Coastline segmentation extraction method and system based on remote sensing data

Abstract

The invention relates to the technical field of coastline segmentation and extraction, and discloses a coastline segmentation and extraction method and system based on remote sensing data. Acquiring multi-source remote sensing image data of the target coastal zone area; constructing a dual-path U-Net model, wherein the dual-path U-Net model comprises a natural shoreline coding path and an artificial shoreline coding path; fusing the output features of the natural shoreline coding path and the artificial shoreline coding path through a dynamic weight fusion module to generate a fused feature map; and performing sea-land segmentation and morphological optimization based on the fused feature map, and performing segmentation extraction on the coastline. Differentiation features of a natural shoreline and an artificial shoreline are respectively processed by constructing a dual-path U-Net model, and adaptive feature fusion of multi-source remote sensing data is realized in combination with a dynamic weight fusion module. Through end-to-end sea-land segmentation and morphological optimization processes, automatic and efficient extraction from multi-source remote sensing data to a coastline vectorization result is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of coastline segmentation and extraction, and particularly to a method and system for coastline segmentation and extraction based on remote sensing data. Background Art

[0002] The coastal zone is the zone where land and sea interact. It is both the "golden zone" for economic development and the "fragile area" of the ecological environment. The coastline is the core division basis and an important part of the coastal zone. Its complexity, activity and marginality make it a key element in marine surveys.

[0003] From the perspective of morphological characteristics, the coastline can be divided into two major types: natural coastline and artificial coastline. The natural shoreline is usually composed of beaches, mudflats, reefs or biological communities (such as mangroves), and its boundary shows the characteristics of gradual change and fuzziness. For example, the alternation of the exposed and submerged states of the intertidal zone during the ebb and flow of tides results in the natural shoreline showing weak edge features with gentle gray-scale transition and poor texture continuity in remote sensing images. The artificial shoreline includes hardened structures formed by human activities such as breakwaters, docks, and reclamation projects. Such shorelines mostly show strong edge features with high contrast and regular geometric shapes in images. The significant differences between the two types of shorelines in terms of spatial distribution, boundary clarity and spectral response. The semantic segmentation model based on deep learning (such as U-Net) has improved the automation level of shoreline extraction to a certain extent through multi-level feature learning. However, it fails to make adaptive adjustments for the morphological differences of shoreline types. When optimizing for a certain type of shoreline, the extraction accuracy of the other type of shoreline often drops significantly, unable to meet the extraction accuracy requirements for weak boundaries (natural shorelines) and strong boundaries (artificial shorelines).

[0004] In addition, the complex lighting conditions and ground object coverage in the coastal zone further increase the segmentation difficulty. The shadows in the intertidal zone (such as cliff occlusion, cloud projection) have similar spectral characteristics to water bodies in the visible light band. Traditional methods are prone to misjudging shadows as water bodies, causing the shoreline to shift landward. Although SAR images have the advantage of all-weather observation, their unique speckle noise and geometric deformation problems make the segmentation results of a single data source lack robustness.

[0005] In summary, the existing coastline segmentation methods still have significant bottlenecks in dealing with the morphological differences between natural and artificial shorelines, multi-source information fusion, etc. How to construct a segmentation model that can adapt to shoreline types, take into account context information and detail retention, and effectively fuse multi-source remote sensing data to improve the ability to distinguish shadows and water bodies has become an urgent problem to be solved in the current segmentation and extraction of the coastal zone. Summary of the Invention

[0006] To solve the above technical problems, the present invention provides a method for segmenting and extracting a coastline based on remote sensing data, including the following steps: Step S1: Obtain multi-source remote sensing image data of the target coastal zone area; Step S2: Construct a dual-path U-Net model, where the dual-path U-Net model includes a natural coastline encoding path and an artificial coastline encoding path; Step S3: The natural coastline encoding path and the artificial coastline encoding path obtain a natural coastline feature map and an artificial coastline feature map according to the multi-source remote sensing image data; Step S4: Fuse the natural coastline feature map and the artificial coastline feature map through a dynamic weight fusion module to generate a fused feature map; Step S4: Based on the fused feature map, perform land-sea segmentation and morphological optimization to segment and extract the coastline.

[0007] Further, the dual-path U-Net model in step S2 includes: The natural coastline encoding path uses a convolutional kernel to extract the edge features of the natural coastline and obtains a natural coastline feature map; The artificial coastline encoding path uses depthwise separable convolution to extract the edge features of the artificial coastline and obtains an artificial coastline feature map; Transfer the natural coastline feature map and the artificial coastline feature map to the decoder through skip connections.

