Port and bridge large-scale target thematic map generation method based on SAR technology

Abstract

The invention belongs to the technical field of information remote sensing, and particularly discloses a port and bridge large-scale target thematic map generation method based on an SAR technology, and the method comprises the steps: obtaining multi-temporal synthetic aperture radar image data of a port region, processing the original signal intensity distribution of the image data through denoising and filtering, and obtaining a smoothed scattering intensity map; according to the scattering intensity graph, extracting echo amplitude and phase information of a target area, fusing multi-polarization channel data, and determining a scattering characteristic parameter set; grouping the scattering characteristic parameter set based on a clustering algorithm, and separating preliminary clustering results of facility classes and residue classes to obtain feature vector representation; matching the feature vector representation with a stored shape template according to a pre-established feature database; the objective of the invention is to solve the problem in the prior art that environment management is difficult to accurately position a business scene due to mixed distribution of facilities and residues in a port area.

Description

Technical Field

[0001] This invention relates to the field of information remote sensing technology, specifically to a method for generating thematic maps of large targets such as ports and bridges based on SAR technology. Background Technology

[0002] As a core hub for global trade and logistics, the operational efficiency and environmental cleanliness of ports directly impact economic benefits and ecological sustainability. Intelligent identification and thematic map generation based on synthetic aperture radar (SAR) technology can achieve precise monitoring of port facilities and residues through high-resolution imagery, providing crucial support for operation and maintenance management and environmental governance. The importance of this field lies in its ability to not only enhance the intelligence level of port operations but also provide a data foundation for environmental protection. However, existing methods have limitations. Traditional optical imagery is restricted by weather and lighting conditions, making it difficult to acquire data stably around the clock. Furthermore, the complex scattering characteristics of SAR imagery challenge the accuracy of target identification and information retrieval, especially in complex port environments where the characteristics of facilities and residues are often confused. The core challenge lies first in the difficulty of analyzing scattering characteristic parameters. The differences in material and shape of targets such as wharves, cranes, storage yards, and abandoned equipment within the port area result in complex and variable intensity and patterns of scattering signals, making it difficult to accurately distinguish target types using a single parameter.

[0003] For example, the scattering characteristics of stockpiled cargo and floating debris can be confused due to their similar materials, affecting the reliability of cleanliness assessments. This problem further leads to the low efficiency of multi-dimensional matching in information retrieval algorithms. In actual operations, when it is necessary to quickly locate specific facilities or residues within a port, existing algorithms often struggle to accurately locate targets and query attributes in a short time due to incomplete feature database coverage or imperfect matching rules.

[0004] For example, when handling abandoned equipment in the storage yard, maintenance personnel may delay cleanup decisions because the retrieval system cannot effectively distinguish the equipment from surrounding cargo. Therefore, how to achieve rapid target location and classification by constructing a comprehensive database of port facilities and residue characteristics, combined with scattering characteristic parameters and shape feature extraction algorithms, has become a key issue in the generation of thematic maps for port operation and maintenance and cleanliness assessment. Summary of the Invention

[0005] This invention provides a method for generating large-scale thematic maps of ports and bridges based on SAR technology, aiming to solve the problem in existing technologies where the mixed distribution of port facilities and residues makes it difficult to accurately locate environmental governance in business scenarios.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for generating thematic maps of large targets such as ports and bridges based on SAR technology includes: acquiring multi-temporal synthetic aperture radar image data of the port area; processing the original signal intensity distribution of the image data through denoising filtering to obtain a smoothed scattering intensity map; extracting echo amplitude and phase information of the target area based on the scattering intensity map, fusing multi-polarization channel data, and determining a set of scattering characteristic parameters; grouping the scattering characteristic parameter set based on a clustering algorithm, and separating the preliminary clustering results of facility classes and residue classes to obtain feature vector representations; matching the feature vector representations with stored shape templates according to a pre-established feature database, and if the match is similar... If the density exceeds a preset threshold, the target type is confirmed, and a classification label is obtained. Based on the classification label, combined with the scattering characteristic parameter set and shape analysis results, a multi-dimensional attribute description is fused to obtain the target's precise spatial coordinates. Based on the precise spatial coordinates, boundary detection is optimized using a convolutional neural network to process edge obfuscation in the image, resulting in a refined target contour map. Based on the refined target contour map, the residue distribution density is extracted. If the density exceeds a preset threshold, high-risk areas are marked, and a cleanliness assessment layer is obtained. Based on the cleanliness assessment layer, facility location information is overlaid to generate a vector thematic map. Multi-layer data is fused to determine the environmental governance reference output.

[0007] In one aspect of this disclosure, the step of acquiring multi-temporal synthetic aperture radar image data of a port area and obtaining a smoothed scattering intensity map by processing the original signal intensity distribution of the image data through denoising filtering includes: Acquire multi-temporal synthetic aperture radar image data of the port area, and separate the original signal intensity through data preprocessing; The original signal strength is processed by adaptive denoising filtering, and a smooth signal strength distribution is generated. Based on the spatial domain analysis, the signal intensity distribution is smoothed to obtain scattering intensity characteristics. If the scattering intensity characteristics meet a preset threshold, the multi-temporal variation trend is extracted through time series analysis to obtain time series scattering characteristics. Based on the temporal scattering characteristics, a convolutional neural network is used to classify the land cover types in the port area and generate classification results. Based on the superimposed classification results and the smoothed signal intensity distribution, a scattering intensity map of the port area is generated; Based on the scattering intensity map, and by optimizing the boundary smoothness through mean filtering, the final scattering intensity map is obtained.

[0008] In one aspect of this disclosure, the step of extracting the echo amplitude and phase information of the target region based on the scattering intensity map, fusing multi-polarization channel data, and determining the scattering characteristic parameter set includes: Based on the original data of the target region obtained from the scattering intensity map, the boundary of the target region is determined by a preset region segmentation algorithm to obtain the pixel set of the target region. Based on the pixel set of the target area, the echo amplitude data is extracted, and the amplitude value of each pixel is calculated using the Fast Fourier Transform algorithm to obtain the echo amplitude distribution. Based on the pixel set of the target region, phase data is extracted, and the phase information of each pixel is processed using a phase unwrapping algorithm to obtain the phase data distribution; The phase unwrapping algorithm is configured as follows:

[0009] in, Represented as the true phase value after untangling; This is represented by the observed entanglement phase value; Represented as at pixel The integer multiple to be added Number of jumps; Based on the fused dataset, scattering characteristics are extracted using principal component analysis algorithm to obtain scattering characteristic parameters; Based on the scattering characteristic parameters, the correlation coefficient between each parameter is calculated. If the correlation coefficient is higher than a preset threshold, the corresponding parameter is retained to obtain the final scattering characteristic parameter set. By normalizing the final scattering characteristic parameter set, a standardized parameter set is generated, thus obtaining a description of the scattering characteristics of the target region.

