Multi-source sea area data intelligent supervision method, system and device and storage medium

Abstract

The invention relates to a multi-source sea area data intelligent supervision method, system and device and a storage medium, and the method comprises the steps: obtaining sea area geographic space data and historical sea area supervision data of a target sea area, and carrying out the monitoring construction, and obtaining an island monitoring grid; acquiring multi-source monitoring data of the target sea area according to the island monitoring grid, and performing grid data identification to obtain a global pattern spot dynamic data set; extracting a suspected specific target area coordinate of the global pattern spot dynamic data set and calculating a coordinate specific target confidence coefficient; and carrying out range comparison on the suspected specific target area coordinates, and carrying out behavior evaluation in combination with the coordinate specific target confidence to obtain a sea area specific target early warning report.

Description

Technical Field

[0001] The present invention relates to the technical field of sea area supervision, and particularly relates to a multi-source sea area data intelligent supervision method, system, device and storage medium. Background Art

[0002] With the rapid development of multi-source perception technologies such as satellite remote sensing and UAV aerial photography, the multi-source heterogeneous sea area geospatial data has shown explosive growth, providing a data basis for intelligent supervision. However, existing technologies mostly use static threshold judgment, which is prone to misjudgment and missed reports. At the same time, existing technologies are difficult to distinguish legal construction and sea use behaviors. Currently, the international community is paying increasing attention to marine ecological protection. How to achieve accurate identification and early warning of specific target behaviors through multi-source data fusion has become a key issue in the field of marine resource management. Summary of the Invention

[0003] The main purpose of the present invention is to provide a multi-source sea area data intelligent supervision method, system, device and storage medium, which can effectively distinguish legal construction and specific target behaviors, reduce supervision misjudgment, and improve the accuracy of aerial photography decision-making.

[0004] To achieve the above purpose, the present invention provides a multi-source sea area data intelligent supervision method, including: Obtain the sea area geospatial data and historical sea area supervision data of the target sea area, and perform monitoring construction to obtain an island monitoring grid; Obtain the multi-source monitoring data of the target sea area according to the island monitoring grid, and perform grid data identification to obtain a global patch dynamic data set; Extract the coordinates of the suspected specific target area in the global patch dynamic data set and calculate the coordinate specific target confidence; Perform range comparison on the coordinates of the suspected specific target area, and combine the coordinate specific target confidence to perform behavior evaluation to obtain a sea area specific target early warning report.

[0005] Further, the step of obtaining the sea area geospatial data and historical sea area supervision data of the target sea area, and performing monitoring construction to obtain an island monitoring grid includes: Based on the target sea area, obtain the sea area boundary data and geographic information, and perform spatial construction to obtain the sea area geospatial data; Obtain the marine aerial photography records and manual inspection reports of the target sea area, perform sea use behavior analysis to obtain the historical sea area supervision data; Perform spatial overlay on the sea area geospatial data according to the historical sea area supervision data, and combine the preset grid division threshold to perform initial grid setting to obtain an initial monitoring grid unit; Iteratively optimize the initial monitoring grid cells according to the resource deployment constraints of the target sea area to obtain the island monitoring grid.

[0006] Furthermore, obtain multi-source monitoring data of the target sea area based on the island monitoring grid, and perform grid data identification to obtain a global patch dynamic data set, including: Plan multi-source monitoring points for the island monitoring grid to obtain the topological positions of monitoring nodes; Obtain multi-spectral remote sensing images, UAV aerial photography data, and nearshore monitoring data of the target sea area according to the topological positions of the monitoring nodes, and perform multi-source alignment and integration to obtain the multi-source monitoring data; Perform land cover classification on the multi-source monitoring data based on a preset land cover classification rule library to obtain a grid-based land cover label distribution map; Perform regional edge detection and segmentation on the grid-based land cover label distribution map to obtain independent patch boundary vector data; Perform global matching on the independent patch boundary vector data according to the historical patch database to obtain the global patch dynamic data set.

[0007] Furthermore, the obtaining of the multi-spectral remote sensing images, UAV aerial photography data, and nearshore monitoring data of the target sea area according to the topological positions of the monitoring nodes, and performing multi-source alignment and integration to obtain the multi-source monitoring data includes: Collect remote sensing images according to the topological positions of the monitoring nodes, and eliminate atmospheric and terrain errors to obtain the multi-spectral remote sensing images; Perform UAV aerial photography planning based on the topological positions of the monitoring nodes to obtain the UAV aerial photography data; Deploy nearshore monitoring nodes according to the topological positions of the monitoring nodes, collect wave monitoring data, tide level monitoring data, and water quality monitoring data, and perform nearshore data integration to obtain the nearshore monitoring data; Perform aerial triangulation solution and position registration on the UAV aerial photography data based on the topological positions of the monitoring nodes to obtain an aerial photography image data set; Perform multi-source registration on the multi-spectral remote sensing images, the aerial photography image data set, and the nearshore monitoring data to obtain the multi-source monitoring data.

[0008] Furthermore, the extraction of the coordinates of suspected specific target areas in the global patch dynamic data set and the calculation of the confidence of the coordinates for specific targets includes: Perform overlay analysis of the protected area space on the global patch dynamic data set according to the sea area ecological protection rules to obtain an initial set of specific target areas; Perform regional change detection on the initial set of specific target areas according to the global patch dynamic data set, and identify abnormal areas to obtain the outlines of suspected specific target areas; Perform spatial coordinate transformation on the outlines of the suspected specific target areas according to the marine geospatial data to obtain the coordinates of the suspected specific target areas; Calculate the probability of specific targets in the area for the coordinates of the suspected specific target areas according to the historical marine supervision data to obtain a preliminary confidence level; Calculate the specific target level for the preliminary confidence level according to the marine ecological protection rules to obtain the confidence level of specific targets for the coordinates.

[0009] Further, perform a range comparison on the coordinates of the suspected specific target areas, and combine the confidence level of specific targets for the coordinates to conduct a behavior assessment to obtain a marine specific target early warning report, including: Analyze the legal sea use space range for the coordinates of the suspected specific target areas according to the preset sea use project database to obtain an abnormal coordinate set in the permitted coverage area and a specific target coordinate set in the non-permitted area; Match the regional sea use types for the abnormal coordinate set in the permitted coverage area based on the sea use permit information in the sea use project database to obtain the determination result of sea use compliance; Screen the regional coordinates for the specific target coordinate set in the non-permitted area according to the preset specific target confidence level threshold to obtain a high-confidence specific target coordinate set; Match the specific target behavior patterns for the high-confidence specific target coordinate set according to the historical marine supervision data to obtain the level of specific target behavior types; Conduct a specific target assessment on the determination result of sea use compliance, the high-confidence specific target coordinate set, and the level of specific target behavior types to obtain the marine specific target early warning report.

[0010] Further, the analysis of the legal sea use space range for the coordinates of the suspected specific target areas according to the preset sea use project database to obtain an abnormal coordinate set in the permitted coverage area and a specific target coordinate set in the non-permitted area includes: Calculate the polygon spatial boundary for the coordinates of the suspected specific target areas according to the legal sea use range data in the sea use project database to obtain a preliminary screening coordinate set; Filter the preliminary screening coordinate set for spatial relationships and time validity to obtain an expired permit coordinate set and a valid permit coverage coordinate set; Compare the attribute of the permit activity type for the valid permit coverage coordinate set to obtain the abnormal coordinate set in the permitted coverage area; Cluster the preliminary screening coordinate set and the expired permit coordinate set according to the legal sea use scope data to obtain the specific target coordinate set of the unpermitted area.

[0011] The present invention also provides a multi-source sea area data intelligent supervision system, which is applied to the multi-source sea area data intelligent supervision method described in any one of the above, and includes: An acquisition module, which is used to acquire the sea area geospatial data and historical sea area supervision data of the target sea area, and perform monitoring construction to obtain an island monitoring grid; An analysis module, which is used to obtain the multi-source monitoring data of the target sea area according to the island monitoring grid, and perform grid data identification to obtain a global patch dynamic data set; An association module, which is used to extract the coordinates of the suspected specific target area in the global patch dynamic data set and calculate the coordinate specific target confidence; A processing module, which is used to compare the range of the coordinates of the suspected specific target area, and combine the coordinate specific target confidence to perform behavior evaluation to obtain a sea area specific target early warning report.

[0012] The present invention also provides a multi-source sea area data intelligent supervision device, including: A memory, which is used to store programs; A processor, which is used to execute the program to implement each step of the multi-source sea area data intelligent supervision method described in any one of the above.

[0013] The present invention also provides a storage medium, which stores computer instructions, and the computer instructions are used to make a computer execute the method described in any one of the above.

