risk index data and the pollution prediction results and pollution error data after the preset classification threshold, the pollution diffusion path prediction model is used to predict the pollution diffusion path in the next N hours and obtain the potential impact area data.

[0055] Step S500.3: Decision presetting is performed based on the obtained risk index data and potential impact area data.

[0056] In high-priority areas, adsorption equipment will be deployed to treat oil films and garbage salvage equipment will be deployed. In medium-priority areas, chemical neutralization agents will be used to decompose pollutants. In low-priority areas, ecological recovery will be promoted by releasing repair bacteria.

[0057] Assign decision presets to clean-up equipment and monitoring personnel based on priority maps.

[0058] Step S500.4: Use the gradient boosting tree to perform decision preset optimization.

[0059] The gradient boosting tree is trained using the historical risk index data and pollutant concentrations in the database as the training set. The model accuracy is evaluated using the cross-validation method, and the gradient boosting tree hyperparameters are adjusted through grid search.

[0060] Step S500.5: Perform visualization processing based on the obtained risk index data and the decision preset results.

[0061] High priority areas are painted red, medium priority areas are painted yellow, and low priority areas are painted green.

[0062] Dynamically display the impact of different decision preset results on pollutant concentrations and generate report content.

[0063] The report content includes: risk index data, priority map, 3D visual model, pollution source marking results and decision preset results.

[0064] As a preferred solution of the present invention, the latest risk index data is obtained, and the historical risk index data is compared with the latest risk index data to obtain the comparison result, and the current execution decision preset is analyzed to obtain the analysis result. The final report result is generated based on the obtained report content, report results, comparison results and analysis results, which specifically includes: Step S600.1, obtain the latest risk index data.

[0065] The risk index data is listed as historical risk index data, and the historical risk index data is compared with the latest risk index data, specifically: Where: Indicates the change in risk index, , it means the risk is increased, , it means the risk is reduced. , it means the risk remains stable. Indicates the change in pollution concentration, , it means that the pollutant concentration has increased and the control measures need to be strengthened. , it means that the pollutant concentration has decreased and the current management is effective.

[0066] according to and The change value of is used to divide the governance effect into three categories: high efficiency, medium efficiency and low efficiency, and generate a report result, which includes: concentration reduction rate, diffusion range control effect and risk reduction range.

[0067] Step S600.2: Analyze the current execution decision preset.

[0068] The analysis includes: resource input and governance coverage.

[0069] like or When When the number of monitoring points increases, the diffusion path is closely tracked. When the current strategy begins to show results, the current strength will be maintained.

[0070] Step S600.3: Generate a final report result based on the obtained report content, report results, comparison results of historical risk index data and latest risk index data, and decision presets.

[0071] It is generated through office and output in docx format. The final report results include: comparative analysis results of historical risk index data and the latest risk index data, pollution concentration distribution, pollution sources, decision-making preset analysis results, risk index changes, and pollution prediction results.

[0072] Compared with the prior art, the present invention has the following beneficial effects: 1. Through the fusion of multimodal data from satellite remote sensing, drone monitoring and deep-water sampling, full coverage monitoring of pollutants from the surface to deep water areas is achieved, overcoming the technical bottleneck that a single data source cannot accurately reflect the three-dimensional pollution characteristics, and greatly improving the comprehensiveness and accuracy of the monitoring system.

[0073] 2. By introducing multimodal data fusion technology and utilizing dynamic time warping algorithm, feature correlation enhancement and weighted fusion method, multi-source heterogeneous data are effectively integrated, the robustness and consistency of the data are improved, and high-quality input data is provided for the subsequent pollution diffusion prediction model, thereby significantly improving the accuracy of the model.

[0074] 3. Based on the hybrid driving model, combining the advantages of the physical driving model and the data-driven model, the present invention can achieve high-precision pollution diffusion prediction and pollution source tracing. At the same time, it can dynamically adjust the model parameters to adapt to real-time monitoring data to ensure the reliability of pollution diffusion path prediction and tracing results.