[0010] In one aspect of this disclosure, the step of grouping the scattering characteristic parameter set based on a clustering algorithm and separating the preliminary clustering results of facility classes and residue classes to obtain feature vector representations includes: The scattering characteristic parameter set is grouped based on the K-means clustering algorithm to obtain preliminary clustering results and feature vector representations for facility categories and residue categories. If the intra-cluster variance of the preliminary clustering results is greater than a preset threshold, the cluster centers are adjusted through iterative optimization to obtain optimized clustering results. Based on the optimized clustering results, the feature vector of each cluster is extracted to obtain the feature vector representations of facility category and residue category; Principal component analysis is used to reduce the dimensionality of the eigenvectors to obtain the dimensionality-reduced eigenvectors. If the dimensionality of the eigenvectors after dimensionality reduction is lower than a preset threshold, linear discriminant analysis is used to further separate the facility category and the residue category to obtain the separated eigenvectors. Based on the separated feature vectors, the inter-cluster distance is calculated, the accuracy of category differentiation is judged, and the final classification result is obtained. Based on the final classification results, feature vector representations of facility categories and residue categories are obtained.

[0011] In one aspect of this disclosure, the step of matching the feature vector representation with a stored shape template based on a pre-established feature database, and confirming the target type and obtaining a classification label if the matching similarity is higher than a preset threshold, includes: Obtain the template shape set and feature vector set based on the preset feature database; The similarity score between the target feature vector and the template shape is calculated based on the cosine similarity algorithm. If the similarity score exceeds the preset threshold, the target matching template is determined and a preliminary classification label is obtained. The preliminary classification labels are verified based on the cluster analysis, and optimized classification labels are obtained. Based on the optimized classification labels, the corresponding target type is extracted from the feature database. If the target type matches the preset classification rule, the final classification label is obtained. The feature database is updated based on the final classification labels to obtain the optimized template shape set.

[0012] In one aspect of this disclosure, the step of fusing the classification labels, the scattering characteristic parameter set, and the shape analysis results into a multi-dimensional attribute description to obtain the precise spatial coordinates of the target includes: Extract target-related tag data from a pre-set database, and use a random forest algorithm to initially filter the tags to obtain a set of categorized tags; Based on the classification label set, a scattering characteristic parameter set is obtained from the sensor data, and the parameters are dimensionality reduced by principal component analysis to obtain a dimensionality-reduced scattering feature set. Based on the dimensionality-reduced scattering feature set, shape analysis results are obtained, and the target shape is feature-extracted through a convolutional neural network to obtain a shape feature description; The scattering feature set and shape feature description after dimensionality reduction are fused. If the fused feature vector meets the preset similarity threshold, a multi-dimensional attribute description is generated. Based on the multi-dimensional attribute description, and by iteratively calculating the spatial position of the target using the Kalman filter algorithm, preliminary spatial coordinates are obtained; Based on the preliminary spatial coordinates and combined with the preset coordinate correction model, if the coordinate deviation is less than the preset threshold, the coordinates are optimized and adjusted to determine the final spatial coordinates. The accuracy of target positioning is determined by using the final spatial coordinates and verifying the distance between the coordinates and the preset reference point using the Euclidean distance calculation method.

[0013] In one aspect of this disclosure, the step of obtaining a target contour map of a string by optimizing boundary detection and processing edge obfuscation in an image based on the precise spatial coordinates and using a convolutional neural network includes: Based on the raw image data obtained from the image data input, the image is denoised and standardized using preprocessing techniques to obtain the first image data; Based on the first image data, feature extraction is performed using a convolutional neural network to obtain a feature map containing edge information; Based on the edge information in the feature map, if the edge pixel intensity is lower than a preset threshold, the edge clarity is enhanced by a boundary optimization algorithm to obtain a second feature map. Based on the second feature map and through spatial location analysis technology, the target region corresponding to the precise spatial coordinates is obtained, and the first target region is generated. Based on the first target region, and by separating the target region from the background using target region segmentation technology, a second target region is obtained; Based on the second target region, a refined target contour map is generated by applying contour generation technology. Based on the refined target contour map, and through edge sharpness enhancement technology, the contour boundary is further optimized to obtain the final target contour map.

[0014] In one aspect of this disclosure, the step of extracting the residue distribution density based on the refined target contour map, and marking high-risk areas if the density exceeds a preset threshold to obtain a cleanliness assessment layer, includes: Based on the target contour map, obtain the residue distribution data, use image segmentation algorithm to determine the residue area and obtain the distribution density. If the distribution density exceeds the preset threshold, generate high-risk areas through marking process to obtain preliminary area division. Based on the preliminary regional division, high-risk areas are classified using a clustering algorithm to obtain classification results; The clustering algorithm for classifying high-risk areas is configured as follows:

[0015] in, This is represented as the sum of squared errors of the entire cluster; This is represented by the number of clusters in the clustering; Represented as the first A cluster; Represented as data points within a cluster; Represented as the first The center point of each cluster; Based on the cleanliness assessment layer, a convolutional neural network is used to analyze the features of the assessment layer to obtain feature distribution data. If the feature distribution data meets the preset conditions, the final result is generated and output through data processing. Based on the final output, a visual cleanliness assessment layer is generated to obtain the final cleanliness level.

[0016] In one aspect of this disclosure, the step of generating a vector thematic map by overlaying facility location information based on the cleanliness assessment layer, fusing multi-layer data, and determining an environmental governance reference output includes: Environmental cleanliness assessment data and facility location information are obtained from the data source, stored as a structured dataset, and an initial dataset is obtained. Spatial analysis is performed based on the initial dataset. Geographic Information System (GIS) tools are used to calculate the spatial relationship between cleanliness assessment data and facility locations to obtain a spatial association dataset. If the cleanliness value in the spatial association dataset is lower than a preset threshold, the facility locations in the corresponding areas are marked to obtain a marked location dataset. Based on the labeled location dataset, vector data processing methods are used to generate vector thematic maps, resulting in visualized thematic map data; Based on a multi-layer data fusion algorithm, combined with vector thematic map data and environmental cleanliness assessment data, comprehensive environmental data is generated to obtain a fused dataset; Based on the fused dataset, a decision tree algorithm is used to analyze the priority of environmental governance and obtain the ranking results of governance schemes. In one aspect of this disclosure, the step of generating a vector thematic map based on the labeled location dataset using a vector data processing method to obtain visualized thematic map data includes: Vector data is obtained from the marked location dataset, and the vector data includes spatial coordinates and attribute information. Based on the vector data, a spatial interpolation algorithm is used to generate a vector thematic map, thereby obtaining preliminary thematic map data; Topological analysis is performed based on the preliminary thematic map data to obtain the spatial relationships between vector elements; Boundary overlap and adjacency attributes are extracted from the spatial relationships to generate a spatial association dataset; Based on the spatial association dataset, a clustering algorithm is used to obtain the features of the classified thematic map. Multi-scale rasterization processing is performed based on the thematic map features to generate multi-resolution raster data; Based on the multi-resolution raster data, and extracting color distribution attributes from the multi-resolution raster data, the visualization rendering parameters are determined; Dynamic rendering is performed based on the visualization rendering parameters to obtain the final visualized thematic map data; Extract vector data from the labeled location dataset, the vector data including spatial coordinates and attribute fields; The vector data is processed using the Kriging interpolation algorithm to generate continuous vector thematic map data; A topology check is performed on the vector thematic map data to obtain the adjacency and overlap relationships between features; Based on the adjacency and overlap relationships, spatial correlation features are extracted from the adjacency and overlap relationships to obtain a spatial weight matrix; Based on the spatial weight matrix, the K-means clustering algorithm is used to obtain the classified thematic map partition data. Adaptive rasterization processing is performed on the partitioned data to obtain a multi-scale raster dataset. Color gradient attributes are extracted from the multi-scale raster dataset to obtain layered rendering parameters; Dynamic visualization rendering is performed based on the layered rendering parameters to obtain high-resolution thematic map data.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention achieves precise classification, spatial positioning, and cleanliness assessment of facilities and residues by integrating technologies such as denoising filtering, multi-polarization data analysis, clustering algorithms, feature matching, convolutional neural network boundary detection, and multi-layer data fusion. First, the original radar image is processed by denoising filtering to extract a smoothed scattering intensity map, and multi-polarization channel data is fused to generate a set of scattering characteristic parameters. Next, a clustering algorithm is used to separate the features of facilities and residues, and a pre-set shape template is used to confirm the target type. Further, a convolutional neural network is used to optimize boundary detection, generating a refined target contour map, and residue distribution density is extracted to mark high-risk areas. Finally, facility positioning information is overlaid to generate a vector thematic map, providing accurate reference for environmental governance. This invention significantly improves the accuracy and efficiency of environmental monitoring and governance in port areas through multi-dimensional data fusion and intelligent analysis. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart of a method for generating large-scale thematic maps of ports and bridges based on SAR technology according to the present invention.