[0014] A multi-source sea area data intelligent supervision method, system, device and storage medium provided by the present invention have the following beneficial effects: By constructing an island monitoring grid and integrating multi-source monitoring data, the dynamic perception of the global patch changes of sea areas and islands is realized, overcoming the lag problem of traditional single remote sensing or manual inspection, and significantly improving the early detection ability of behaviors. Calculating the confidence of the suspected specific target area avoids the misjudgment and missed reporting problems of static threshold judgment, making the identification of specific target behaviors more scientific and reliable. By combining the sea use project database with real-time monitoring data for range comparison, legal construction and specific target behaviors can be effectively distinguished, reducing supervision misjudgment and improving the accuracy of aerial photography decision-making. By comprehensively analyzing the specific target confidence and behavior evaluation results, a precise sea area specific target early warning report is generated, providing a data-driven decision-making basis for the supervision department and improving the supervision response speed and governance efficiency. Description of the Drawings

[0015] Figure 1 It is a flowchart of an intelligent supervision method for multi-source sea area data provided by the present invention; Figure 2 It is a structural diagram of an intelligent supervision system for multi-source sea area data provided by the present invention; Figure 3 It is a structural diagram of an intelligent supervision device for multi-source sea area data provided by the present invention.

[0016] The realization of the purpose, functional characteristics and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0017] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, and are not used to limit the present invention.

[0018] Next, with reference to the accompanying drawings and specific embodiments, the present invention will be further described.

[0019] Referring to Figure 1 As shown, the present invention provides an intelligent supervision method for multi-source sea area data, including: Step S101: Obtain the sea area geographical space data and historical sea area supervision data of the target sea area, and perform monitoring construction to obtain an island monitoring grid; Step S201: Obtain the multi-source monitoring data of the target sea area according to the island monitoring grid, and perform grid data identification to obtain a global patch dynamic data set; Step S301: Extract the coordinates of the suspected specific target area in the global patch dynamic data set and calculate the coordinate specific target confidence; Step S401: Compare the range of the coordinates of the suspected specific target area, and combine the coordinate specific target confidence to perform behavior evaluation to obtain a sea area specific target early warning report.

[0020] Based on the above steps, the detailed step process is as follows: Step S101: Geospatial data includes vector or raster data such as island topography, shoreline changes, and current marine use status, which are usually sourced from satellite remote sensing, UAV aerial photography, or marine survey results. Historical marine supervision data is extracted from aerial photography records, supervision information, or manual inspection reports, annotating the spatial locations and timestamps of events such as specific target sea uses and illegal reclamation. The construction of the monitoring grid uses geospatial grid technology to divide the target sea area into regular or irregular grid cells, with each grid associated with geographical attributes, ecological sensitivity indicators, and historical specific target probabilities. Grid division needs to consider the balance between data resolution and computational efficiency, for example, using quadtrees or hexagonal grids to adapt to different scale analyses. Through spatial overlay analysis, historical hotspots are associated with grid attributes to form an island monitoring grid with risk identification, providing a spatial reference for subsequent dynamic monitoring. The grid data output by this step implies the regional risk level, directly affecting the acquisition priority and recognition accuracy of the dynamic dataset of the entire region's patches.

[0021] Step S201: Based on the spatial framework of the island monitoring grid, this step focuses on the real-time acquisition and dynamic identification of multi-source monitoring data. Multi-source monitoring data covers heterogeneous sources such as satellite remote sensing images, ship AIS trajectories, offshore radar signals, and buoy sensor data. Through spatio-temporal alignment and grid mapping, the raw data is transformed into feature vectors within grid cells. For example, satellite images extract island shoreline change patches through semantic segmentation, AIS data analyzes abnormal ship berthing coordinates, and radar signals detect unreported offshore structures. Grid data identification uses deep learning models (such as U-Net or YOLOv5) to automatically classify patches, distinguish legal sea use activities from potential specific targets, and output a dynamic dataset containing spatio-temporal coordinates, type labels, and change intensities. The dynamics of the entire region's patches are reflected in the differential detection in the time series, such as comparing pre- and post-period images to identify newly reclaimed areas. This step depends on the rationality of grid division. If the monitoring grid is too coarse-grained, small-scale specific target patches may be missed. The generated dynamic dataset provides a preliminary screening result of specific target suspects for calculating the confidence level of subsequent suspected specific target areas.

[0022] Step S301: The preset rules include spatial constraint conditions such as the boundaries of ecological red lines, restrictions on sea use types (such as prohibited aquaculture areas), and the scope of the core areas of nature reserves. The patch coordinates conflicting with the rules are extracted through GIS spatial analysis. For example, the construction patches appearing within the mangrove reserve will be marked as high-suspicion targets. The confidence calculation combines multi-dimensional evidence: the overlapping ratio of the patch and the reserve, the frequency of historical similar specific targets, and the reliability of sensor data (such as the cross-validation of radar and optical images). Bayesian network or random forest models are used to quantify the probability of specific targets, and the confidence scores of specific targets for the coordinates are output. The high-score areas are related to illegal reclamation or illegal sand mining in the sea, while the low-score areas need further manual verification. The accuracy of this step is affected by the quality of the dynamic data set. If there is noise in the data (such as misjudgment caused by cloud cover), the confidence needs to be corrected by backtracking the time-series data. The output high-confidence coordinate set becomes the direct input of the early warning report for specific targets in the sea area.

[0023] Step S401: The sea use project database stores the approved coordinates, uses, and validity periods of legal sea use projects. Whether the suspected coordinates fall within the approved scope is verified through spatial join analysis (such as overlay analysis or buffer analysis). The unmatched coordinates are processed according to their confidence levels: the high-confidence coordinates directly trigger a red alert, and the medium- and low-confidence coordinates are associated with ship registration information or enterprise credit records for secondary filtering. The time dimension is introduced into the behavior assessment. For example, the association between the short-term and high-frequency AIS signals and the patch changes implies poaching behavior. The early warning report includes specific target types (such as illegal aquaculture beyond the boundary), spatial locations, confidence levels, and optimization suggestions, and is automatically pushed to the aerial photography terminal. This step depends on confidence calibration. If the threshold is set too high, it will lead to missed alarms, and the update frequency of dynamic data determines the timeliness of the early warning. After the report is generated, it is fed back to the monitoring grid to optimize the historical distribution data, forming a closed-loop management.

[0024] A multi-source sea area data intelligent supervision method provided by the present invention realizes the dynamic perception of the whole-region patch changes of sea areas and islands by constructing an island monitoring grid and integrating multi-source monitoring data, overcomes the lag problem of traditional single remote sensing or manual inspection, and significantly improves the early detection ability of behaviors. The confidence calculation for the suspected specific target areas avoids the misjudgment and missed alarm problems of static threshold judgment, making the identification of specific target behaviors more scientific and reliable. By comparing the scope by combining the sea use project database and real-time monitoring data, legal construction and specific target behaviors can be effectively distinguished, reducing supervision misjudgment and improving the accuracy of aerial photography decision-making. By comprehensively analyzing the confidence of specific targets and the results of behavior assessment, a precise early warning report for specific targets in the sea area is generated, providing a data-driven decision-making basis for the supervision department and improving the supervision response speed and governance efficiency.

[0025] In one embodiment, the marine geographical spatial data and historical marine supervision data of the target sea area are obtained and monitored to construct an island monitoring grid, including: The sea area boundary data and geographical information are the basic elements for constructing the geographical spatial data. The sea area boundary data is sourced from official nautical charts, remote sensing images, or electronic navigation charts provided by the marine surveying and mapping department, and includes vector data such as coastline, island contour, territorial sea baseline, etc. Geographical information covers dynamic hydrological data such as seabed topography, water depth, tides, ocean currents, as well as fixed feature information such as reefs, channels, and anchorages. These data are spatially registered and coordinate system unified through a Geographic Information System (GIS) platform to ensure that data from different sources are aligned under the same projection coordinate system.

[0026] The topological relationship of the original marine vector data is verified, and after eliminating geometric errors, a continuous marine surface layer and discrete island point layers are generated. The water depth sampling point data is converted into a regular grid surface, and a digital elevation three-dimensional model of water depth is established. The marine surface layer and the digital elevation model of water depth are spatially superimposed, and after integrating the island point layers, three-dimensional terrain model data is formed; the generated three-dimensional terrain model data is stored in layers according to the boundary layer, terrain layer, and feature layer to construct a structured database supporting spatial analysis.

[0027] Marine aerial records are from the case database, including time, location, type (such as illegal fishing, waste dumping, cross-border navigation, etc.) and penalty results. Manual inspection reports are filled in by island residents or patrol boats, recording non-filing events such as suspicious activities and equipment damage. The two types of data are subject to spatio-temporal matching and standardized cleaning to eliminate duplicate or invalid records. Behavioral analysis uses the kernel density estimation method to convert discrete case points into a continuous spatial probability distribution surface to identify high-frequency hot spots. For different types of behaviors, their spatial autocorrelation is calculated respectively to verify whether the events show clustering characteristics. The historical marine supervision data is finally output as a raster layer, and each pixel value represents the risk intensity at that location, while retaining the type classification attribute, providing a risk weight basis for subsequent grid division.