[0075] 4. Through the preprocessing and fusion of multimodal data, combined with range pollution data, precise pollution data and deep-water data, the pollution diffusion path prediction model and data-driven model are dynamically adjusted to generate more accurate pollution prediction results and pollution error data. At the same time, through reverse diffusion simulation and pollution source tracing analysis, the efficiency of pollution source positioning and decision-making support is significantly improved, providing accurate and real-time decision-making basis for rapid response to marine pollution incidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1A flow chart of a method for tracing and monitoring marine pollution based on satellite remote sensing and drones provided in an embodiment of the present application. DETAILED DESCRIPTION

[0077] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0078] See also Figure 1 , Figure 1 A flowchart of a method for tracing and monitoring marine pollution based on satellite remote sensing and drones is provided for an embodiment of the present application.

[0079] In this embodiment, the marine pollution source tracing and monitoring method based on satellite remote sensing and drones may include steps S100, S200, S300, S400, S500 and S600; Step S100: Obtain range pollution data, precise pollution data and deep water data.

[0080] Step S200: pre-processing and multi-modal data fusion are performed based on the obtained range pollution data, precise pollution data and deep water data.

[0081] Step S300: Based on the range pollution data, precise pollution data and deep-water data after multimodal data fusion, a pollution diffusion path prediction model and a data-driven model are configured to obtain pollution prediction results and pollution error data, and hybrid-driven fusion is performed based on the pollution prediction results and pollution error data, and the initial source of the pollutants is tracked to obtain pollution source tracing data.

[0082] Step S400: Generate three-dimensional volume data based on the obtained pollution source tracing data and the pollution prediction results and pollution error data after hybrid drive fusion, and obtain a three-dimensional visual model through time series processing. Based on the obtained pollution source tracing data, mark the location of the pollution source through the three-dimensional visual model.

[0083] Step S500: Preset classification thresholds and risk index calculations are performed based on the pollution prediction results and pollution error data after hybrid drive fusion to obtain risk index data and potential impact area data, and decision presets and decision preset optimizations are performed based on the obtained risk index data and potential impact area data, and visualization processing is performed and report content is generated.

[0084] Step S600, obtain the latest risk index data, compare the historical risk index data with the latest risk index data to obtain the comparison result, analyze the current execution decision preset to obtain the analysis result, and generate the final report result based on the obtained report content, report results, comparison results and analysis results.

[0085] In a specific implementation, obtaining range pollution data, precise pollution data and deep water data specifically includes: Step S100.1, obtaining range pollution data, precise pollution data and deep water data.

[0086] Through satellite remote sensing, we conduct large-scale coverage monitoring of pollutants in the ocean surface and obtain range pollution data, which include: pollutant distribution range, pollutant concentration, pollutant type, location of potential pollution sources, ocean current speed and direction, surface water temperature and surface salinity.

[0087] The range pollution data is obtained by preliminarily identifying the distribution and concentration of pollutants through satellite remote sensing high-resolution multispectral images, extracting the spectral characteristics of pollutants using hyperspectral images to classify substances, and capturing areas of abnormal temperature using thermal infrared to locate pollution sources.

[0088] Through the use of drones, high-precision monitoring of hot spots and areas that are difficult to cover by satellites can be carried out to obtain accurate pollution data, which includes: refined pollutant distribution, pollutant concentration, pollutant characteristics, specific pollution source types, biological pollution sources, local water temperature, local water flow speed and local water transparency.

[0089] Accurate pollution data is obtained by using drones to carry spectral sensors to identify the optical properties of different pollutants, using high-resolution cameras to obtain clear pollutant distribution maps, and using environmental DNA sampling devices to collect water samples for biological traceability analysis.

[0090] The deepwater data are obtained by collecting dissolved organic matter, heavy metal content and pH value of columnar water through buoy monitoring. The deepwater data include: deep pollutant concentration, vertical distribution of pollutants, type of deepwater pollution source, intensity of pollution source, deep water temperature, deep salinity, deep water flow speed and direction and water density.

[0091] Deepwater data are obtained by moving underwater robots in the target area and collecting deepwater samples of pollutants.

[0092] In a specific implementation, preprocessing and multimodal data fusion are performed based on the obtained range pollution data, precise pollution data and deep water data, specifically: Step S200.1: pre-processing is performed based on the obtained range pollution data, precise pollution data and deep water data.

[0093] Preprocessing includes: image denoising, numerical data cleaning and spectral data normalization.

[0094] Image denoising uses convolutional neural networks to denoise range pollution data, identify and repair noise areas, and perform geometric distortion correction on precise pollution data. Numerical data cleaning uses an RNN-based time series anomaly detection model to identify and remove sudden error data points. Spectral data standardization uses normalization to uniformly adjust spectral data collected by multiple sensors.