[0020] Figure 2This is one of the schematic diagrams of a method for generating large-scale target thematic maps of ports and bridges based on SAR technology according to the present invention.

[0021] Figure 3 This is a second schematic diagram of a method for generating large-scale target thematic maps of ports and bridges based on SAR technology according to the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to embodiments. These embodiments are merely some, not all, of the embodiments described. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the protection scope of the present invention.

[0023] Example 1 Please see Figure 1 As shown, this embodiment discloses a method for generating thematic maps of large targets such as ports and bridges based on SAR technology, including: acquiring multi-temporal synthetic aperture radar image data of a port area; processing the original signal intensity distribution of the image data through denoising filtering to obtain a smoothed scattering intensity map; extracting echo amplitude and phase information of the target area based on the scattering intensity map, fusing multi-polarization channel data, and determining a set of scattering characteristic parameters; grouping the scattering characteristic parameter set based on a clustering algorithm, and separating the preliminary clustering results of facility classes and residue classes to obtain feature vector representations; matching the feature vector representations with stored shape templates according to a pre-established feature database. If the similarity score is higher than a preset threshold, the target type is confirmed and a classification label is obtained. Based on the classification label, combined with the scattering characteristic parameter set and shape analysis results, a multi-dimensional attribute description is fused to obtain the target's precise spatial coordinates. Based on the precise spatial coordinates, boundary detection is optimized through a convolutional neural network to process edge obfuscation in the image and obtain a refined target contour map. Based on the refined target contour map, the residue distribution density is extracted. If the density exceeds a preset threshold, high-risk areas are marked to obtain a cleanliness assessment layer. Based on the cleanliness assessment layer, facility location information is overlaid to generate a vector thematic map. Multi-layer data is fused to determine the environmental governance reference output.

[0024] This invention achieves precise classification, spatial positioning, and cleanliness assessment of facilities and residues by integrating technologies such as denoising filtering, multi-polarization data analysis, clustering algorithms, feature matching, convolutional neural network boundary detection, and multi-layer data fusion. First, the original radar image is processed by denoising filtering to extract a smoothed scattering intensity map, and multi-polarization channel data is fused to generate a set of scattering characteristic parameters. Next, a clustering algorithm is used to separate the features of facilities and residues, and a pre-set shape template is used to confirm the target type. Further, a convolutional neural network is used to optimize boundary detection, generating a refined target contour map, and residue distribution density is extracted to mark high-risk areas. Finally, facility positioning information is overlaid to generate a vector thematic map, providing accurate reference for environmental governance. This invention significantly improves the accuracy and efficiency of environmental monitoring and governance in port areas through multi-dimensional data fusion and intelligent analysis.

[0025] Example 2 Please see Figures 1-3 As shown, this embodiment is a further optimization based on Embodiment 1. In this embodiment, the specific steps of the method for generating large-scale target thematic maps of ports and bridges based on SAR technology are as follows: S101. Acquire multi-temporal synthetic aperture radar image data of the port area, and use noise reduction filtering to process the original signal intensity distribution of the image data to obtain a smoothed scattering intensity map.

[0026] Multi-temporal synthetic aperture radar (SAR) imagery data of the port area was acquired, and the original signal intensity was separated through data preprocessing. Adaptive denoising filtering was used to process the original signal intensity, generating a smooth signal intensity distribution. Spatial domain analysis was performed on the smoothed signal intensity distribution to determine scattering intensity characteristics. If the scattering intensity characteristics met a preset threshold, multi-temporal variation trends were extracted through temporal analysis to obtain temporal scattering characteristics. Based on the temporal scattering characteristics, a convolutional neural network was used to classify the land cover types in the port area, generating classification results. By overlaying the classification results with the smoothed signal intensity distribution, a scattering intensity map of the port area was generated. For the scattering intensity map, mean filtering was used to optimize the boundary smoothness, resulting in the final scattering intensity map.

[0027] Specifically, to acquire multi-temporal synthetic aperture radar (SAR) imagery data of the port area, C-band SAR images can be downloaded from the space agency's satellite data platform via a programming interface. The VV polarization mode is selected, with a spatial resolution of 10 meters. The time series covers January to March 20XX, with one image acquired per month. The image size is 1000×1000 pixels. After downloading, the image data is stored in single-view complex (SLC) format, containing amplitude and phase information, with amplitude values ​​typically ranging from 0 to 65535.

[0028] Next, the original signal intensity distribution is subjected to noise reduction filtering. An improved Lee filtering algorithm is used, the filtering window is set to 5×5 pixels, and the noise standard deviation is estimated to be 0.2. The calculation formula is:

[0029] in, The average pixel value within the window; As a weighting factor;

[0030] in, For signal variance, This represents the noise variance.

[0031] After filtering, the noise in the signal strength distribution is significantly reduced, with the standard deviation decreasing from 0.2 to 0.05, while preserving the edge details of ships and docks in the port area.

[0032] Based on the filtered image, a scattering intensity map is calculated by converting the filtered amplitude values ​​into backscattering coefficients. ; The formula is:

[0033] in, This is a calibration constant, provided by Sentinel-1 metadata, with a typical value of 10^5.

[0034] When analyzing the scattering intensity map, the ships in the port area The values ​​are between -10 and -5 dB, and between -15 and -12 dB in the dock area, indicating that the scattering intensity from ships is higher than that from the dock, which is consistent with the high reflectivity of metal structures.