[0028] The spatial superposition operation is achieved through a weighted superposition tool, and algebraic operations are performed on the distribution raster and layers such as boundaries and topography in the geographical spatial data. Higher weights are assigned to high-risk areas, and after superimposing with dangerous sea areas where the water depth exceeds the warning value, a comprehensive risk score layer is generated. The grid division threshold is set according to the requirements of supervision accuracy, such as a fixed size of 1km×1km, or a variable size dynamically adjusted according to the sea area area. The initial monitoring grid cells are generated using the regular quadrilateral triangulation method to ensure full coverage and no overlap. Each grid cell inherits the underlying geographical attributes (such as average water depth, main feature type) and risk level labels to form a spatial management unit with multi-dimensional attributes. The grid boundary is topologically consistent with the marine administrative jurisdiction boundary to avoid cross-jurisdiction segmentation.

[0029] Resource deployment constraints include physical limitations such as the voyage radius of patrol boats, the endurance time of drones, and the coverage range of monitoring equipment. The graph iteration optimization takes the initial monitoring grid cells as nodes to construct an adjacency graph, and adjusts cyclically through the following steps: calculate the supervision demand index of each grid (comprehensive area, risk level, accessibility); evaluate the actual coverage ability of existing resources for this grid; perform merging or splitting operations on grids where the supervision gap exceeds the threshold. The merging operation preferentially selects grids with similar risk levels and spatial proximity, and the splitting operation divides large grids with high demand according to terrain features. The iteration termination condition is that the resource supply-demand ratio of all grids reaches the balance threshold, and the finally output island monitoring grid not only meets the hot spot and key monitoring requirements, but also conforms to the actual equipment and manpower deployment capabilities, forming an operable supervision unit system.

[0030] In this embodiment, by integrating geospatial data and historical sea area supervision data, an island monitoring grid with dynamic optimization capabilities is constructed. Based on the spatial construction of the sea area boundary and geographical information, the accurate coverage of the supervision scope is ensured, and the arbitrariness of traditional manual demarcation is avoided. Through the in-depth analysis of ocean aerial photography records and manual inspection reports, the scientific identification of hot spots is realized, and the targeting of supervision resources is significantly improved. The combination of spatial superposition and grid division threshold enables the initial monitoring grid to not only reflect the natural characteristics of the sea area but also match the risk distribution, solving the problem that static grids cannot adapt to dynamic supervision requirements. The graph iteration optimization process incorporates resource deployment constraints into grid adjustment, enabling the finally generated island monitoring grid to take into account the actual operation capabilities of equipment such as patrol boats and drones while meeting supervision requirements, greatly improving the supervision efficiency and resource utilization rate.

[0031] In one embodiment, multi-source monitoring data of the target sea area is obtained based on the island monitoring grid, and grid data identification is performed to obtain a global patch dynamic data set, including: The planning of the island monitoring grid is based on the principle of geospatial grid division. The target sea area is divided into regular or irregular geographical unit grids, and the grid size is jointly determined by the island area, the complexity of the coastline, and the requirements for supervision accuracy. The grid division uses the spatial grid generation tool of Geographic Information System (GIS). After inputting the vector boundary data of the island, the basic grid layer is generated according to the equal area principle or the equal longitude and latitude interval. The planning of the topological positions of the monitoring nodes needs to meet the spatial coverage constraints of multi-source data acquisition devices: satellite remote sensing monitoring requires no cloud cover in the grid and conforms to the satellite transit time window; UAV monitoring needs to ensure that the route planning complies with air traffic control requirements; the deployment of nearshore monitoring devices needs to avoid tidal erosion areas. The spatial distribution of the monitoring nodes follows the gradient priority strategy, and differential node density configurations are implemented in areas with intensive human activities, ecological protection areas, and historical specific target areas. For grids with terrain undulations, three-dimensional terrain modeling technology is used to calculate the visible range of the devices, and by adjusting the node coordinates, the coverage area of the monitoring devices forms a continuous and blind area-free monitoring network within the grid. The final output of the topological positions of the monitoring nodes is a metadata file containing device types, spatial coordinates, coverage radius, and acquisition frequency, providing a spatial index basis for subsequent multi-source data acquisition.

[0032] Multi-source data acquisition relies on the spatial index of the topological positions of the monitoring nodes to trigger the corresponding data acquisition process. Satellite remote sensing data is obtained by the ground receiving station calling archived data or issuing programming tasks to obtain multi-spectral images of the specified grid. The spatial resolution of the images is not less than 10 meters, and the wavelength range covers the visible light to near-infrared spectral bands. UAV aerial photography data acquisition uses an automatic route planning module to generate a terrain-following flight route based on the coordinates of the monitoring nodes, carrying a high-resolution optical camera and a lidar device to obtain orthophotos and three-dimensional point cloud data with centimeter-level ground resolution. Nearshore monitoring data is collected in real-time through high-definition cameras and hydrological sensors deployed on fixed facilities such as docks and watchtowers, including video streams and physical and chemical parameters such as water temperature and salinity. Multi-source alignment and integration include two core links: spatial reference unification and time synchronization. Spatial reference unification uses the control point registration method, selecting permanent ground feature points on the island as the reference, and uniformly transforming satellite images, UAV data, and nearshore video frames into the same plane coordinate system. Time synchronization normalizes the time tags of multi-source data through the Global Navigation Satellite System timing module of the data acquisition device, eliminating time deviations caused by device response delays or transmission lags. The integrated multi-source monitoring data is stored as a spatio-temporally consistent raster-vector hybrid dataset, where satellite images provide large-scale spectral information, UAV data supplements detailed textures, and nearshore monitoring data embeds dynamic change features, forming a multi-dimensional data cube.

[0033] The feature classification rule library consists of three parts: the spectral feature library, the texture feature library, and the morphological rules of typical features in the sea area and islands. The spectral feature library defines the reflectance threshold ranges of each band in the multispectral image for features such as vegetation, water bodies, beaches, and artificial buildings. The texture feature library extracts parameters such as contrast and entropy based on the gray-level co-occurrence matrix of UAV data to distinguish features with similar surface materials. The morphological rules set the area thresholds, shape indices, and spatial distribution patterns of different features. The classification process adopts a hierarchical decision-making mechanism: first, pixel-level classification is performed on satellite multispectral images, and the normalized difference water index and the normalized difference vegetation index are used to separate water areas and vegetation areas; subsequently, the high-resolution texture features of UAV aerial photography data are introduced, and details such as building edges and reclamation project traces are identified through the sliding window analysis method; nearshore monitoring data is used to assist in discriminating dynamic change areas in the intertidal zone through HSV color space analysis of video key frames. During the classification process, spatial context verification is implemented. When the classification results of adjacent pixels conflict, logical correction is performed based on the spatial distribution law of features (such as buildings being discontinuously distributed in deep water areas). The classification results are spatially aggregated according to the original monitoring grid to generate a statistical table of the proportion of each feature type and a spatial distribution raster map within each grid, and the classification confidence of each pixel is recorded at the same time. The grid-based feature label distribution map is stored in the GeoTIFF format, and the attribute table contains feature codes, area statistics, and data source identifiers, providing semantic input for subsequent polygon segmentation.

[0034] The feature classification results of the grid-based feature label distribution map contain discrete pixel-level label information, and independent polygons with geographical significance need to be formed through spatial aggregation and boundary extraction. Region edge detection and segmentation are based on the spatial continuity characteristics of feature categories, and a method combining morphological filtering and edge tracking is used to eliminate classification noise and extract accurate boundaries. A closing operation is performed on each feature category layer, and a circular structuring element is used to fill small holes, smooth irregular edges, and enhance the spatial connectivity of pixels of the same category. Edge detection adopts a multi-scale gradient detection strategy: for large-area continuous features such as vegetation and water bodies, an operator is used to extract weak edges, and false boundaries are filtered in combination with the feature area threshold; for complex detail areas such as buildings and reclamation projects, the Laplacian of Gaussian operator is used to enhance the response of corner and linear features. The detected edge lines are converted into continuous polygon contours, and topological repair is performed on the contour lines with intersections or breaks to ensure the geometric integrity of each closed polygon.

[0035] In the vector data generation stage, the contour polygons are converted into vector boundaries. While maintaining the shape features, the number of redundant vertices is reduced, and the simplification tolerance is set to 0.5 times the image spatial resolution. Topological consistency verification is performed on the simplified vector boundaries to eliminate boundary overlaps or gaps between adjacent patches. The attribution relationship between patches and the original grid is established through spatial indexing. The vector data of the independent patch boundaries records the identifier, feature type, area, and the grid number to which each patch belongs in the layer attribute table. The output format uses the standardized Shapefile or GeoPackage, supporting spatial query and overlay analysis. The boundary accuracy of this step is directly affected by the resolution of the feature classification result and the edge detection parameters. The areas with a classification confidence lower than 85% trigger the manual review mechanism during segmentation.

[0036] The historical patch database stores the geometric boundaries, feature attributes, and change event records of all patches in previous monitoring cycles in a spatio-temporal cube structure. Global matching identifies the evolution relationship between the current patch and the historical data through spatio-temporal sequence comparison and spatial relationship reasoning. The matching process is divided into two stages: spatial overlay and attribute association. In the spatial overlay stage, the vector data of the current patch boundary is layer-by-layer overlaid and analyzed with the patches in the same geographic grid in the historical database to calculate the spatial intersection area, overlap rate, and shape similarity index between the two. In the attribute association stage, the feature types, area change rate, and spatial distribution patterns of the current patch and the historical patch are compared, and the patch status is determined by combining the preset change thresholds.