[0008] Further, in step S4, fusing the natural coastline feature map and the artificial coastline feature map through a dynamic weight fusion module to generate a fused feature map includes: Concatenate the natural coastline feature map output by the natural coastline encoding path and the artificial coastline feature map output by the artificial coastline encoding path along the channel dimension; Perform spatial attention calculation and channel attention calculation on the concatenated feature map respectively to obtain a spatial attention weight and a channel attention weight; Generate a dynamic weight α based on the spatial attention weight and the channel attention weight; Perform weighted fusion on the natural coastline feature map and the artificial coastline feature map according to the dynamic weight α to generate the fused feature map.

[0009] Further, the spatial attention calculation uses dilated convolution to extract context information to obtain a spatial attention weight; The channel attention calculation generates a channel attention weight through global average pooling and max pooling.

[0010] Further, the calculation formula of the dynamic weight α is: ; In the formula, W s (i,j) is the spatial attention weight at the spatial position in the spliced feature map (i,j) ; W c (i,j) is the channel attention weight at the spatial position in the spliced feature map, and H and W are the height and width of the fused feature map respectively. (i,j)

[0011] Furthermore, in step S5, performing land-sea segmentation based on the fused feature map includes: Step S5A1: Performing multi-modal feature splicing on the fused feature map and the multi-source remote sensing image data to obtain multi-modal input data; Step S5A2: Inputting the multi-modal input data into a cascaded ensemble learning module to obtain a natural shoreline classification result and an artificial shoreline classification result; Step S5A3: Performing spatial splicing on the natural shoreline classification result and the artificial shoreline classification result to generate an initial coastline.

[0012] Furthermore, in step S5A1, performing multi-modal feature splicing on the fused feature map and the multi-source remote sensing image data to obtain multi-modal input data includes: Step S5A11: Obtaining a normalized difference water index according to the green light band and the near-infrared band in the multi-source remote sensing image data; Step S5A12: Extracting the VV polarization data of the synthetic aperture radar data in the multi-source remote sensing image data; Step S5A13: Splicing the normalized difference water index, the VV polarization data, and the fused feature map along the channel dimension to obtain the multi-modal input data.

[0013] Furthermore, when inputting the multi-modal input data into a cascaded ensemble learning module in step S5A2 to obtain a natural shoreline classification result and an artificial shoreline classification result, performing region-based classification when inputting the multi-modal input data into the cascaded ensemble learning module specifically includes: For the natural shoreline area, using the random forest classifier in the cascaded ensemble learning module to perform threshold segmentation on the normalized difference water index and the near-infrared band to obtain the natural shoreline classification result; For the artificial shoreline area, using the support vector machine classifier in the cascaded ensemble learning module to perform edge enhancement on the VV polarization data and the red light band to obtain the artificial shoreline classification result.

[0014] Furthermore, the morphological optimization includes: ​Step S5B1: Denoise the initial coastline through opening operation; Step S5B2: Perform closing operation on the denoised initial coastline to obtain an optimized coastline; Step S5B3: Perform boundary tracking based on the optimized coastline to obtain the final coastline.

[0015] On the other hand, the present invention also provides a coastline segmentation and extraction system based on remote sensing data, which executes the coastline segmentation and extraction method based on remote sensing data described in any one of the above. The system includes: a data acquisition module, a model construction module, a fused feature map generation module, and a result output module; The data acquisition module is used to acquire multi-source remote sensing image data of the target coastal zone area; The model construction module is used to construct a dual-path U-Net model, and the dual-path U-Net model includes a natural shoreline encoding path and an artificial shoreline encoding path; The fused feature map generation module is used to fuse the output features of the natural shoreline encoding path and the artificial shoreline encoding path through a dynamic weight fusion module to generate a fused feature map; The result output module is used to perform land-sea segmentation and morphological optimization based on the fused feature map, and segment and extract the coastline.