[0035] Based on this, dynamic targets in the port area can be further extracted, and multi-temporal difference analysis can be used to calculate the values ​​of adjacent temporal images. The difference, with a threshold set to 2dB, identifies the areas where ships move and generates dynamic change maps, providing data support for port flow monitoring.

[0036] S102. Based on the scattering intensity map, extract the echo amplitude and phase information of the target area, fuse the multi-polarization channel data, and determine the scattering characteristic parameter set.

[0037] The raw data of the target region is obtained from the scattering intensity map. A pre-defined region segmentation algorithm is used to determine the boundary of the target region, resulting in a pixel set. Based on this pixel set, echo amplitude data is extracted, and the amplitude value of each pixel is calculated using a Fast Fourier Transform algorithm to obtain the echo amplitude distribution. Finally, based on the pixel set, phase data is extracted, and the phase information of each pixel is processed using a phase unwrapping algorithm to obtain the phase data distribution.

[0038] The phase unwrapping algorithm is configured as follows:

[0039] This represents the true phase value after untangling. This represents the observed entanglement phase value. Indicates at pixel point The integer multiple to be added The number of jumps, as described in the formula above, describes the basic mathematical relationship of phase unwrapping. Channel information is obtained from the multi-polarization channel. If the signal-to-noise ratio of the channel data is higher than a preset threshold, a weighted average method is used to fuse the echo amplitude distribution and phase data distribution to obtain a fused data set. Based on the fused data set, principal component analysis is used to extract scattering characteristics and determine scattering characteristic parameters. Based on the scattering characteristic parameters, the correlation coefficients between each parameter are calculated. If the correlation coefficients are higher than a preset threshold, the corresponding parameters are retained, resulting in the final scattering characteristic parameter set. By normalizing the final scattering characteristic parameter set, a standardized parameter set is generated, determining the scattering characteristic description of the target region.

[0040] Specifically, based on the input scattering intensity map, the echo amplitude and phase information of the target region are first extracted using an image preprocessing algorithm. Assuming the input is a SAR image, a threshold segmentation algorithm based on grayscale values ​​is used, with a threshold T=0.75, to extract pixels in the target region. For each pixel, the echo amplitude is calculated:

[0041] in and These are in-phase and quadrature components, respectively. Assume a certain pixel... =0.6, =0.8, therefore =1.0; Phase information is calculated using the arctangent function. .

[0042] To ensure accuracy, a medium-range filter (3x3 window) is applied to remove noise. Then, the multi-polarization channel data (HH, HV, VV) are fused to construct the covariance matrix C, with matrix elements... ,in Let be the complex scattered signal of the i-th channel, and <*> denote the time average.

[0043] For example, HH channel signal S HH =0.7+0.3i, HV channel S HV =0.2+0.1i, calculate C HH,HV =0.17+0.05i. Based on the covariance matrix, the eigenvalues ​​λ1=1.2, λ2=0.3, and λ3=0.1 are extracted using the eigenvalue decomposition algorithm, and the scattering entropy is calculated: This characterizes the randomness of scattering.

[0044] Further eigenvector analysis determined that the main scattering mechanism was surface scattering.

[0045] To form a rigorous logic, and considering business scenarios such as land cover classification, the scattering characteristic parameters—amplitude A, phase φ, and entropy H—are input into the support vector machine model. The kernel function chosen is RBF, and the parameters... γ =0.5, C =1.0, with a classification accuracy of 90%. Through the above steps, from amplitude and phase extraction to multi-polarization fusion, and then to characteristic parameter determination, a complete technology chain is formed, which is applicable to scenarios such as ground object monitoring.

[0046] S103. The scattering characteristic parameter set is grouped using a clustering algorithm to separate the facility class and the residue class into preliminary clustering results, and feature vector representations are obtained.

[0047] K-means clustering is used to group the scattering characteristic parameter set to obtain preliminary clustering results for facility categories and residue categories, resulting in feature vector representations. If the intra-cluster variance of the preliminary clustering results is greater than a preset threshold, the cluster centers are adjusted iteratively to obtain optimized clustering results. Based on the optimized clustering results, feature vectors are extracted for each cluster to obtain feature vector representations for facility categories and residue categories. Principal component analysis is used to reduce the dimensionality of the feature vectors to obtain dimensionality-reduced feature vectors. If the dimensionality of the dimensionality-reduced feature vectors is lower than a preset threshold, linear discriminant analysis is used to further separate facility categories and residue categories, obtaining separated feature vectors. Based on the separated feature vectors, inter-cluster distances are calculated to determine the accuracy of category differentiation, resulting in the final classification results. Feature vector representations for facility categories and residue categories are generated based on the final classification results.

[0048] Specifically, for the grouping of scattering characteristic parameter sets, the K-means clustering algorithm is used to initially separate the facility class and the residue class. It is assumed that the input parameter set contains 1000 samples, each with a 3-dimensional feature vector: scattering intensity (0 to 100), angular deviation (-30° to 30°), and frequency response (1 to 10 GHz). First, data preprocessing normalizes the features through Z-score standardization; The calculation formula is:

[0049] in The mean, The standard deviation is used to obtain the normalized eigenvectors.

[0050] For example, if the mean scattering intensity is 50 and the standard deviation is 10, then the standardized value of a sample with an intensity of 60 is (60-50) / 10=1. Next, the K-means algorithm is applied, setting the cluster number K=2 to distinguish between facility and residue categories, and two random centroids are initialized, such as (0.5, 0.2, 0.3) and (-0.5, -0.2, -0.3). The distance from each sample to the centroid is calculated using Euclidean distance, with the formula: The sample is assigned to the nearest centroid, and the centroid is iteratively updated until convergence.

[0051] For example, the distance from a sample (1, 0.1, 0.4) to the centroid (0.5, 0.2, 0.3) is... After allocation, the centroid is recalculated as the mean of the samples within the cluster. After 10 iterations, two clusters are obtained: cluster 1 (facility class, 600 samples, centroid (0.8, 0.15, 0.35)) and cluster 2 (residue class, 400 samples, centroid (-0.7, -0.25, -0.3)).

[0052] To verify the separation effect, the intra-cluster variance and inter-cluster distance were calculated. The formula for intra-cluster variance is: The variance of cluster 1 is 0.12, that of cluster 2 is 0.15, and the inter-cluster distance is... This indicates good separation. Finally, the cluster centroids are used as feature vectors, with the facility class feature vector being (0.8, 0.15, 0.35) and the residue class feature vector being (-0.7, -0.25, -0.3), for subsequent classification tasks to ensure relevance to the business scenario and improve classification accuracy.

[0053] S104. Using a pre-established feature database, match the feature vector representation with the stored shape template. If the similarity is higher than a preset threshold, confirm the target type and obtain the classification label.