[0095] It should be noted that, since preprocessing is a well-known technology in the art, it will not be described in detail here.

[0096] Step S200.2: Perform multimodal data fusion based on the pre-processed range pollution data, precise pollution data and deep water data.

[0097] Multimodal data fusion includes: spatiotemporal alignment, feature association enhancement and weighted fusion.

[0098] Spatiotemporal alignment is based on the dynamic time warping algorithm to align the time series of multi-source data, use the geospatial interpolation algorithm to fill in the data of uncovered areas, strengthen the feature correlation, and map the spectral data, thermal infrared data and deep-water physical parameters to a unified feature space through multimodal feature maps. The graph neural network is used to establish a correlation model between data to improve the accuracy and robustness of data fusion. Weighted fusion combines spatiotemporal alignment and feature correlation enhancement to dynamically assign weights to the range pollution data, precision pollution data and deep-water data quality. Specifically: Where: is the initial weight of the data, is the resolution factor, adjusted according to the relative quality of resolution. is the timeliness factor, adjusted according to the impact of data delay, is the data integrity factor, adjusted for the missing rate, It is the weighted influencing factor corresponding to resolution, timeliness and completeness.

[0099] It should be noted here that since multimodal data fusion is a well-known technology in the art, it will not be elaborated here.

[0100] The experimental settings are: Because the resolution weight has a lower impact, The timeliness impact is medium. It has a great impact on data integrity.

[0101] The weight factor is set as: Resolution factor , Timeliness Factor and integrity factor .

[0102] Resolution is less than 1 meter, , resolution 1m to 10m, , with a resolution better than 10 meters, .

[0103] Data delay is less than 30 minutes, , data delay 30 to 60 minutes, , data delay exceeds 60 minutes, .

[0104] The data missing rate is lower than , data missing rate to , the data missing rate is higher than .

[0105] If the satellite data has a low resolution of 30 meters, a delay of 30 minutes, and a completeness of 100%, the drone data has a medium resolution of 10 meters, a delay of 15 minutes, and a completeness of 95%, and the deep-water data has no resolution, a delay of 1 hour, and a completeness of 92%, then the calculation weights are: Satellite data weight : Drone data weight : Deepwater acquisition data weight : The final fusion feature value is: In a specific implementation, a pollution diffusion path prediction model and a data-driven model are configured based on the range pollution data, precise pollution data, and deep-water data after multimodal data fusion to obtain pollution prediction results and pollution error data. Hybrid drive fusion is performed based on the pollution prediction results and pollution error data, and the initial source of the pollutants is tracked to obtain pollution source tracing data, specifically: Step S300.1: Based on the range pollution data, precise pollution data and deep water data after multimodal data fusion, a pollution diffusion path prediction model is configured, specifically: Where: is the pollutant concentration, is the fluid velocity field, is the diffusion coefficient, which is dynamically adjusted to accommodate differences in water quality, is the pollutant source term, representing the intensity of the new pollutant.

[0106] The finite volume method is used to discretize the partial differential equations, and the parallel computing is used to accelerate the solution to obtain the pollution prediction results.

[0107] Step S300.2: configuring a data-driven model based on the range pollution data, precision pollution data and deepwater data after multimodal data fusion.

[0108] Based on deep learning, a physically enhanced neural network is used, and the law of conservation of diffusion is added during the training process. The contamination error data is obtained by combining the physical error and the data error through the loss function, which is as follows: Where: It is used to optimize data-driven models. is the residual of the diffusion equation, is the error between pollution prediction and observation, and are the weight factors of physical error and data error respectively.

[0109] Step S300.3: Perform hybrid drive fusion based on the obtained pollution prediction results and pollution error data, specifically: Where: is the final fused pollutant concentration prediction value, is the pollutant concentration predicted by the physical driving model, is the pollutant concentration predicted by the data-driven model and are the weight factors of the physics-driven model and the data-driven model respectively.

[0110] Step S300.4: Based on the pollution prediction results and pollution error data after hybrid drive fusion, the initial source of the pollutants is tracked through reverse diffusion simulation, specifically: Where: is the geographical location of the pollution source, is the current distribution location of pollutants, is the fluid velocity field, describing the path of the pollutants moving with the environmental dynamics is the current time, is the initial time when pollution occurs.