[0037] For patches with a spatial intersection area ratio exceeding 90% and the same feature type, they are marked as "stable patches"; for patches with an intersection area ratio lower than 30% or a change in feature type, they are marked as "new patches" or "disappeared patches"; for patches with an intersection area between 30% - 90%, their morphological change trajectories are further analyzed. If the area continuously increases and conforms to the characteristics of reclamation projects, they are marked as "expanding patches". The construction of the dynamic dataset adopts an incremental update mechanism. The spatio-temporal fingerprint features of the newly added or changed patches are extracted, including the center point coordinates, change time window, associated monitoring device type, and original data traceability information, and similarity matching is performed with the historical specific target case library. The matching results are stored as the global patch dynamic data. Each patch record contains the change type, confidence score, and associated historical event number, supporting the playback of the feature evolution process along the time axis.

[0038] In this embodiment, through the multi-source monitoring point planning of the island monitoring grid and the collaborative deployment of satellites, unmanned aerial vehicles (UAVs) and inshore devices, full-coverage three-dimensional monitoring of the target sea area is achieved, effectively improving the integrity of monitoring data in complex terrains and dynamically changing areas. Based on the multi-source alignment and integration technology, the spatio-temporal reference differences of heterogeneous data are eliminated, ensuring the multi-dimensional data fusion accuracy of spectral, texture and dynamic features, and providing a highly consistent data basis for ground object classification and evolution analysis. Relying on the grid-based ground object label distribution map and the vector boundary data generated by edge detection and segmentation, high-precision spatial expression of natural landforms and artificial buildings is realized, supporting millimeter-level boundary recognition of the dynamic changes of map patches. Through the global matching mechanism of the historical map patch database and the current data, a spatio-temporal associated traceability chain of specific target behaviors is constructed, significantly enhancing the intelligent discovery and early warning capabilities for problems such as reclamation and ecological damage, and providing a full-cycle closed-loop decision-making basis for the supervision of sea areas and islands.

[0039] In one embodiment, multi-spectral remote sensing images, UAV aerial photography data and inshore monitoring data of the target sea area are obtained according to the topological positions of the monitoring nodes, and multi-source alignment and integration are performed to obtain multi-source monitoring data, including: For remote sensing image acquisition, based on the deployment coordinates and time window parameters of the satellite equipment in the topological positions of the monitoring nodes, the satellite task scheduling system is triggered to obtain the original multi-spectral data of the target grid. After the completion of the original data acquisition, the historical remote sensing spectral database is called to extract the ground object spectral reflectance curves archived under the same geographical location, similar seasons and meteorological conditions as the reference benchmark. Spectral matching is achieved through the alignment of the band response function, and the convolution operation is performed on the sensor band range of the current image and the band parameters of the historical spectral library to generate comparable equivalent spectral features.

[0040] Atmospheric error elimination adopts the method of joint inversion based on the radiative transfer model and historical spectra: after inputting real-time meteorological parameters (aerosol optical depth, water vapor content), by minimizing the differences between the current image spectrum and the historical reference spectrum on typical ground objects (such as pure water bodies, bare soil), the atmospheric correction parameters are iteratively optimized to improve the inversion accuracy of the surface reflectance. Under the support of DEM data, terrain error correction combines the reflectance attenuation laws of ground objects with different slopes and aspects in the historical spectral library, and adopts a dynamic normalization model to eliminate the terrain shadow effect. The corrected multi-spectral remote sensing image retains the spectral features of ground objects in bands such as blue, green, red, and near-infrared, and the consistency error of the spectral curve with the historical spectral database is less than 5%, providing a high-fidelity input for subsequent multi-source data fusion.

[0041] The UAV aerial photography plan takes the coordinates and coverage radius of the aerial photography equipment in the monitoring node topology position as input parameters, combines the UAV flight performance (endurance, climb rate) and camera parameters (focal length, pixel size), and generates a three-dimensional waypoint sequence. The waypoint coordinates are spatially encrypted based on the grid vertex coordinates, and the longitudinal flight strip spacing is calculated according to the lateral field of view angle of the camera to ensure that the forward overlap rate is not less than 70% and the side overlap rate is not less than 50%. During the execution stage of the aerial photography task, the UAV is equipped with a differential GNSS module and an inertial navigation system (INS), and performs terrain-following flight along the planned waypoints, and adjusts the flight altitude in real time to maintain the consistency of the ground resolution.

[0042] The camera trigger mode adopts equidistant interval shooting, and synchronously records the POS data (latitude, longitude, altitude, attitude angle) of each image. The preprocessing of the aerial photography data includes lens distortion correction and radiometric consistency processing: geometric correction of the original image based on the camera calibration parameters (radial distortion coefficient, tangential distortion coefficient); histogram matching algorithm is used to equalize the brightness and contrast of adjacent images to eliminate the influence of illumination changes. The preprocessed aerial photography data set is output as an orthophoto mosaic map with georeference and a digital surface model (DSM), with a spatial resolution better than 5 cm and a planar positioning accuracy reaching the sub-meter level.

[0043] The deployment of the nearshore monitoring nodes strictly follows the preset fixed equipment coordinates in the monitoring node topology position, and installs wave radars, pressure tide gauges and multispectral water quality meters in key areas such as the intertidal zone, reef groups, and ports. The wave monitoring data emits electromagnetic waves through a high-frequency radar and receives the sea surface echo to invert the wave height, wave direction and period parameters, with a sampling frequency not less than 2 Hz, and the spatial coverage range matches the field of view angle of the monitoring node. The tide level monitoring data is collected by an underwater pressure sensor, and the original pressure value is converted into the absolute water level elevation through atmospheric pressure compensation and temperature drift correction, and the timestamp is synchronized with the GNSS clock.

[0044] The water quality monitoring data obtains the chlorophyll a concentration, turbidity and dissolved oxygen content through an optical sensor, and is collected once every 30 minutes. The data is calibrated on-site to eliminate the interference of environmental temperature and salinity. During the data integration stage, the multi-source sensor data is aligned in time and space and spatially interpolated: time alignment uses the sliding window method to unify the time bases of each device and eliminate the time series misalignment caused by transmission delay; spatial interpolation is based on the Kriging algorithm to convert the discrete point data into a continuous raster surface covering the nearshore grid. The wave data generates an isoline map of wave height distribution, the tide level data constructs a water level elevation surface model, and the water quality data forms a spatial distribution map of parameter concentration. The integrated nearshore monitoring data is stored as a time series multi-dimensional raster data set, supporting spatial overlay analysis with remote sensing images and UAV data.

[0045] The aerial triangulation solution of UAV aerial photography data uses the coordinates of control points preset in the topological position of the monitoring nodes as geometric constraint conditions. The selection of control points includes permanent features (such as reef vertices and dock corners) or artificially arranged reflective targets. The coordinates of control points are measured on-site by GNSS-RTK equipment, with a planar accuracy better than 2 cm and an elevation accuracy better than 5 cm, and are strictly aligned with the spatial reference of the topological position of the monitoring nodes. The aerial triangulation adopts the method of block adjustment, and jointly optimizes by combining the POS data (position and orientation system data) recorded by the UAV and the automatically extracted image feature points (such as coastline inflection points and building outlines): the homologous connection points between adjacent images are generated through the scale-invariant feature matching algorithm to ensure that each image contains no less than 20 evenly distributed feature points; with the coordinates of control points as strong constraints and connection points as weak constraints, iteratively adjust the exterior orientation elements (latitude, longitude, altitude, roll angle, pitch angle, yaw angle) of the image and the three-dimensional coordinates of ground points to minimize the sum of the squared projection errors of all images. After the adjustment converges, the residual threshold is controlled within 1.5 times the ground resolution of the image, and the over-limit area triggers manual verification or supplementary control points.

[0046] In the position registration stage, the adjusted image coordinate system is transformed to the planar projection coordinate system (such as UTM projection) adopted by the topological position of the monitoring nodes through the seven-parameter Helmert transformation model to eliminate the translation, rotation, and scale deviations caused by coordinate system differences. For the residual geometric distortion, the thin plate spline function (TPS) is used to compensate for the local non-linear deformation of the image. With the control point residuals as input parameters, the planar positioning error after correction is better than 0.3 times the ground resolution of the image (for example, the error of a 5 cm resolution image is less than 1.5 cm). The aerial photography images after solution are cropped into blocks according to the monitoring grid numbers to generate orthophoto image tiles and digital surface model (DSM) tiles. Each tile is associated with a metadata file, recording the acquisition time, sensor parameters, adjustment accuracy, and residual statistical values. The dataset is stored in the GeoTIFF format, embedded with spatial reference and timestamp information, and the planar error is controlled within 0.3 times the resolution to ensure pixel-level spatial consistency with multi-spectral remote sensing images and provide input data with millimeter-level accuracy for multi-source registration.