[0054] A set of template shapes and a set of feature vectors are obtained from a pre-defined feature database. A cosine similarity algorithm is used to calculate the similarity score between the target feature vector and the template shape. If the similarity score exceeds a pre-defined threshold, the target matching template is determined, and a preliminary classification label is obtained. Cluster analysis is used to verify the preliminary classification label, and optimized classification labels are obtained. Based on the optimized classification labels, the corresponding target type is extracted from the feature database. If the target type matches the pre-defined classification rules, a final classification label is generated. The feature database is updated using the final classification labels, and the set of template shapes is optimized.

[0055] Specifically, by matching the shape templates represented and stored in the feature vector representation with a pre-established feature database, feature vectors are first extracted from the input image. For example, a convolutional neural network is used to extract 128-dimensional feature vectors, specifically ResNet. 50 The model takes an input image normalized to 224×224 pixels. After multiple convolutional and pooling layers, a feature vector v = [v1, v2, ..., v128] is obtained. This vector is then matched against shape templates in a feature database. The database contains 1000 templates, each corresponding to a target type, and each template is stored as a 128-dimensional vector. The matching uses a cosine similarity algorithm, calculated using the following formula: ; Where t is the template vector.

[0056] For example, the similarity between input vector v and template t1 is 0.92. A preset threshold is set to 0.85. If the similarity is higher than 0.85, a match is considered successful, confirming the target type as the category corresponding to template t1. If multiple templates have similarities higher than the threshold, the highest similarity is selected. For example, if v has similarities of 0.92 with t1 and 0.88 with t2, t1 is selected.

[0057] After the category labels are generated, the output is a circle with a confidence score of 0.92. To ensure robustness, if the similarity score is below 0.85, a second verification can be performed based on the contextual business logic, and the category labels can be updated. This entire process is completed automatically by the algorithm, with rigorous logic to ensure efficient classification.

[0058] S105. Obtain the classification label, combine it with the scattering characteristic parameter set and shape analysis results, and fuse them into a multi-dimensional attribute description to determine the precise spatial coordinates of the target.

[0059] The process involves obtaining classification labels by extracting target-related label data from a pre-defined database and using a random forest algorithm for initial label filtering to obtain a classification label set. Based on this set, a scattering characteristic parameter set is obtained from sensor data, and principal component analysis (PCA) is used to reduce the dimensionality of these parameters, resulting in a dimensionality-reduced scattering feature set. Using this dimensionality-reduced feature set, shape analysis results are obtained, and a convolutional neural network is used to extract target shape features, yielding a shape feature description. The dimensionality-reduced scattering feature set and shape feature description are then fused. If the fused feature vector meets a pre-defined similarity threshold, a multi-dimensional attribute description is generated. Based on this multi-dimensional attribute description, a Kalman filter algorithm is used to iteratively calculate the target's spatial location, obtaining preliminary spatial coordinates. Using these preliminary spatial coordinates and a pre-defined coordinate correction model, if the coordinate deviation is less than a pre-defined threshold, the coordinates are optimized and adjusted to determine the final spatial coordinates. Finally, the final spatial coordinates are obtained, and the distance between the coordinates and a pre-defined reference point is verified using Euclidean distance calculation to assess the accuracy of target positioning.

[0060] Specifically, firstly, classification labels are obtained by semantic segmentation of the target using a deep learning model. A convolutional neural network combined with the ResNet-50 architecture is used as input, with an RGB image of the target region at a resolution of 512x512 pixels, and the output classification labels such as vehicles, pedestrians, and buildings.

[0061] For example, given an input image of a city street scene, the model predicts the target as a "vehicle" with a confidence level of 0.95. Next, the scattering characteristic parameter set is obtained through radar point cloud data analysis using a millimeter-wave radar at a frequency of 77 GHz. The radial velocity of the target is extracted, for example, 5.2 m / s, and the scattering cross-section is set to 1.5 m. 2 The distance is 50.3m. The specific algorithm uses Fast Fourier Transform to process the radar echo, calculates the Doppler frequency shift, and obtains the velocity parameters; the RCS is estimated through echo intensity integration. Shape analysis results utilize 3D point cloud data and employ a volume parallax algorithm from the point cloud library to calculate the dimensions of the target's circumscribed rectangle, for example, 4.8m long, 1.9m wide, and 1.5m high. Principal component analysis is then used to extract the target's principal axis skew angle of 30°.

[0062] The multi-dimensional attribute descriptions, fused with the above results, construct a feature vector containing classification labels, scattering parameters (velocity: 5.2 m / s, RCS 1.5 m), and other parameters. 2 Distance: 50.3m and shape: dimensions 4.8x1.9x1.5m, deflection angle 30°.

[0063] Finally, the precise spatial coordinates of the target were determined by using a Kalman filter algorithm, fusing radar range (50.3m), angles (azimuth 45°, elevation 5°), and point cloud data, and iteratively updating the target position to obtain three-dimensional coordinates x=35.5m, y=35.5m, z=1.2m.

[0064] To ensure logical rigor, a weighted average method was used during the fusion process to optimize coordinate accuracy: radar 0.6 and point cloud 0.4, keeping the error within 0.1m. If the original data is missing, it can be generated through interpolation from historical data. For example, when point cloud data is missing, shape parameters can be predicted using data from the previous frame to maintain continuity.

[0065] S106. For the precise spatial coordinates, a convolutional neural network is used to optimize boundary detection, process edge obfuscation in the image, and obtain a refined target contour map.

[0066] Raw image data is acquired from the image data input. Preprocessing techniques are used to denoise and standardize the images, resulting in first image data. Based on the first image data, a convolutional neural network is used for feature extraction to obtain a feature map containing edge information. For the edge information in the feature map, if the edge pixel intensity is lower than a preset threshold, a boundary optimization algorithm is applied to enhance edge sharpness, resulting in a second feature map. Using the second feature map, spatial location analysis techniques are used to determine the target region corresponding to precise spatial coordinates, generating a first target region. Based on the first target region, target region segmentation techniques are used to separate the target region from the background, resulting in a second target region. For the second target region, contour generation techniques are applied to generate a refined target contour map. Using the refined target contour map, edge sharpness enhancement techniques are used to further optimize the contour boundaries, obtaining the final target contour map.

[0067] Specifically, for boundary detection optimization with precise spatial coordinates, the input image is first processed using a convolutional neural network. Assume the image size is 512x512 pixels with a grayscale value of 0-255. The U-Net architecture is adopted, consisting of 4 downsampling layers and 4 upsampling layers. Each layer uses a 3x3 convolutional kernel with a stride of 1, padding of 1, and the ReLU activation function. The input image is first preprocessed using a Gaussian filter (σ=1.5) to smooth noise and enhance edge features. Next, the CNN extracts features at multiple scales to generate feature maps, capturing the spatial information of the edges.