[0111] Extract the spectral features of range pollution data and precise pollution data, match them with the known pollutant feature library, extract the water sample DNA information in the precise pollution data, compare it with the biological database of polluted areas, identify possible biological pollution sources, and obtain pollution source tracing data.

[0112] In a specific implementation, three-dimensional volume data is generated based on the pollution source tracing data obtained, the pollution prediction results after hybrid drive fusion, and the pollution error data, and a three-dimensional visual model is obtained through time series processing. Based on the pollution source tracing data obtained, the pollution source position is marked through the three-dimensional visual model, specifically: Step S400.1: Generate three-dimensional data based on the obtained pollution source tracing data, the pollution prediction results after hybrid drive fusion, and the pollution error data.

[0113] The three-dimensional volume data is based on the grid unit of 10m×10m×1m, and the three-dimensional grid is obtained. The pollutant concentration values ​​are directly mapped to the 3D grid, and a color gradient is used to represent the concentration range.

[0114] Color gradients include: blue, yellow, and red.

[0115] Blue represents low concentration 0-10 mg / L, yellow represents medium concentration 10-50 mg / L, and red represents high concentration greater than 50 mg / L.

[0116] Arrange the blue below, the yellow in the middle, and the red above, and generate a frame-by-frame animation of pollutant diffusion through time series, combined with the fluid velocity field , vector arrows are superimposed on the three-dimensional grid to intuitively display the diffusion direction and obtain a three-dimensional visual model.

[0117] In the three-dimensional grid, the length of the vector arrow represents the flow velocity, and the direction indicates the direction of fluid movement. It is dynamically superimposed on the three-dimensional visual model to show the trend and direction of pollutant diffusion.

[0118] Step S400.2: Based on the obtained pollution source tracing data, the pollution source locations are marked through a three-dimensional visual model to obtain the pollution source marking results.

[0119] The labels include: location of pollution source, intensity of pollution source and type of pollution source.

[0120] When the user clicks on the pollution source label, detailed information is provided, including: spectral characteristics, environmental DNA and pollution prediction results.

[0121] In a specific implementation, based on the pollution prediction results and pollution error data after hybrid drive fusion, preset classification thresholds and risk index calculations are performed to obtain risk index data and potential impact area data, and decision presets and decision preset optimizations are performed based on the obtained risk index data and potential impact area data, and visualization processing and report content are generated, specifically including: Step S500.1: Preset a classification threshold based on the pollution prediction result and pollution error data after hybrid drive fusion.

[0122] Classification thresholds include: low pollution area, medium pollution area and high pollution area.

[0123] when , then it is a low pollution area. , then it is a medium pollution area. , then it is: highly polluted area.

[0124] Highly polluted areas represent a serious threat to the ecosystem, moderately polluted areas require monitoring, and low polluted areas are basically harmless.

[0125] Based on the three-dimensional visual model with the location of the pollution source marked, the pollution area is divided by preset classification thresholds to generate a pollution distribution zoning map, and the pollutant types are marked and classified in combination with the pollution source tracing data.

[0126] Step S500.2: Calculate the risk index based on the pollution prediction result after the preset classification threshold and the pollution error data to obtain the risk index data, specifically: Where: It is a risk index that quantifies the threat posed by pollutants to the environment. is the pollutant concentration, is the environmental sensitivity factor, is the intensity of the pollution source.

[0127] Priority areas are set based on the resulting risk index data.

[0128] The priority area is set as follows: , then it is a high priority area. , then it is a medium priority area. , it is a low priority area, and combined with the pollution distribution zoning map, a priority map is generated to mark the high-risk areas.

[0129] Based on the obtained risk index data and the pollution prediction results and pollution error data after the preset classification threshold, the pollution diffusion path prediction model is used to predict the pollution diffusion path in the next N hours and obtain the potential impact area data.

[0130] Step S500.3: Decision presetting is performed based on the obtained risk index data and potential impact area data.

[0131] In high-priority areas, adsorption equipment will be deployed to treat oil films and garbage salvage equipment will be deployed. In medium-priority areas, chemical neutralization agents will be used to decompose pollutants. In low-priority areas, ecological recovery will be promoted by releasing repair bacteria.