[0016] The embodiments of the present invention have the following technical effects: The coastline segmentation and extraction method based on remote sensing data provided by the present invention processes the differential features of natural shorelines and artificial shorelines by constructing a dual-path U-Net model, and combines a dynamic weight fusion module to achieve adaptive feature fusion of multi-source remote sensing data. Finally, it effectively solves the technical problem that it is difficult to balance the weak boundary blur of natural shorelines and the strong boundary features of artificial shorelines in coastline segmentation. Through the independently designed natural shoreline encoding path and artificial shoreline encoding path, it is possible to perform differential feature extraction for the gradual blur characteristics of natural shorelines and the regular hardening structure of artificial shorelines, overcoming the defect of poor adaptability of traditional single models to different types of shoreline features; using a dynamic weight fusion module to perform adaptive weighted fusion on the dual-path features not only retains the context continuity features of natural shorelines but also enhances the edge sharpness features of artificial shorelines, significantly improving the overall segmentation accuracy of mixed shorelines in complex coastal zone environments; through an end-to-end land-sea segmentation and morphological optimization process, it effectively suppresses tidal zone shadow interference and multi-source data noise, and realizes the automatic and efficient extraction from multi-source remote sensing data to the vectorized result of the coastline, providing high-precision technical support for coastal zone monitoring. Description of the Drawings

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0018] Figure 1 is the step flowchart of a coastline segmentation and extraction method based on remote sensing data provided by an embodiment of the present invention; Figure 2 is the module schematic diagram of a coastline segmentation and extraction system based on remote sensing data provided by an embodiment of the present invention; Figure 3 is the comparison chart of the coastline segmentation results of the single-path U-Net and the dual-path U-Net for the coastal area provided by an embodiment of the present invention. Specific Embodiments

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope protected by the present invention.

[0020] Based on the problems of the existing coastline segmentation methods in dealing with the morphological differences between natural shorelines and artificial shorelines, multi-source information fusion, etc., the present invention provides a coastline segmentation and extraction method based on remote sensing data, as Figure 1 shown, including the following steps: Step S1: Obtain multi-source remote sensing image data of the target coastal zone area; Step S2: Construct a dual-path U-Net model, and the dual-path U-Net model includes a natural shoreline encoding path and an artificial shoreline encoding path; In some embodiments, the dual-path U-Net model in Step S2 includes: Step S21: The natural shoreline encoding path uses a convolution kernel to extract the edge features of the natural shoreline and obtain a natural shoreline feature map.

[0021] Exemplarily, a standard convolution kernel is used to construct the natural shoreline encoding path. Its larger receptive field is beneficial to capturing the context-related features of natural landforms such as beaches and mudflats. When the three-dimensional convolution kernel slides in the spatial dimension, it gradually abstracts the gradual change features of the intertidal zone through multi-level downsampling, and uses the semantic information contained in the deep features to identify low-contrast boundaries.

[0022] Step S22: The artificial shoreline encoding path uses depthwise separable convolution to extract the edge features of the artificial shoreline, obtaining an artificial shoreline feature map; Exemplarily, the artificial shoreline encoding path selects depthwise separable convolution, decomposing the standard convolution into two steps: depth convolution and pointwise convolution. While ensuring the feature extraction ability, it significantly reduces the number of parameters. This lightweight design is particularly suitable for extracting the regular edge features of artificial structures such as breakwaters and docks. By enhancing the frequency of local convolution operations, detailed information of high-contrast edges can be captured in the shallow network.

[0023] Step S23: The natural shoreline feature map and the artificial shoreline feature map are transmitted to the decoder through skip connections.

[0024] Exemplarily, the feature maps output by the two encoding paths are directly transmitted to the decoder through skip connections, maintaining the spatial correspondence of the original features before feature fusion. During the process of the decoder gradually restoring the spatial resolution through upsampling, it synchronously receives the skip connection features from the dual paths to achieve complementary enhancement of multi-scale features. This dual-stream architecture enables the model to process the significantly different shoreline type features in parallel, avoiding the feature compromise problem of the single-path model when dealing with mixed shorelines.

[0025] There are significant differences in the morphology, spectrum, and texture features between natural shorelines and artificial shorelines: Natural shorelines (such as beaches, mudflats, and mangroves) usually exhibit weak edge features that are gradual and blurred, and are greatly affected by tides, light, and shadows. The model needs to have stronger context awareness ability to capture continuous but low-contrast boundaries. Artificial shorelines (such as docks, breakwaters, and reclamation projects) have strong edge features with high contrast and geometric regularity, and the model needs to be able to accurately extract sharp contours and avoid edge blurring caused by traditional convolution operations. If a single-path U-Net with a shared encoder is used, the model will forcefully learn the compromise features of the two types of shorelines, resulting in the loss of details of the weak boundaries of natural shorelines due to over-smoothing; the strong edges of artificial shorelines become blurred due to feature confusion. Based on this, the present invention constructs a dual-path U-Net model. Through their respective independent encoding paths: the natural shoreline encoding path uses standard convolution kernels to enhance the extraction of context information and adapt to fuzzy boundaries; the artificial shoreline encoding path uses depthwise separable convolution to reduce parameter redundancy and focus on local edge enhancement. Finally, the contributions of the two types of features are adaptively adjusted through dynamic weight fusion to ensure that the segmentation result not only retains the continuity of natural shorelines but also maintains the clarity of artificial shorelines.