[0068] To optimize edge obfuscation, a boundary loss function is introduced, combining cross-entropy loss and Dice loss with weights of 0.7 and 0.3 respectively, ensuring the model's sensitivity to edge pixels. During training, the Adam optimizer is used with a learning rate of 0.001, a batch size of 16, and 100 epochs. The training dataset contains 1000 labeled images, the validation set contains 200 images, and the test set contains 200 images. The model output is a probability map with a threshold of 0.5, and the probability values ​​are binarized into edge or non-edge pixels. To further refine the target contour, a conditional random field post-processing is used, with Gaussian kernel parameters set. =80, =15, iterated 5 times, and optimized edge continuity. Analysis shows that the model achieved an F1 score of 0.92 on the test set, a significant improvement compared to traditional Canny edge detection (F1=0.75). The pixel classification accuracy of edge-confused regions improved from 85% to 93%, indicating that CNN combined with CRF effectively reduced edge blurring. The final output contour map maintained a spatial coordinate accuracy of ±1 pixel, meeting the requirements for accurate boundary detection. By combining with the subsequent object tracking module, the contour map can be used as input to support real-time object localization, logically forming a complete technology chain from image input to contour output.

[0069] S107. Extract the residue distribution density from the refined target contour map. If the density exceeds a preset threshold, mark the high-risk area to obtain the cleanliness assessment layer.

[0070] Residue distribution data is obtained from the target contour map. Image segmentation algorithms are used to determine the residue regions and obtain the distribution density. If the distribution density exceeds a preset threshold, high-risk areas are generated through labeling, resulting in preliminary region division. Based on the preliminary region division, clustering algorithms are used to classify the high-risk areas, yielding the classification results.

[0071] J represents the sum of squared errors of the clustering, k represents the number of clusters, C_i represents the i-th cluster, x represents the data points in the cluster, and μ_i represents the centroid of the i-th cluster. This formula measures the compactness of the clustering results; a smaller value indicates better clustering. Cleanliness assessment indicators are extracted from the classification results to generate a cleanliness assessment layer. For the cleanliness assessment layer, a convolutional neural network is used to analyze its features, obtaining feature distribution data. If the feature distribution data meets preset conditions, the final output is generated through data processing. Based on the final output, a visualized cleanliness assessment layer is generated to determine the cleanliness level.

[0072] Specifically, the residue distribution density is extracted from the refined target contour image. First, image processing algorithms are used to perform grayscale conversion and edge detection on the contour image. The Canny edge detection algorithm is employed, with a low threshold of 50 and a high threshold of 150 to extract residue boundaries. Next, the average grayscale value of each pixel within a 5x5 pixel neighborhood is calculated to generate a density distribution map, with density values ​​ranging from 0 to 255. Areas with density values ​​greater than 180 are considered high-density residue areas and marked as initial high-risk areas. Based on this, a cleanliness assessment layer is designed, using k... means A clustering algorithm with k=3 categorizes density distribution into low, medium, and high-density groups. High-density groups with a mean >180 are directly marked as high-risk areas, generating a binary mask image where high-risk areas are assigned a value of 1, and other areas are assigned a value of 0. To ensure accuracy, considering business scenarios such as semiconductor wafer cleaning, residual chemical composition analysis is introduced. Assuming that residual types are identified using infrared spectroscopy data, a threshold of 0.8 for the absorption peak intensity of organic residues is set. If a region has a density >180 and an absorption peak intensity >0.8, it is confirmed as a high-risk area, and a cleanliness assessment layer is output. If the proportion of high-risk areas in the assessment layer exceeds 5%, cleaning process optimization is triggered, generating optimization parameters, such as increasing the cleaning time by 10 seconds or increasing the cleaning solution concentration by 0.2%. Through these steps, from density extraction to risk labeling to assessment layer generation, a closed-loop logic is formed, ensuring the accuracy and operability of the cleanliness assessment.

[0073] S108. Based on the cleanliness assessment layer, a vector thematic map is generated by overlaying facility location information, and multi-layer data is integrated to determine the environmental governance reference output.

[0074] Environmental cleanliness assessment data and facility location information are acquired from data sources and stored as a structured dataset to obtain an initial dataset. Spatial analysis is performed on this initial dataset using Geographic Information System (GIS) tools to calculate the spatial relationship between cleanliness assessment data and facility locations, determining a spatially correlated dataset. If the cleanliness value in the spatially correlated dataset is lower than a preset threshold, the facility locations in the corresponding areas are marked, resulting in a marked location dataset. Based on the marked location dataset, vector data processing methods are used to generate vector thematic maps, obtaining visualized thematic map data. A multi-layer data fusion algorithm is used to combine the vector thematic map data and the environmental cleanliness assessment data to generate comprehensive environmental data, resulting in a fused dataset. For the fused dataset, a decision tree algorithm is used to analyze environmental governance priorities and determine the ranking of governance schemes. Based on the ranking of governance schemes, an environmental governance reference output is generated, resulting in the final governance recommendation dataset.

[0075] Specifically, the implementation method of generating vector thematic maps based on cleanliness assessment layers and integrating multi-layer data to determine environmental governance reference outputs can be achieved through geographic information systems and data analysis techniques. First, assuming that the cleanliness assessment layer is based on the Air Quality Index, taking data from 100 monitoring points in a city as an example, the AQI value ranges from 0 to 300, divided into excellent (0-50), good (51-100), lightly polluted (101-150), moderately polluted (151-200), and heavily polluted (201-300).

[0076] Using the Kriging interpolation algorithm, spatial interpolation is performed on the AQI data of the monitoring points to generate a gridded cleanliness assessment layer with a resolution of 500 meters × 500 meters. The calculation formula is as follows:

[0077] in Given the AQI value of a point, The weights are determined by the distance and the covariance function. Assume the covariance model is a Gaussian model with a parameter range of a = 2 kilometers.

[0078] Next, facility location information, such as the coordinates of sewage treatment plants and waste treatment stations, is overlaid. Assuming the city has 10 sewage treatment plants (coordinates in WGS84 latitude and longitude, e.g., [120.15, 30.25]) and 15 waste treatment stations, the facility data is stored in the GIS database as a point vector layer. Using buffer analysis, the influence radius of the sewage treatment plants is set to 3 kilometers, and that of the waste treatment stations to 1 kilometer, generating a buffer vector layer. The calculation formula is D = √((x - xi)). 2 +(y-yi) 2 ), where D is the distance from the point to the facility, and xi and yi are the facility coordinates. The cleanliness assessment layer and the facility buffer layer are merged, and a weighted overlay algorithm is used, with a weight allocation of 70% for cleanliness and 30% for facility impact, to generate a comprehensive environmental index map. The index formula is:

[0079] The normalized AQI value (0-1). The facility impact factor is set (1 within the buffer zone, 0 outside). Finally, a vector thematic map is generated through classification rendering, with 5 color zones set (excellent: green, good: yellow, lightly polluted: orange, moderately polluted: red, heavily polluted: purple). Based on areas with a comprehensive index E>0.8, environmental governance priority suggestions are output, such as suggesting the addition of purification facilities in areas with E>0.8. Specific coordinates are extracted through spatial query, such as [120.20, 30.30]. The entire process is automated through the Python API of ArcGIS or QGIS, ensuring efficient and reproducible data processing. The generated environmental governance reference output can be directly used for urban planning.

[0080] Based on the labeled location dataset, vector data processing methods are used to generate vector thematic maps, resulting in visualized thematic map data.