[0047] Multi-source registration takes the spatial coordinate system of the monitoring node's topological position as the unified benchmark, and performs hierarchical processing on the spatial alignment and attribute association of heterogeneous data. In the spatial alignment stage, multi-spectral remote sensing images, aerial orthophoto image blocks, and near-shore monitoring grid data are imported into the same geospatial platform, and geometric registration is carried out using feature point matching and least squares transformation methods: for satellite and aerial images, stable feature points such as coastline and building outlines are extracted for affine transformation; for near-shore wave and water level grid data, a control point forced matching method is used to ensure that the water level contour line coincides with the water-land boundary of the remote sensing image. Time alignment is achieved through a unified timestamp, and the satellite transit time, UAV aerial photography period, and time series of near-shore monitoring data are interpolated and aligned to generate a synchronous time slice data set.

[0048] In the attribute association stage, a semantic mapping relationship between multi-source data is established: the texture features of aerial images are associated with the spectral bands of remote sensing images, the near-shore water quality parameters are associated with the remote sensing chlorophyll-a inversion results, and the water level data is associated with the inundation area extracted from the UAV DSM. The registered multi-source monitoring data is stored as a spatio-temporally consistent multi-layer grid-vector hybrid data set, which contains multi-dimensional attribute fields such as spectrum, texture, wave, water level, and water quality, and supports fast retrieval and overlay analysis by grid number or timestamp.

[0049] In this embodiment, by fusing the historical remote sensing spectral library and real-time acquisition data for joint inversion, the accuracy of atmospheric correction and topographic correction is significantly improved, effectively overcoming the spectral distortion caused by complex meteorological conditions and terrain undulations, and ensuring the authenticity and reliability of the ground object reflectance characteristics of multi-spectral remote sensing images. Based on the air-space-ground-sea multi-source data collaborative acquisition mechanism of the monitoring node's topological position, the advantages of satellite wide-area coverage, UAV high-precision detailed inspection, and near-shore dynamic monitoring are complementary to each other, forming a three-dimensional monitoring network in the full spatial dimension. The strict unification of the spatio-temporal benchmark and semantic association mapping in the multi-source registration process eliminates the spatial misalignment and attribute deviation between heterogeneous data, providing a millimeter-level consistent data base for the analysis of island ground object evolution. The combination of aerial triangulation solution and closed-loop verification mechanism reduces the UAV positioning error while reversely optimizing the monitoring node deployment strategy, forming an intelligent supervision closed-loop for self-improving data quality, and greatly enhancing the credibility and timeliness of aerial photography evidence such as reclamation monitoring and ecological damage tracing.

[0050] In one embodiment, based on the preset sea area ecological protection rules, the coordinates of suspected specific target areas are extracted from the global patch dynamic data set, and confidence calculation is performed to obtain the coordinate specific target confidence, including: The marine ecological protection rules are a series of predefined constraints used to regulate the development and utilization of marine resources, including but not limited to spatial boundaries and control requirements such as prohibited development zones, restricted development zones, and ecological red line zones. The global dynamic patch dataset is a spatiotemporal continuous data on sea and island surface cover changes acquired through multi-source remote sensing images, drone aerial photography, and ship monitoring. It is stored in the form of vector patches and contains patch boundaries, attribute labels, and timestamp information. The core of the spatial overlay analysis of protected areas is to perform a geometric intersection operation on the global dynamic patch dataset and the spatial control layer in the marine ecological protection rules to screen out patches that overlap or conflict with the boundaries of the protected areas.

[0051] Specifically, marine ecological protection rules usually exist in the form of polygon layers in geographic information systems, such as marine nature reserves, important fishing waters, or prohibited land reclamation areas. Each patch in the global patch dynamic dataset needs to be spatially related to these protected area layers. If the patch falls completely or partially within the protected area, a specific target condition is triggered. For example, if a patch shows newly reclaimed land, and the area is designated as a prohibited development area in the ecological protection rules, the patch will be marked as the initial specific target area. Spatial overlay analysis relies on the geometric calculation functions of GIS tools or spatial databases.

[0052] The generation of the initial set of specific target areas needs to be verified in conjunction with the time dimension. For example, if a certain patch already exists in historical data and has not changed, even if it is located in a protected area, it is a legal legacy project and needs to be excluded; while newly added or expanded patches need to be marked as highlights. In addition, attribute matching is also crucial. For example, if specific types of scientific research activities are allowed in a protected area, further filtering is required through patch attribute fields (such as "land use type"). The final output of the initial set of specific target areas contains the patch's unique identifier, spatial boundary, specific target type, and time information, providing a basis for subsequent change detection.

[0053] The initial set of specific target areas only reflects spatial location conflicts and does not distinguish between natural changes and human-specific target activities. The goal of regional change detection is to eliminate changes in patches caused by legitimate or natural factors through time series analysis, focusing on anomalous human activities. Global patch dynamic datasets contain multi-temporal data, such as patch sequences derived from daily, weekly, or monthly remote sensing imagery. Change detection is achieved by comparing patch attributes (such as area, shape, and cover type) within the same area at different times.

[0054] The key to abnormal area recognition lies in setting change thresholds and pattern rules. For example, if the area of a patch within a protected area expands rapidly in the short term, it implies illegal reclamation; if the shape of the patch changes from a natural coastline to a regular rectangle, it indicates the construction of artificial structures. Change detection algorithms can use simple difference methods (such as an area change rate exceeding 20%) or complex models (such as detecting mutation points based on time series), but it is necessary to avoid over-reliance on algorithms and instead set rules by combining domain knowledge. For example, if the patch coverage type in a mangrove reserve changes from "vegetation" to "bare land", it is directly marked as abnormal; if the change occurs after the typhoon season, it is necessary to combine meteorological data to determine whether it is caused by natural disasters.

[0055] The generation of the contour of the suspected specific target area requires integrating geometric and semantic information. For example, after change detection on the initial specific target patch, only the areas with high characteristics of human intervention are retained, and its contour is obtained through vector boundary refinement or raster-to-vector operations. The contour data needs to be accompanied by evidence of changes, such as screenshots of images from different time phases, statistical values of the change amount (such as the percentage of area increase or decrease), for subsequent manual review or confidence calculation. The contour of the suspected specific target area output in this step is a fused expression of spatial and attribute information, providing an accurate geometric basis for coordinate transformation.

[0056] The contour of the suspected specific target area usually exists in the form of vector polygons (such as WKT format) or raster masks, and needs to be converted into standardized spatial coordinates for positioning and visualization in the supervision system. Geospatial data includes coordinate system parameters, elevation benchmarks, and projection methods (such as Mercator projection). The purpose of coordinate transformation is to unify the reference frameworks of different data sources and ensure that the coordinates accurately correspond to the field positions.

[0057] Specifically, if the original contour data is inconsistent with the target coordinate system, projection transformation or datum conversion is required. For example, if the UAV aerial photography data uses a local independent coordinate system, while the sea area supervision requires the use of the national geodetic coordinate system, conversion needs to be carried out through a seven-parameter or four-parameter model. For raster data (such as the contour extracted from remote sensing images), it is necessary to map the pixel coordinates to geographic coordinates through a georeference file. The converted coordinates need to retain sufficient precision. For example, the boundary of a reclamation project requires sub-meter precision, while large-scale ecological monitoring can accept meter-level errors.

[0058] The output format of the coordinates of the suspected specific target area needs to be adapted to the subsequent analysis requirements. Common forms include the center point coordinates (longitude, latitude), the sequence of boundary vertices, or the coordinates of the minimum bounding rectangle. For example, for the expansion of a specific target in a circular aquaculture area, the center point and radius are output; for an irregular reclamation area, the longitude and latitude sequences of all boundary points are output. The coordinate data needs to be accompanied by metadata, such as the coordinate system declaration, accuracy evaluation value, and conversion method, to avoid misjudgment due to coordinate ambiguity in subsequent steps. The standardized output of this step provides a unified spatial reference benchmark for specific target probability calculation and confidence evaluation.

[0059] Historical sea area supervision data is a spatial dataset of specific target events in the sea area recorded in the past, usually including information such as specific target types, occurrence times, processing results, and geographical locations. These data can reflect the specific target hotspots, high-incidence types, and time patterns in a specific area, providing a statistical basis for the probability assessment of the current suspected specific target area. The goal of calculating the probability of a specific target in a region is to quantify the matching degree between the current suspected specific target area and the historical specific target pattern through spatial association and statistical analysis, so as to generate a preliminary confidence level.

[0060] In specific implementation, historical sea area supervision data is usually stored in the form of point layers or polygon layers, such as the location records of illegal reclamation, specific target fishing, or sewage discharge events. The calculation process first conducts a spatial association analysis between the coordinates of the suspected specific target area and the historical data. For example, it determines whether the current area is located in the high-incidence area of historical specific targets through buffer analysis or spatial join. If a suspected specific target area overlaps or is adjacent to illegal reclamation events that have occurred multiple times in the past five years, its specific target probability will increase significantly.