[0132] Assign decision presets to clean-up equipment and monitoring personnel based on priority maps.

[0133] Step S500.4: Use the gradient boosting tree to perform decision preset optimization.

[0134] The gradient boosting tree is trained using the historical risk index data and pollutant concentrations in the database as the training set. The model accuracy is evaluated using the cross-validation method, and the gradient boosting tree hyperparameters are adjusted through grid search.

[0135] Step S500.5: Perform visualization processing based on the obtained risk index data and the decision preset results.

[0136] High priority areas are painted red, medium priority areas are painted yellow, and low priority areas are painted green.

[0137] Dynamically display the impact of different decision preset results on pollutant concentrations and generate report content.

[0138] The report content includes: risk index data, priority map, 3D visual model, pollution source marking results and decision preset results.

[0139] In a specific implementation method, the latest risk index data is obtained, and the historical risk index data is compared with the latest risk index data to obtain a comparison result, and the current execution decision preset is analyzed to obtain an analysis result. The final report result is generated based on the obtained report content, report results, comparison results and analysis results, which specifically includes: Step S600.1, obtain the latest risk index data.

[0140] The risk index data is listed as historical risk index data, and the historical risk index data is compared with the latest risk index data, specifically: Where: Indicates the change in risk index, , it means the risk is increased, , it means the risk is reduced. , it means the risk remains stable. Indicates the change in pollution concentration, , it means that the pollutant concentration has increased and the control measures need to be strengthened. , it means that the pollutant concentration has decreased and the current management is effective.

[0141] according to and The change value of is used to divide the governance effect into three categories: high efficiency, medium efficiency and low efficiency, and generate a report result, which includes: concentration reduction rate, diffusion range control effect and risk reduction range.

[0142] Step S600.2: Analyze the current execution decision preset.

[0143] The analysis includes: resource input and governance coverage.

[0144] like or When When the number of monitoring points increases, the diffusion path is closely tracked. When the current strategy begins to show results, the current strength will be maintained.

[0145] Step S600.3: Generate a final report result based on the obtained report content, report results, comparison results of historical risk index data and latest risk index data, and decision presets.

[0146] It is generated through office and output in docx format. The final report results include: comparative analysis results of historical risk index data and the latest risk index data, pollution concentration distribution, pollution sources, decision-making preset analysis results, risk index changes, and pollution prediction results.

[0147] In the above content, in practical application, first, obtain range pollution data, precise pollution data and deep water data, conduct large-scale coverage monitoring of ocean surface pollutants through satellite remote sensing, obtain range pollution data, including pollutant distribution range, pollutant concentration, pollutant type, potential pollution source location, ocean current speed and direction, surface water temperature and surface salinity, preliminarily identify the distribution and concentration of pollutants through satellite remote sensing high-resolution multispectral images, use hyperspectral to extract pollutant spectral characteristics to complete material classification, and locate pollution sources through thermal infrared capture of temperature anomaly areas, then, use drones to conduct high-precision monitoring of hot spots and areas that are difficult for satellites to cover, and obtain precise pollution data, including pollutant distribution refinement information, concentration, characteristics, specific pollution source types, biological pollution sources, local water temperature, local water flow speed and local water transparency, finally, collect dissolved organic matter, heavy metal content and pH value data of columnar water through buoy monitoring, and collect deep water samples with underwater robots to obtain deep water data, including deep pollutant concentration, vertical distribution of pollutants, pollution source type and intensity, as well as deep water temperature, salinity, water flow speed and direction and water density.

[0148] Then, the obtained range pollution data, precise pollution data and deepwater data are preprocessed and multimodal data fused. The preprocessing includes image denoising, numerical data cleaning and spectral data standardization. The image data is denoised and the noise area is repaired through convolutional neural networks. The precise pollution data is corrected for geometric distortion. The numerical data cleaning adopts a method based on the time series anomaly detection model to eliminate sudden error data points. The spectral data is standardized through normalization technology to eliminate the differences in data collected by different sensors. Multimodal data fusion includes spatiotemporal alignment, feature association enhancement and weighted fusion. The dynamic time warping algorithm is used to align the time series of multi-source data, and the geospatial interpolation method is combined to fill the data of uncovered areas. The correlation of multimodal features is further strengthened through graph neural networks, and dynamic weighted fusion is performed according to the resolution, timeliness and completeness of the data to generate high-quality fused data.