[0026] Exemplarily, during the implementation process of the present invention, the coastline segmentation results of the single-path U-Net and the dual-path U-Net for coastal areas are compared. Among them, the single-path U-Net is set as a shared encoder with a unified convolutional kernel; the dual-path U-Net is set as an independent encoding path + dynamic weight fusion. The intersection over union (IoU) and the edge accuracy (F1-score) are used as evaluation indicators: the intersection over union (IoU) is used to measure the overlap degree of the overall segmentation area; the edge accuracy (F1-score) is used to specifically evaluate the accuracy of the shoreline boundary pixels (natural shorelines rely more on continuity, and artificial shorelines rely more on sharpness).

[0027] It can be clearly seen from Figure 3 Obviously, in the segmentation and extraction results of the natural shoreline by the dual-path U-Net, the IoU is increased by 18% and the F1-score is increased by 21%. The dual-path model can better capture the gradual change features through context-aware convolution, and the dynamic weight fusion effectively enhances the continuity of weak edges and reduces breaks. In the segmentation and extraction results of the artificial shoreline by the dual-path U-Net, the IoU is increased by 14% and the F1-score is increased by 15%. It can be seen that the depthwise separable convolution can accurately extract regularized structures; the edge sharpness is significantly improved, avoiding the blurring effect of traditional convolutions. Therefore, the independent dual-path design is superior to the single-path in overall indicators and has better generalization ability.

[0028] Step S3: The natural shoreline encoding path and the artificial shoreline encoding path obtain a natural shoreline feature map and an artificial shoreline feature map according to the multi-source remote sensing image data; Through the independently designed natural shoreline encoding path and artificial shoreline encoding path, it is possible to perform differential feature extraction for the gradual change and blurring characteristics of natural shorelines and the regular hardening structures of artificial shorelines, overcoming the defect that traditional single models have poor adaptability to different types of shoreline features.

[0029] Step S4: The dynamic weight fusion module fuses the natural shoreline feature map and the artificial shoreline feature map to generate a fused feature map; including: Step S41: Concatenate the natural shoreline feature map output by the natural shoreline encoding path and the artificial shoreline feature map output by the artificial shoreline encoding path along the channel dimension; Step S42: Perform spatial attention calculation and channel attention calculation on the concatenated feature map respectively to obtain spatial attention weights and channel attention weights; In some embodiments, the spatial attention calculation uses dilated convolution to extract context information to obtain spatial attention weights; Exemplarily, the spatial attention weight calculation adopts dilated convolution operations. Convolution kernels with different dilation rates are set to process the input features in parallel, capturing local details and long-range context information respectively. After element-wise addition of the output features of each branch, a spatial weight map is generated through the Sigmoid activation function. This design enables the model to adaptively focus on key regions at different scales. Preferably, a dilation rate greater than 1 is set to expand the receptive field of the convolution kernel, capturing a larger range of context information without increasing the computational load. This design is particularly suitable for processing the spatial continuity features of natural shoreline areas, enhancing the recognizability of weak edges by perceiving the relevance of surrounding pixels.

[0030] Channel attention calculation generates channel attention weights through global average pooling and max pooling.

[0031] Exemplarily, channel attention calculation synchronously implements global average pooling and max pooling operations. Global average pooling captures the overall distribution characteristics of the feature channels, and max pooling extracts the significant response regions of the feature map. After concatenating the results of the two pooling operations along the channel dimension, a two-layer fully connected network is used to learn the non-linear relationships between channels, and a squeeze-and-excitation network structure is used to generate channel attention weights.

[0032] This dual-attention mechanism effectively overcomes the limitations of a single attention model. Spatial attention focuses on the spatial distribution characteristics of features, and channel attention optimizes the information utilization rate of feature channels. The two work together to enhance the discriminative ability of feature fusion, effectively identifying the feature channels important for the current classification task, such as enhancing the feature channels with strong edge responses in the artificial shoreline feature map.