[0061] The statistical method for probability calculation can use kernel density estimation to measure the frequency of specific targets in a region, or adjust the confidence level based on the Bayesian method combined with prior probability. For example, there have been 10 illegal aquaculture events in a sea area in the past three years, and the current suspected specific target area happens to be located in the high-value area of the kernel density of this sea area, and its preliminary confidence level can be set to a relatively high value (such as 0.8). If there is no historical specific target record in this area, but it meets the typical specific target characteristics (such as the sudden appearance of a large area of artificial structures), then other factors (such as supervision intensity, ecological sensitivity) need to be combined to adjust the probability.

[0062] The output of the preliminary confidence level needs to include the probability value (such as in the range of 0 - 1) and supporting evidence. For example, the preliminary confidence level of a certain coordinate is 0.75, based on the fact that there have been 3 similar specific target behaviors in this area in the past two years, and the current change pattern highly coincides with historical cases. The result of this step provides data support for the subsequent calculation of the specific target level, but it has not yet combined the strictness of specific protection rules, so it still needs to be further corrected.

[0063] The sea area ecological protection rules not only define the spatial scope of prohibited or restricted activities, but also stipulate the control levels of different regions and the severity of specific targets. For example, any human activity in the core protection area is a serious specific target, while partial compliant development is allowed in the general control area. The purpose of calculating the specific target level is to weight and adjust the preliminary confidence level in combination with the strictness of the protection rules, and finally generate the coordinate-specific target confidence level to reflect the actual supervision priority of specific target behaviors.

[0064] Specifically, the rules for marine ecological protection usually exist in the form of hierarchical control layers. For example: Core protection area (primary control): Any development is strictly prohibited, and the confidence level of specific target behaviors is directly adjusted to the highest level (such as 1.0).

[0065] Ecological restoration area (secondary control): Restricted development. If the initial confidence level is 0.6, it is increased to 0.8 in combination with the rules.

[0066] Moderate utilization area (tertiary control): Compliance activities are allowed. If the initial confidence level is low (such as 0.3), it is reduced to 0.1.

[0067] The calculation of the specific target level can use a rule engine or a decision matrix. For example: If the area belongs to the core protection area and the initial confidence level ≥ 0.5, then the confidence level of the specific target of the coordinate = 1.0.

[0068] If the area belongs to the ecological restoration area and the initial confidence level ≥ 0.7, then the confidence level of the specific target of the coordinate = 0.9.

[0069] If the area belongs to the moderate utilization area and the initial confidence level ≤ 0.4, then the confidence level of the specific target of the coordinate = 0.2.

[0070] The finally output confidence level of the specific target of the coordinate is a comprehensive evaluation value, which takes into account both the historical probability of the specific target and the protection level of the current area. For example, a reclamation patch is in a historically high-incidence area (initial confidence level 0.8), but is located in a general control area, and the final confidence level is adjusted to 0.7; if the same patch is located in the core protection area, the confidence level is directly set to 1.0, triggering the highest-priority supervision response. This step ensures the reasonable allocation of supervision resources and gives priority to dealing with specific target behaviors with the most serious ecological impacts.

[0071] This embodiment combines protected area spatial overlay analysis and regional change detection, which can effectively distinguish natural changes from human specific target activities and greatly improve the accuracy of specific target area identification. Using historical marine supervision data for probability calculation enables the system to have learning ability, be able to automatically adjust the monitoring focus according to the specific target characteristics of the area, and significantly improve the supervision efficiency. By combining the initial confidence level with the marine protection level, the priority classification management of specific target behaviors is realized, enabling limited supervision resources to be accurately invested in areas with the highest ecological risks. The whole set of methods constructs a complete supervision closed-loop from data collection to intelligent decision-making through multi-source data fusion and spatial analysis technology, providing a scientific and reliable intelligent supervision means for marine ecological protection.

[0072] In one embodiment, the scope of the suspected specific target area coordinates is compared based on the sea use project database, and the behavior is evaluated in combination with the confidence level of the specific target of the coordinate, obtaining a marine specific target early warning report, including: The coordinates of the suspected specific target area are derived from the central point coordinates of the newly added or changed patches in the global patch dynamic dataset. The legal sea use scope data is extracted from the sea use project database, including the vector boundaries of the approved sea use blocks and their attribute information (such as sea use type, approval time, validity period). The spatial analysis is based on the spatial overlay analysis function of the Geographic Information System (GIS), and the layer of suspected specific target coordinate points is overlaid with the vector surface layer of the legal sea use scope.

[0073] For the coordinate points that completely fall within the legal sea use boundary, extract the attribute conflict information with the approved sea use type and time validity period, and generate an abnormal coordinate set for the permitted coverage area (such as the approved type is wharf construction but the actual monitoring is reclamation); for the coordinate points outside the legal sea use boundary, exclude the reasonable error range of the approved boundary extension through buffer analysis (such as the dynamic boundary fluctuation affected by tides), and the coordinate points beyond the buffer are marked as the specific target coordinate set for the non-permitted area. The attribute tables of the two types of coordinate sets record the timestamps of the coordinate points, the source data (satellite, drone or inshore equipment), and the original patch numbers, providing a traceable basis for subsequent analysis.

[0074] The sea use permit information includes constraints such as the approved sea use type (land reclamation, channel dredging, wharf construction, etc.), the permitted construction period, and the upper limit of the sea use area. The regional sea use type matching is performed for the abnormal coordinate set of the permitted coverage area, and spatial connection and attribute association are carried out: spatially associate the abnormal coordinate points with the corresponding legal sea use blocks, and extract the approved sea use type; through time validity verification, eliminate the coordinate points that exceed the approval validity period (such as the construction permit has expired); semantically match the actually monitored sea use type (judged by the ground object classification rule base) with the approved type (such as the approval is for a temporary storage yard but the monitoring is for a permanent building).

[0075] The matching results are divided into three categories: consistent type and valid time (compliant), deviated type but area not exceeding the limit (requiring manual review), and type conflict or area exceeding the limit (specific target). The judgment results are written into the coordinate point attribute table, and the approval document number and responsible entity information in the sea use project database are associated to form a complete sea use compliance judgment result.

[0076] The confidence threshold for specific targets is comprehensively set according to the reliability of the monitoring data source, the frequency of coordinate point occurrences, and the distance to the adjacent historical specific target area. The reliability weight distribution rule for the data source is: the confidence of satellite monitoring coordinates is 0.7, the confidence of drone monitoring coordinates is 0.9, and the confidence of inshore equipment coordinates is 0.8; the frequency of coordinate point occurrences is calculated through time series analysis, and the frequency weight of the coordinate points that appear continuously more than 3 times is 1.0, and the weight of single occurrences is 0.5; the distance weight to the adjacent historical specific target area adopts an exponential decay model, and the weight decreases by 0.2 for every 100 meters increase in distance.

[0077] The confidence calculation integrates three types of weights, and the calculation formula is: Confidence = Data source weight × Frequency weight × Distance weight. During the screening process, the threshold range is dynamically adjusted (usually 0.6 - 0.9), and only the coordinate points with a confidence higher than the threshold are retained. Spatial clustering analysis is performed on the screened coordinate points to eliminate isolated points (such as no other specific target points within 500 meters around a single coordinate point), and finally a specific target coordinate set with high confidence is generated. The coordinate set attribute table records the confidence score, associated data source, and clustering number, providing input for behavior pattern matching.

[0078] Historical sea area supervision data stores the spatial coordinates, behavior types (such as illegal reclamation, over-range aquaculture, illegal sand mining) and handling results of all verified specific target events within the past cycle. Specific target behavior pattern matching uses dual analysis of spatial association and behavior characteristics: The high-confidence specific target coordinate set is spatially superimposed with the distribution heat map of historical specific target events to calculate the proximity of the current coordinate point to the historical similar specific target area (such as coordinate points less than 500 meters from the nearest historical specific target point are marked as high-risk); at the same time, the land cover type of the patch where the current coordinate point is located (obtained through the dynamic dataset of the entire region patches), the change trajectory (such as the monthly increase in the reclamation area), and the type of adjacent sea use projects (such as close to ecological protection areas or waterways) are extracted to construct a specific target behavior feature vector.

[0079] Through a predefined specific target pattern rule library (such as "continuous reclamation and adjacent to the ecological area" corresponding to "ecological destruction type specific target"), pattern matching is performed on the feature vector, and the specific target type (reclamation, sewage discharge, illegal sand mining, etc.) and level (low, medium, high) are output. The level is divided based on weighted calculation of the specific target area, duration, and ecological sensitivity, and finally a structured specific target behavior type and level table containing type codes and risk levels is generated.

[0080] Specific target assessment integrates the output data of the first three stages: the compliance determination results of the abnormal coordinate set in the permitted coverage area (such as type conflict or area overrun), the high-confidence specific target coordinate set in the non-permitted area, and its specific target behavior type and level. During the assessment process, priority sorting and evidence chain integration are performed: for the type conflict coordinate points in the permitted coverage area, the sea use approval document number and responsible entity information are associated and marked as "approval specific target"; for the high-risk level coordinate points in the high-confidence specific target coordinate set, their specific target type (such as illegal sand mining) is superimposed with the aerial photography records of historical similar events and marked as "repeated specific target"; for the coordinate points adjacent to the ecological red line or sensitive area, the "ecological high-risk specific target" is output in combination with the ecological impact assessment model.