[0149] Next, the pollution diffusion path prediction model and data-driven model are configured based on the multimodal data fusion results to generate pollution prediction results and pollution error data. The pollution diffusion path prediction model is based on the diffusion equation: in, represents the pollutant concentration, is the fluid velocity field, is the diffusion coefficient, For the intensity of pollution sources, the partial differential equations are discretely solved by the finite volume method, combined with parallel computing acceleration to obtain the pollution diffusion prediction results. The data-driven model combines the law of conservation of pollution diffusion through a physically enhanced neural network. During the training process, the physical error and data error are combined in the loss function for optimization to generate pollution error data.

[0150] Then, a hybrid driven fusion prediction of pollutant concentration is performed based on the pollution prediction results and pollution error data, combining the prediction results of the physical driven model and the data driven model, using the weighted formula: in, and The weight factors are adjusted dynamically, and the advantages of the physical model and the data model are integrated to generate the final fused pollutant concentration prediction value.

[0151] Next, the pollution source is traced through reverse diffusion simulation and characteristic analysis, and the source of the pollutants is traced back according to the pollution diffusion path prediction results: in, is the location of the pollution source, is the current pollutant distribution location, It is the fluid velocity field. On this basis, the spectral characteristics of pollutants are combined with environmental DNA analysis, and the pollution feature library and regional biological database are matched to further verify the type and specific source of pollution sources.

[0152] Finally, based on the pollution source tracing results and pollution diffusion prediction, pollution control optimization suggestions are generated, and the allocation of governance resources is adjusted according to the high, medium and low risk area divisions. Suitable governance measures are recommended based on the pollutant types, including: mechanical cleaning, chemical neutralization or ecological restoration. The prediction model is dynamically updated, the decision-making strategy is optimized, and the pollution distribution, diffusion path and governance effect are displayed through a three-dimensional visualization platform, providing comprehensive support for subsequent pollution monitoring and governance.

[0153] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for tracing and monitoring marine pollution based on satellite remote sensing and drones, characterized in that: The steps include: S100, obtain range pollution data, precise pollution data and deep water data; S200, preprocessing and multi-modal data fusion based on the obtained range pollution data, precise pollution data and deep water data; S300, based on the range pollution data, precise pollution data and deep-water data after multimodal data fusion, configure the pollution diffusion path prediction model and the data-driven model to obtain the pollution prediction results and pollution error data, perform hybrid-driven fusion based on the pollution prediction results and pollution error data, and track the initial source of the pollutants to obtain the pollution source tracing data; S400, generating three-dimensional volume data based on the obtained pollution source tracing data and the pollution prediction results and pollution error data after hybrid drive fusion, and obtaining a three-dimensional visual model through time series processing, and marking the location of the pollution source through the three-dimensional visual model based on the obtained pollution source tracing data; S500: Preset classification thresholds and risk index calculations are performed based on the pollution prediction results and pollution error data after hybrid drive fusion to obtain risk index data and potential impact area data, and decision presets and decision preset optimizations are performed based on the obtained risk index data and potential impact area data, and visualization processing is performed and report content is generated; S600, obtaining the latest risk index data, comparing the historical risk index data with the latest risk index data to obtain the comparison result, analyzing the current execution decision preset to obtain the analysis result, and generating the final report result based on the obtained report content, report results, comparison results and analysis results.

2. The method for tracing and monitoring marine pollution based on satellite remote sensing and unmanned aerial vehicles according to claim 1, characterized in that: Obtain range pollution data, precision pollution data and deepwater data, including: S100.

1. Obtain range pollution data, precision pollution data and deepwater data; Through satellite remote sensing, we can monitor the pollutants on the ocean surface over a large area and obtain the pollution data, which include: the distribution range of pollutants, the concentration of pollutants, the type of pollutants, the location of potential pollution sources, the speed and direction of ocean currents, the surface water temperature and the surface salinity; Use drones to conduct high-precision monitoring of hot spots and areas that are difficult for satellites to cover, and obtain accurate pollution data, including: detailed distribution of pollutants, pollutant concentrations, pollutant characteristics, specific types of pollution sources, biological pollution sources, local water temperature, local water flow speed, and local water transparency; The deepwater data are obtained by collecting dissolved organic matter, heavy metal content and pH value of columnar water through buoy monitoring. The deepwater data include: deep pollutant concentration, vertical distribution of pollutants, type of deepwater pollution source, intensity of pollution source, deep water temperature, deep salinity, deep water flow speed and direction and water density.