[0081] Vector data, including spatial coordinates and attribute information, is obtained from a labeled location dataset. A spatial interpolation algorithm is used to generate a vector thematic map based on this vector data, yielding preliminary thematic map data. Topological analysis is performed on the preliminary thematic map data to determine the spatial relationships between vector features. Boundary overlap and adjacency attributes are extracted from these spatial relationships to generate a spatial association dataset. A clustering algorithm is used on the spatial association dataset to obtain classified thematic map features. Multi-scale rasterization is performed on these thematic map features to generate multi-resolution raster data. Color distribution attributes are extracted from the multi-resolution raster data to determine visualization rendering parameters. Dynamic rendering is performed based on these visualization rendering parameters to generate the final visualized thematic map data. Solution: Vector data, including spatial coordinates and attribute fields, is extracted from a labeled location dataset. Kriging interpolation is used to process the vector data, generating continuous vector thematic map data. Topological checks are performed on the vector thematic map data to determine the adjacency and overlap relationships between features. Spatial association features are extracted from these adjacency and overlap relationships to generate a spatial weight matrix. K-means clustering is used on the spatial weight matrix to obtain classified thematic map partition data. Adaptive rasterization is performed on the partitioned data to generate a multi-scale raster dataset. Color gradient attributes are extracted from the multi-scale raster dataset to determine layered rendering parameters. Dynamic visualization rendering is then performed using these layered rendering parameters to generate high-resolution thematic map data.

[0082] In the description of this invention, it should be understood that the terms "coaxial," "bottom," "one end," "top," "middle," "other end," "upper," "side," "top," "inner," "front," "center," "both ends," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0083] Furthermore, the terms “first,” “second,” “third,” and “fourth” are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as “first,” “second,” “third,” or “fourth” may explicitly or implicitly include at least one of those features.

[0084] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "setting," "connection," "fixing," "screw connection," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal connection of two components or the interaction between two components. Unless otherwise explicitly limited, those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0085] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for generating thematic maps of large targets such as ports and bridges based on SAR technology, characterized in that, include: Acquire multi-temporal synthetic aperture radar image data of the port area, and obtain a smoothed scattering intensity map by denoising and filtering the original signal intensity distribution of the image data. Based on the scattering intensity map, the echo amplitude and phase information of the target area are extracted, and the multi-polarization channel data are fused to determine the scattering characteristic parameter set. The scattering characteristic parameter set is grouped based on a clustering algorithm, and the preliminary clustering results of facility class and residue class are separated to obtain feature vector representation; Based on the pre-established feature database, the feature vector representation is matched with the stored shape template. If the similarity is higher than a preset threshold, the target type is confirmed and a classification label is obtained. Based on the classification labels, combined with the scattering characteristic parameter set and shape analysis results, a multi-dimensional attribute description is fused to obtain the target's precise spatial coordinates; Based on the precise spatial coordinates, and by optimizing boundary detection through a convolutional neural network, edge obfuscation in the image is processed to obtain the target contour map of the refinement; Based on the refined target contour map, the residue distribution density is extracted. If the density exceeds a preset threshold, high-risk areas are marked to obtain a cleanliness assessment layer. Based on the cleanliness assessment layer, a vector thematic map is generated by overlaying facility location information, and multi-layer data is integrated to determine the reference output for environmental governance.

2. The method for generating large-scale thematic maps of ports and bridges based on SAR technology according to claim 1, characterized in that: The step of acquiring multi-temporal synthetic aperture radar image data of the port area and processing the original signal intensity distribution of the image data through denoising filtering to obtain a smoothed scattering intensity map includes: Acquire multi-temporal synthetic aperture radar image data of the port area, and separate the original signal intensity through data preprocessing; The original signal strength is processed by adaptive denoising filtering, and a smooth signal strength distribution is generated. Based on the spatial domain analysis, the signal intensity distribution is smoothed to obtain scattering intensity characteristics. If the scattering intensity characteristics meet a preset threshold, the multi-temporal variation trend is extracted through time series analysis to obtain time series scattering characteristics. Based on the temporal scattering characteristics, a convolutional neural network is used to classify the land cover types in the port area and generate classification results. Based on the superimposed classification results and the smoothed signal intensity distribution, a scattering intensity map of the port area is generated; Based on the scattering intensity map, and by optimizing the boundary smoothness through mean filtering, the final scattering intensity map is obtained.

3. The method for generating large-scale thematic maps of ports and bridges based on SAR technology according to claim 1, characterized in that: The step of extracting the echo amplitude and phase information of the target region based on the scattering intensity map, fusing multi-polarization channel data, and determining the scattering characteristic parameter set includes: Based on the original data of the target region obtained from the scattering intensity map, the boundary of the target region is determined by a preset region segmentation algorithm to obtain the pixel set of the target region. Based on the pixel set of the target area, the echo amplitude data is extracted, and the amplitude value of each pixel is calculated using the Fast Fourier Transform algorithm to obtain the echo amplitude distribution. Based on the pixel set of the target region, phase data is extracted, and the phase information of each pixel is processed using a phase unwrapping algorithm to obtain the phase data distribution; The phase unwrapping algorithm is configured as follows: in, Represented as the true phase value after untangling; This is represented by the observed entanglement phase value; Represented as at pixel The integer multiple to be added at the location Number of jumps; Based on the fused dataset, scattering characteristics are extracted using principal component analysis algorithm to obtain scattering characteristic parameters; Based on the scattering characteristic parameters, the correlation coefficient between each parameter is calculated. If the correlation coefficient is higher than a preset threshold, the corresponding parameter is retained to obtain the final scattering characteristic parameter set. By normalizing the final scattering characteristic parameter set, a standardized parameter set is generated, thus obtaining a description of the scattering characteristics of the target region.

4. The method for generating large-scale thematic maps of ports and bridges based on SAR technology according to claim 1, characterized in that: The step of grouping the scattering characteristic parameter set based on a clustering algorithm, separating the facility class and the residue class from the preliminary clustering results, and obtaining the feature vector representation includes: The scattering characteristic parameter set is grouped based on the K-means clustering algorithm to obtain preliminary clustering results and feature vector representations for facility categories and residue categories. If the intra-cluster variance of the preliminary clustering results is greater than a preset threshold, the cluster centers are adjusted through iterative optimization to obtain optimized clustering results. Based on the optimized clustering results, feature vectors of each cluster are extracted to obtain feature vector representations of facility categories and residue categories; Principal component analysis is used to reduce the dimensionality of the eigenvectors to obtain the dimensionality-reduced eigenvectors. If the dimensionality of the eigenvectors after dimensionality reduction is lower than a preset threshold, linear discriminant analysis is used to further separate the facility category and the residue category to obtain the separated eigenvectors. Based on the separated feature vectors, the inter-cluster distance is calculated, the accuracy of category differentiation is judged, and the final classification result is obtained. Based on the final classification results, feature vector representations of facility categories and residue categories are obtained.