[0081] The warning reports for specific targets in the sea area are divided into three levels according to the degree of urgency: red (immediate disposal), orange (rectification within a time limit), and yellow (attention and monitoring). Each level is associated with a specific list of coordinate points, specific target types, legal bases, and optimization suggestions. The report generation uses a standardized template, embeds a dynamic map visualization module (such as a heat map of specific target points and a historical comparison layer), and attaches data traceability information (monitoring equipment type, processing timestamp, confidence score). It is pushed to the aerial photography terminal through a secure interface to support on-site verification and the retrieval of case-filing bases.

[0082] In this embodiment, through the spatial overlay analysis of the legal sea use range and the coordinates of suspected specific targets, the accurate separation of abnormal behaviors inside and outside the approved area is realized, effectively improving the positioning efficiency and spatial accuracy of specific target screening. Based on the multi-dimensional compliance judgment mechanism of sea use type and timeliness, combined with the semantic association of approval attributes and real-time monitoring data, the hidden problem of "approval compliance but actual specific targets" in traditional supervision is solved. The dynamic weighted model of the specific target confidence threshold fuses the credibility of multi-source data, spatio-temporal frequency, and historical distribution characteristics, significantly reducing the interference of isolated false alarms and enhancing the recognition robustness of high-risk specific target areas. The spatial coupling analysis of the historical specific target behavior pattern and the ecological sensitive area gives prior knowledge support to the classification of specific target types, making the warning results have both legal bases and ecological risk prediction values.

[0083] In one embodiment, spatial analysis is performed on the coordinates of the suspected specific target area according to the legal sea use range data in the sea use project database, and an abnormal coordinate set in the permitted coverage area and a specific target coordinate set in the non-permitted area are obtained, including: The coordinates of the suspected specific target area are derived from the newly added or changed patches in the global patch dynamic dataset, and are extracted through the centroid coordinates of the patches or the vertex coordinates of the largest area sub-patch. The legal sea use range data is stored in the form of a standardized vector layer, and each polygon is associated with an approval document number, a sea use type, spatial boundary coordinates, and attribute metadata. The spatial boundary calculation of the polygon adopts the spatial inclusion criterion of the geospatial engine: the spatial relationship of each suspected specific target coordinate point is determined with the legal sea use polygon layer. The topological tolerance mechanism is introduced in the determination process, and the buffer distance is set to twice the accuracy of the approval boundary coordinates (usually 5-10 meters) to eliminate the boundary blur problem caused by surveying errors or tidal dynamic changes. The ray method penetration detection is performed on each coordinate point to calculate the number of intersections with the polygon boundary. An odd number of intersections is determined as an internal point, and an even number is determined as an external point. The determination results are divided into "candidate points in the permitted coverage area" (internal points) and "candidate points in the non-permitted area" (external points), and the confidence score of the penetration detection is recorded at the same time (for example, the confidence of the points near the boundary is reduced by 30%). After the preliminary screening coordinate set is generated, it is associated with the approval document number and sea use type attributes and stored as an event log table in the spatial database. The log table contains the unique identifier of the coordinate point, the determination timestamp, the device number of the original data source, and the spatial inclusion status flag, providing a traceable intermediate data layer for subsequent refined filtering.

[0084] Spatial relationship filtering is based on the spatial metadata of the approved sea use blocks, and multi-level verification is performed on the "candidate points in the permitted coverage area". First, the candidate points are associated with the approval validity period field in the sea use project database through a spatial join operation, and the start time and end time attributes of the approval block corresponding to each point are extracted. The time validity filtering adopts a sliding time window match: if the timestamp of the coordinate point (derived from the monitoring data collection time) exceeds the approval end time by more than 3 months, it is marked as an "expired permitted coordinate point"; if it is within the 3-month window period of the approval end time, it is marked as a "pending verification point" and the manual review process is triggered. A secondary spatial fine screening is performed on the valid permitted coverage coordinates: based on the approval block boundary, a dynamic buffer is expanded outward, and the buffer radius is set according to the sea use type (for example, 5% of the approval area for reclamation projects and 10 meters for waterway projects), and the coordinate points exceeding the buffer are screened out, marked as "suspected over-border points" and associated with the over-border distance attribute. The expired permitted coordinate set and the valid permitted coverage coordinate set are stored in independent spatial layers respectively. The layer attribute table records the approval expiration date, the over-border distance, the associated aerial photography responsible person, and the review status flag, and at the same time retains the foreign key association with the event log table of the preliminary screening coordinate set to ensure the full-link data traceability.

[0085] The licensing activity type attribute comparison establishes a semantic mapping relationship table between the approved sea use types and the monitored feature types. The mapping table defines type compatibility rules (e.g., "wharf construction" allows subclasses such as "concrete structure" and "yard", but prohibits the appearance of "sand mining equipment") and area floating thresholds (e.g., the area error of earthwork is ±15%). The comparison process is divided into two stages: type matching and area verification: Type matching: Extract the monitored feature types of the valid license coverage coordinate points (from the feature classification results of the global patch dynamic dataset) and perform rule mapping with the approved sea use types. If the monitored type is not in the license compatibility list (e.g., the approval is for "dredging project" but "permanent building" is monitored), it is marked as "type conflict".

[0086] Area verification: Calculate the actual area of the monitored sea use block (the polygon area based on the patch vector boundary) and compare it with the maximum allowable area of the approved license. If the overrun ratio exceeds the floating threshold (e.g., the approved area is 100 hectares and the actual monitored area is 120 hectares), it is marked as "area overrun"; if there are both type conflicts and area overruns, it is marked as "compound specific target".

[0087] The abnormal coordinate set is output as a spatial hotspot layer. The layer attribute table records the specific target type codes (e.g., T1 for type conflict, A2 for area overrun of 20 - 30%), overrun values, associated approval document numbers, and the original monitored image IDs. This layer establishes an associated index with the approval details table in the sea use project database, supporting one-click retrieval of design drawings and responsible party information in the approval file as auxiliary evidence for specific target determination.

[0088] Non - licensed area clustering performs a joint analysis of spatial density and time series for "non - licensed area candidate points" and "expired license coordinate points". Spatial clustering uses an improved DBSCAN model, and the parameter settings follow the spatial distribution characteristics of specific target behaviors in the sea area: the neighborhood radius is adaptively calculated based on the average distribution density of historical specific target hotspot areas (usually 200 - 1000 meters), and the minimum point threshold is dynamically adjusted according to the credibility of the monitoring data source (≥5 points for satellite data source, ≥3 points for UAV data source). The time weight factor is introduced in the clustering process: coordinate points that appear in three consecutive monitoring cycles are given a weight of 2, and the weight of isolated single - time points is reduced to 0.5.

[0089] The clustering results generate the spatial boundaries of specific target clusters. The boundary generation adopts the α-shape algorithm to adapt to the complex coastline morphology and avoid the errors caused by regular convex hulls. A legality review is performed on each specific target cluster: spatially overlay it with the pre-application blocks to be approved in the sea use project database. If the overlapping area exceeds 50%, it is marked as a "cluster related to pending approval" and the warning level is reduced. Finally, the specific target coordinate set of the unauthorized area is stored as a specific target cluster thematic layer. The attribute table records the number of coordinate points within the cluster, the spatio-temporal activity index (calculated based on the occurrence frequency and duration), the distance to adjacent sensitive areas (such as ecological red lines), and the penalty records of historical similar specific targets. A second-level association query is realized through the spatio-temporal indexing mechanism of the spatial database.

[0090] In this embodiment, by combining the topological tolerance mechanism and the dynamic buffer, the spatial attribution of coordinate points and the sea use boundary is accurately determined, effectively eliminating the surveying and mapping errors and the interference of tidal dynamics, and improving the recognition accuracy of specific target behaviors inside and outside the legal range. The combination of dynamically adaptive clustering parameters and the analysis of historical specific target hotspots enhances the environmental adaptability of the identification of specific target clusters in unlicensed areas, and simultaneously suppresses the interference of isolated noise points and the misjudgment of new specific target patterns. The full-link data lineage management runs through the original monitoring, approval attributes, and processing intermediate results, constructs a spatio-temporal evidence chain for specific target events, and ensures the judicial credibility and aerial photography operability of the warning report.

[0091] Refer to Figure 2 As shown, the present invention also provides a multi-source sea area data intelligent supervision system, which is applied to the multi-source sea area data intelligent supervision method of any one of the above, including: A collection module, which is used to obtain the sea area geographical space data and historical sea area supervision data of the target sea area, and perform monitoring construction to obtain an island monitoring grid; An analysis module, which is used to obtain the multi-source monitoring data of the target sea area based on the island monitoring grid, and perform grid data identification to obtain a global patch dynamic data set; An association module, which is used to extract the coordinates of the suspected specific target areas in the global patch dynamic data set and calculate the specific target confidence of the coordinates; A processing module, which is used to compare the ranges of the coordinates of the suspected specific target areas, and perform behavior evaluation in combination with the specific target confidence of the coordinates to obtain a sea area specific target warning report.