3. The method for tracing and monitoring marine pollution based on satellite remote sensing and unmanned aerial vehicles according to claim 1, characterized in that: Preprocessing and multimodal data fusion are performed based on the obtained range pollution data, precise pollution data and deepwater data, specifically: S200.

1. Preprocessing based on the obtained range pollution data, precise pollution data and deep water data; Preprocessing includes: image denoising, numerical data cleaning, and spectral data normalization; Image denoising uses convolutional neural networks to denoise range pollution data, identify and repair noise areas, and perform geometric distortion correction on precise pollution data. Numerical data cleaning uses an RNN-based time series anomaly detection model to identify and remove sudden error data points. Spectral data standardization uses normalization to uniformly adjust spectral data collected by multiple sensors. S200.

2. Multimodal data fusion based on pre-processed range pollution data, precise pollution data and deep water data; Multimodal data fusion includes: spatiotemporal alignment, feature correlation enhancement and weighted fusion; Spatiotemporal alignment is based on a dynamic time warping algorithm to align time series of multi-source data, use a geospatial interpolation algorithm to complete the data in uncovered areas, enhance feature correlation, and map spectral data, thermal infrared data, and deepwater physical parameters to a unified feature space through a multimodal feature map. A graph neural network is used to establish a correlation model between data to improve the accuracy and robustness of data fusion. Weighted fusion combines spatiotemporal alignment with enhanced feature correlation to dynamically assign weights to range pollution data, precision pollution data, and deepwater data quality.

4. The method for tracing and monitoring marine pollution based on satellite remote sensing and unmanned aerial vehicles according to claim 1, characterized in that: Based on the range pollution data, precise pollution data and deep-water data after multimodal data fusion, the pollution diffusion path prediction model and the data-driven model are configured to obtain the pollution prediction results and pollution error data. Based on the pollution prediction results and pollution error data, hybrid drive fusion is performed, and the initial source of pollutants is tracked to obtain pollution source tracing data, specifically: S300.

1. Configure a pollution diffusion path prediction model based on range pollution data, precise pollution data and deep water data after multimodal data fusion; The finite volume method is used to discretize partial differential equations, and the solution is accelerated by parallel computing to obtain pollution prediction results; S300.

2. Configure data-driven model based on range pollution data, precision pollution data and deepwater data after multimodal data fusion; Based on deep learning, a physically enhanced neural network is used, the diffusion conservation law is added in the training process, and the contamination error data is obtained by combining the physical error and the data error through the loss function; S300.

3. Perform hybrid drive fusion based on the obtained pollution prediction results and pollution error data, specifically: Where: is the final fused pollutant concentration prediction value, is the pollutant concentration predicted by the physical driving model, is the pollutant concentration predicted by the data-driven model and are the weight factors of the physics-driven model and the data-driven model respectively; S300.

4. Based on the pollution prediction results and pollution error data after hybrid drive fusion, the initial source of pollutants is tracked through reverse diffusion simulation; Extract the spectral features of range pollution data and precise pollution data, match them with the known pollutant feature library, extract the water sample DNA information in the precise pollution data, compare it with the biological database of polluted areas, identify possible biological pollution sources, and obtain pollution source tracing data.

5. The method for tracing and monitoring marine pollution based on satellite remote sensing and unmanned aerial vehicles according to claim 1, characterized in that: Based on the obtained pollution source tracing data and the pollution prediction results and pollution error data after hybrid drive fusion, three-dimensional volume data is generated, and a three-dimensional visual model is obtained through time series processing. Based on the obtained pollution source tracing data, the location of the pollution source is marked through the three-dimensional visual model, specifically: S400.

1. Generate three-dimensional volume data based on the obtained pollution source tracing data and the pollution prediction results and pollution error data after hybrid drive fusion; The three-dimensional volume data is based on the grid unit of 10m×10m×1m, and the three-dimensional grid is obtained. The pollutant concentration values ​​are directly mapped to the three-dimensional grid, and the concentration range is represented by a color gradient; Color gradients include: blue, yellow, and red; Arrange the blue below, the yellow in the middle, and the red above, and generate a frame-by-frame animation of pollutant diffusion through time series, combined with the fluid velocity field , superimposing vector arrows on the three-dimensional grid, intuitively showing the diffusion direction, and obtaining a three-dimensional visual model; S400.