5. The method for generating large-scale target thematic maps of ports and bridges based on SAR technology according to claim 1, characterized in that: The step of matching the feature vector representation with the stored shape template based on a pre-established feature database, and confirming the target type and obtaining the classification label if the similarity is higher than a preset threshold, includes: Obtain the template shape set and feature vector set based on the preset feature database; The similarity score between the target feature vector and the template shape is calculated based on the cosine similarity algorithm. If the similarity score exceeds the preset threshold, the target matching template is determined and a preliminary classification label is obtained. The initial classification labels are verified based on the cluster analysis, and optimized classification labels are obtained. Based on the optimized classification labels, the corresponding target type is extracted from the feature database. If the target type matches the preset classification rule, the final classification label is obtained. The feature database is updated based on the final classification labels to obtain the optimized template shape set.

6. The method for generating large-scale target thematic maps of ports and bridges based on SAR technology according to claim 1, characterized in that: The step of fusing the classification labels, the scattering characteristic parameter set, and the shape analysis results into a multi-dimensional attribute description to obtain the precise spatial coordinates of the target includes: Extract target-related tag data from a pre-set database, and use a random forest algorithm to initially filter the tags to obtain a set of categorized tags; Based on the classification label set, a scattering characteristic parameter set is obtained from the sensor data, and the parameters are dimensionality reduced by principal component analysis to obtain a dimensionality-reduced scattering feature set. Based on the reduced scattering feature set, shape analysis results are obtained, and the target shape is extracted using a convolutional neural network to obtain a shape feature description. The scattering feature set and shape feature description after dimensionality reduction are fused. If the fused feature vector meets the preset similarity threshold, a multi-dimensional attribute description is generated. Based on the multi-dimensional attribute description, and by iteratively calculating the spatial position of the target using the Kalman filter algorithm, preliminary spatial coordinates are obtained; Based on the preliminary spatial coordinates and combined with the preset coordinate correction model, if the coordinate deviation is less than the preset threshold, the coordinates are optimized and adjusted to determine the final spatial coordinates. The accuracy of target positioning is determined by using the final spatial coordinates and verifying the distance between the coordinates and the preset reference point using the Euclidean distance calculation method.

7. The method for generating large-scale thematic maps of ports and bridges based on SAR technology according to claim 1, characterized in that: The step of obtaining the target contour map of the image by optimizing boundary detection and processing edge obfuscation in the image based on the precise spatial coordinates includes: Based on the raw image data obtained from the image data input, the image is denoised and standardized using preprocessing techniques to obtain the first image data; Based on the first image data, feature extraction is performed using a convolutional neural network to obtain a feature map containing edge information; Based on the edge information in the feature map, if the edge pixel intensity is lower than a preset threshold, the edge clarity is enhanced by a boundary optimization algorithm to obtain a second feature map. Based on the second feature map and through spatial location analysis technology, the target region corresponding to the precise spatial coordinates is obtained, and the first target region is generated. Based on the first target region, and by separating the target region from the background using target region segmentation technology, a second target region is obtained; Based on the second target region, a refined target contour map is generated by applying contour generation technology. Based on the refined target contour map, and through edge sharpness enhancement technology, the contour boundary is further optimized to obtain the final target contour map.

8. The method for generating large-scale target thematic maps of ports and bridges based on SAR technology according to claim 1, characterized in that: The step of extracting the residue distribution density based on the refined target contour map, and marking high-risk areas if the density exceeds a preset threshold to obtain a cleanliness assessment layer includes: Based on the target contour map, obtain the residue distribution data, use image segmentation algorithm to determine the residue area and obtain the distribution density. If the distribution density exceeds the preset threshold, generate high-risk areas through marking process to obtain preliminary area division. Based on the preliminary regional division, high-risk areas are classified using a clustering algorithm to obtain classification results; The clustering algorithm for classifying high-risk areas is configured as follows: in, This is represented as the sum of squared errors of the entire cluster; This is represented by the number of clusters in the clustering; Represented as the first A cluster; Represented as data points within a cluster; Represented as the first The center point of each cluster; Based on the cleanliness assessment layer, a convolutional neural network is used to analyze the features of the assessment layer to obtain feature distribution data. If the feature distribution data meets the preset conditions, the final result is generated and output through data processing. Based on the final output, a visual cleanliness assessment layer is generated to obtain the final cleanliness level.

9. The method for generating large-scale target thematic maps of ports and bridges based on SAR technology according to claim 1, characterized in that: The steps of generating a vector thematic map by overlaying facility location information based on the cleanliness assessment layer, fusing multi-layer data, and determining the environmental governance reference output include: Environmental cleanliness assessment data and facility location information are obtained from the data source, stored as a structured dataset, and an initial dataset is obtained. Spatial analysis is performed based on the initial dataset. Geographic Information System (GIS) tools are used to calculate the spatial relationship between cleanliness assessment data and facility locations to obtain a spatial association dataset. If the cleanliness value in the spatial association dataset is lower than a preset threshold, the facility locations in the corresponding areas are marked to obtain a marked location dataset. Based on the labeled location dataset, vector data processing methods are used to generate vector thematic maps, resulting in visualized thematic map data; Based on a multi-layer data fusion algorithm, combined with vector thematic map data and environmental cleanliness assessment data, comprehensive environmental data is generated to obtain a fused dataset; Based on the fused dataset, a decision tree algorithm is used to analyze the priorities of environmental governance and obtain a ranking of governance schemes.

10. A method for generating large-scale thematic maps of ports and bridges based on SAR technology according to claim 9, characterized in that: The step of generating a vector thematic map and obtaining visualized thematic map data based on the labeled location dataset using vector data processing methods includes: Vector data is obtained from the marked location dataset, and the vector data includes spatial coordinates and attribute information. Based on the vector data, a spatial interpolation algorithm is used to generate a vector thematic map, thereby obtaining preliminary thematic map data; Topological analysis is performed based on the preliminary thematic map data to obtain the spatial relationships between vector elements; Boundary overlap and adjacency attributes are extracted from the spatial relationships to generate a spatial association dataset; Based on the spatial association dataset, a clustering algorithm is used to obtain the features of the classified thematic map. Multi-scale rasterization processing is performed based on the thematic map features to generate multi-resolution raster data; Based on the multi-resolution raster data, and extracting color distribution attributes from the multi-resolution raster data, the visualization rendering parameters are determined; Dynamic rendering is performed based on the visualization rendering parameters to obtain the final visualized thematic map data; Extract vector data from the labeled location dataset, the vector data including spatial coordinates and attribute fields; The vector data is processed using the Kriging interpolation algorithm to generate continuous vector thematic map data; A topology check is performed on the vector thematic map data to obtain the adjacency and overlap relationships between features; Based on the adjacency and overlap relationships, spatial correlation features are extracted from the adjacency and overlap relationships to obtain a spatial weight matrix; Based on the spatial weight matrix, the K-means clustering algorithm is used to obtain the classified thematic map partition data. Adaptive rasterization processing is performed on the partitioned data to obtain a multi-scale raster dataset. Color gradient attributes are extracted from the multi-scale raster dataset to obtain layered rendering parameters; Dynamic visualization rendering is performed based on the layered rendering parameters to obtain high-resolution thematic map data.

Images