[0092] A multi-source sea area data intelligent supervision system provided by the present invention realizes the dynamic perception of the whole-region patch changes of sea areas and islands by constructing an island monitoring grid and integrating multi-source monitoring data, overcomes the lag problem of traditional single remote sensing or manual inspection, and significantly improves the early detection ability of behaviors. By calculating the confidence level of suspected specific target areas, it avoids the misjudgment and missed reporting problems of static threshold judgment, making the identification of specific target behaviors more scientific and reliable. By comparing the scope by combining the sea use project database with real-time monitoring data, it can effectively distinguish legal construction from specific target behaviors, reduce supervision misjudgment, and improve the accuracy of aerial photography decision-making. By comprehensively analyzing the confidence level of specific targets and the results of behavior evaluation, it generates a precise early warning report for specific targets in the sea area, provides a data-driven decision-making basis for the supervision department, and improves the supervision response speed and governance efficiency.

[0093] Referring to Figure 3 As shown, the present invention also provides a multi-source sea area data intelligent supervision device, including: A memory for storing programs; A processor for executing programs to implement each step of the multi-source sea area data intelligent supervision method in any one of the above.

[0094] In this embodiment, the processor and the memory can be connected by a bus or other means. The memory may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as read-only memory, flash memory, hard disk or solid-state drive. The processor may be a general-purpose processor, such as a central processing unit, a digital signal processor, an application-specific integrated circuit, or one or more integrated circuits configured to implement the embodiments of the present invention.

[0095] The present invention also provides a storage medium storing computer instructions for causing a computer to execute the method according to any one of the above.

[0096] It should be noted that those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system and each module can refer to the corresponding processes in the foregoing method embodiments and will not be repeated herein.

[0097] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied to other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. An intelligent supervision method for multi-source sea area data, characterized in that Including: Obtain the sea area geospatial data and historical sea area supervision data of the target sea area, and construct an island monitoring grid; Obtain the multi-source monitoring data of the islands in the target sea area according to the island monitoring grid, and perform grid data identification to obtain a global patch dynamic data set; Extract the coordinates of the suspected specific target area in the global patch dynamic data set and calculate the coordinate specific target confidence; Compare the ranges of the coordinates of the suspected specific target area, and combine the coordinate specific target confidence to conduct a behavior assessment to obtain a sea area specific target early warning report.

2. The multi-source marine data intelligent supervision method according to claim 1 is characterized in that: The obtaining of the sea area geospatial data and historical sea area supervision data of the target sea area, and the monitoring construction to obtain an island monitoring grid includes: Based on the target sea area, obtain the sea area boundary data and geographic information, and perform spatial construction to obtain the sea area geospatial data; Obtain the marine aerial photography records and manual inspection reports of the target sea area, conduct analysis of sea use behaviors to obtain the historical sea area supervision data; Perform spatial overlay on the sea area geospatial data according to the historical sea area supervision data, and combine with a preset grid division threshold to perform initial grid setting to obtain initial monitoring grid units; Perform graph iteration optimization on the initial monitoring grid units according to the resource deployment constraint conditions of the target sea area to obtain the island monitoring grid.

3. The intelligent supervision method for multi-source sea area data according to claim 1, wherein The obtaining of the multi-source monitoring data of the target sea area according to the island monitoring grid, and the grid data identification to obtain a global patch dynamic data set includes: Plan multi-source monitoring points for the island monitoring grid to obtain the topological positions of monitoring nodes; Obtain the multi-spectral remote sensing images, UAV aerial photography data and inshore monitoring data of the target sea area according to the topological positions of the monitoring nodes, and perform multi-source alignment and integration to obtain the multi-source monitoring data; Perform ground object classification processing on the multi-source monitoring data based on a preset ground object classification rule library to obtain a grid ground object label distribution map; Perform regional edge detection and segmentation on the grid ground object label distribution map to obtain independent patch boundary vector data; Perform global matching on the independent patch boundary vector data according to the historical patch database to obtain the global patch dynamic data set.

4. The intelligent supervision method for multi-source sea area data according to claim 3, wherein The obtaining of the multi-spectral remote sensing images, UAV aerial photography data and inshore monitoring data of the target sea area according to the topological positions of the monitoring nodes, and the multi-source alignment and integration to obtain the multi-source monitoring data includes: Collect remote sensing images according to the topological positions of the monitoring nodes, and eliminate atmospheric and terrain errors to obtain the multi-spectral remote sensing images; Perform UAV aerial photography planning based on the topological positions of the monitoring nodes to obtain the UAV aerial photography data; Deploy inshore monitoring nodes according to the topological positions of the monitoring nodes, collect wave monitoring data, tide level monitoring data and water quality monitoring data, and perform inshore data integration to obtain the inshore monitoring data; Perform aerial triangulation solution and position registration on the UAV aerial photography data based on the topological positions of the monitoring nodes to obtain an aerial photography image data set; Perform multi-source registration on the multi-spectral remote sensing image, the aerial photography image dataset, and the inshore monitoring data to obtain the multi-source monitoring data.

5. The multi-source marine data intelligent supervision method according to claim 1 is characterized in that: Extracting the coordinates of the suspected specific target area of the whole-region patch dynamic dataset and calculating the coordinate-specific target confidence includes: Performing protected area spatial overlay analysis on the whole-region patch dynamic dataset according to the preset sea area ecological protection rules to obtain an initial set of specific target areas; Performing regional change detection on the initial set of specific target areas according to the whole-region patch dynamic dataset and identifying abnormal areas to obtain the outline of the suspected specific target area; Performing spatial coordinate transformation on the outline of the suspected specific target area according to the sea area geospatial data to obtain the coordinates of the suspected specific target area; Calculating the regional specific target probability for the coordinates of the suspected specific target area according to the historical sea area supervision data to obtain a preliminary confidence level; Calculating the specific target level for the preliminary confidence level according to the sea area ecological protection rules to obtain the coordinate-specific target confidence level.

6. The intelligent supervision method for multi-source sea area data according to claim 1, wherein Comparing the ranges of the coordinates of the suspected specific target area and combining the coordinate-specific target confidence level to perform behavior evaluation to obtain a sea area specific target early warning report, including: Performing legal sea use space range analysis on the coordinates of the suspected specific target area according to the preset sea use project database to obtain an abnormal coordinate set in the permitted coverage area and a specific target coordinate set in the non-permitted area; Performing regional sea use type matching on the abnormal coordinate set in the permitted coverage area based on the sea use permit information in the sea use project database to obtain a sea use compliance determination result; Filtering the regional coordinates of the specific target coordinate set in the non-permitted area according to the preset specific target confidence threshold to obtain a high-confidence specific target coordinate set; Performing specific target behavior pattern matching on the high-confidence specific target coordinate set according to the historical sea area supervision data to obtain a specific target behavior type level; Performing specific target evaluation on the sea use compliance determination result, the high-confidence specific target coordinate set, and the specific target behavior type level to obtain the sea area specific target early warning report.

7. The multi-source sea area data intelligent supervision method according to claim 6, characterized in that Performing legal sea use space range analysis on the coordinates of the suspected specific target area according to the preset sea use project database to obtain an abnormal coordinate set in the permitted coverage area and a specific target coordinate set in the non-permitted area, including: Performing polygon spatial boundary calculation on the coordinates of the suspected specific target area according to the legal sea use range data in the sea use project database to obtain a preliminary screening coordinate set; Filtering the preliminary screening coordinate set for spatial relationship and time validity to obtain an expired permit coordinate set and a valid permit coverage coordinate set; Performing comparison of the permit activity type attributes on the valid permit coverage coordinate set to obtain the abnormal coordinate set in the permitted coverage area; Performing non-permitted area clustering on the preliminary screening coordinate set and the expired permit coordinate set according to the legal sea use range data to obtain the specific target coordinate set in the non-permitted area.

8. A multi-source marine data intelligent monitoring system, characterized by: Applied to the multi-source sea area data intelligent supervision method described in any one of the above claims 1-7, including: A collection module, which is used to obtain the marine geospatial data and historical marine supervision data of the target sea area, and conduct monitoring construction to obtain an island monitoring grid; An analysis module, which is used to obtain the multi-source monitoring data of the target sea area based on the island monitoring grid, and conduct grid data identification to obtain a dynamic data set of global patches; An association module, which is used to extract the coordinates of suspected specific target areas in the dynamic data set of global patches and calculate the confidence level of specific targets of the coordinates; A processing module, which is used to compare the ranges of the coordinates of the suspected specific target areas, and conduct behavior evaluation in combination with the confidence level of specific targets of the coordinates to obtain a warning report on specific targets in the sea area.

9. An intelligent supervision device for multi-source sea area data, characterized in that, Including: A memory for storing programs; A processor for executing the program to implement each step of a multi-source marine data intelligent supervision method according to any one of claims 1-7.

10. A storage medium, characterized in that, Computer instructions are stored, and the computer instructions are used to cause a computer to execute the method according to any one of claims 1 to 7.

Images