2. Based on the obtained pollution source tracing data, the pollution source locations are marked through a three-dimensional visual model to obtain the pollution source marking results; The annotations include: location of pollution source, intensity of pollution source and type of pollution source; When the user clicks on the pollution source label, detailed information is provided, including: spectral characteristics, environmental DNA and pollution prediction results.

6. The method for tracing and monitoring marine pollution based on satellite remote sensing and unmanned aerial vehicles according to claim 1, characterized in that: Based on the pollution prediction results and pollution error data after hybrid drive fusion, preset classification thresholds and risk index calculations are performed to obtain risk index data and potential impact area data. Based on the obtained risk index data and potential impact area data, decision presets and decision preset optimization are performed, and visualization processing and report content are generated, including: S500.

1. Preset classification thresholds based on the pollution prediction results and pollution error data after hybrid drive fusion; Classification thresholds include: low pollution area, medium pollution area and high pollution area; Highly polluted areas represent a serious threat to the ecosystem, medium polluted areas require monitoring, and low polluted areas are basically harmless; Based on the three-dimensional visual model with the pollution source location marked, the pollution area is divided by preset classification thresholds to generate a pollution distribution zoning map, and the pollutant types are marked and classified in combination with the pollution source tracing data; S500.

2. Calculate the risk index based on the pollution prediction results after the preset classification threshold and the pollution error data to obtain the risk index data, specifically: Where: It is a risk index that quantifies the threat posed by pollutants to the environment. is the pollutant concentration, is the environmental sensitivity factor, is the intensity of pollution sources; priority areas are set based on the obtained risk index data; Based on the obtained risk index data and the pollution prediction results and pollution error data after the preset classification threshold, the pollution diffusion path prediction model is used to predict the pollution diffusion path in the next N hours and obtain the potential impact area data; S500.

3. Make decision presets based on the obtained risk index data and potential impact area data; In high-priority areas, adsorption equipment will be deployed to treat oil film and garbage salvage equipment. In medium-priority areas, chemical neutralization agents will be used to decompose pollutants. In low-priority areas, restoration bacteria will be deployed to promote ecological recovery. Assign decision presets to cleanup equipment and monitoring personnel based on priority maps; S500.4, using gradient boosting tree to optimize decision presets; The gradient boosting tree is trained using the historical risk index data and pollutant concentrations in the database as the training set, the model accuracy is evaluated using the cross-validation method, and the hyperparameters of the gradient boosting tree are adjusted through grid search; S500.

5. Perform visualization based on the obtained risk index data and decision preset results; Draw high priority areas in red, medium priority areas in yellow, and low priority areas in green; Dynamically display the impact of different decision-making preset results on pollutant concentrations and generate report content; The report content includes: risk index data, priority map, 3D visual model, pollution source marking results and decision preset results.

7. The method for tracing and monitoring marine pollution based on satellite remote sensing and unmanned aerial vehicles according to claim 1, characterized in that: Obtain the latest risk index data, compare the historical risk index data with the latest risk index data, obtain the comparison results, analyze the current execution decision presets, obtain the analysis results, and generate the final report results based on the obtained report content, report results, comparison results and analysis results, including: S600.

1. Obtain the latest risk index data; The risk index data is listed as historical risk index data, and the historical risk index data is compared with the latest risk index data, specifically: Where: Indicates the change in risk index, , it means the risk is increased, , it means the risk is reduced. , it means the risk remains stable. Indicates the change in pollution concentration, , it means that the pollutant concentration has increased and the control measures need to be strengthened. , it means that the pollutant concentration is reduced and the current treatment is effective; according to and The change value of is used to classify the governance effect into three categories: high efficiency, medium efficiency and low efficiency, and generate report results, which include: concentration reduction rate, diffusion range control effect and risk reduction range; S600.

2. Analyze the current execution decision preset; The analysis includes: resource input and governance coverage; like or When When the number of monitoring points increases, the diffusion path is closely tracked. When the current strategy is initially effective, the current strength should be maintained; S600.

3. Generate the final report results based on the obtained report content, report results, comparison results of historical risk index data and latest risk index data and decision presets.

Images