[0033] Step S43: Generate a dynamic weight α based on the spatial attention weight and the channel attention weight; In some embodiments, the calculation formula for the dynamic weight α is: ; In the formula, W s (i,j) is the spatial attention weight at the spatial position (i,j) in the concatenated feature map, W c (i,j) is the channel attention weight at the spatial position (i,j) in the concatenated feature map, and H and W are the height and width of the fused feature map respectively.

[0034] Spatial attention weights reflect the degree of importance of each pixel position for the current task. For example, the intertidal zone requires higher weights due to the spectral similarity between shadows and water bodies. Channel attention weights indicate the information contribution of different feature channels, such as the channel weight differences between the near-infrared band for water body identification and radar data for artificial structure detection. The two achieve attention coupling in the spatial and channel dimensions through pointwise multiplication, and then obtain the overall fusion coefficient through global averaging, ensuring that the α value contains both local feature response intensity information and global statistical characteristics. When processing the area dominated by natural shorelines, the spatial attention weights show a continuous distribution at the gradual boundaries, and the channel attention weights enhance the feature channels containing texture continuity, making the α value automatically biased towards natural shoreline features. When processing artificial shorelines, the spatial attention weights are concentrated at the sharp edges, and the relevant channel weights are increased, prompting the fused features to strengthen the artificial structure features. When the feature map contains gradual boundaries such as mangroves and beaches, the spatial attention captures the large-scale context, and the channel attention strengthens the low-contrast features, and α approaches 1, with a higher weight for the natural shoreline path; when the feature map contains sharp edges such as ports and dams, the channel attention suppresses the noise bands (such as shadows), and the spatial attention focuses on local details, and α approaches 0, with a higher weight for the artificial shoreline path.

[0035] Step S44: Weightedly fuse the natural shoreline feature map and the artificial shoreline feature map according to the dynamic weight α to generate a fused feature map.

[0036] According to the adaptive weight assignment of the input features, the segmentation accuracy of artificial and natural shorelines is significantly improved.

[0037] Step S5: Perform land-sea segmentation and morphological optimization based on the fused feature map to segment and extract the coastline.

[0038] Perform land-sea segmentation based on the fused feature map to accurately classify and identify artificial and natural shorelines, and achieve the morphological integrity of the coastline through morphological optimization, jointly realizing the accurate segmentation and extraction of the coastline.

[0039] In some embodiments, in step S5, performing land-sea segmentation based on the fused feature map includes: Step S5A1: Perform multi-modal feature stitching on the fused feature map and multi-source remote sensing image data to obtain multi-modal input data; In some embodiments, in step S5A1, performing multi-modal feature stitching on the fused feature map and multi-source remote sensing image data to obtain multi-modal input data includes: Step S5A11: Obtain the normalized difference water index according to the green band and the near-infrared band in the multi-source remote sensing image data; ; Among them, NDWI represents the Normalized Difference Water Index, Green represents the green light band, and NIR represents the near-infrared band.

[0040] Step S5A12: Extract the VV polarization data of the synthetic aperture radar data from the multi-source remote sensing image data; Step S5A13: Concatenate the Normalized Difference Water Index, VV polarization data, and the fused feature map along the channel dimension to obtain multi-modal input data.

[0041] Exemplarily, the multi-source remote sensing image data includes optical images, SAR image data, and vegetation indices. First, preprocess the multi-source remote sensing image data. For example, perform radiometric correction and geometric correction on the optical images; perform speckle noise suppression and terrain correction on the SAR images; register the multi-source data to the same coordinate system and resolution.

[0042] The optical images can include the green light band, red light band, blue light band, and near-infrared band, etc. The synthetic aperture radar (SAR) image data includes VV polarization and VH polarization data; the VV polarization data represents the backscattering signal of the ground objects collected by the radar system in the polarization mode of vertical transmission (V) and vertical reception (V) at the same time, and is used for edge enhancement in the artificial shoreline area. The VV polarization data has a strong response to artificial buildings. For example, the geometric structures (right angles, metal surfaces) of artificial shorelines (such as docks, breakwaters) will produce strong specular reflections. The echo signal intensity of the VV polarization for such targets is significantly higher than that of natural shorelines. Usually, the VV backscattering value of concrete dams is usually 10 - 15 dB higher than that of the adjacent water bodies; at the same time, it is also sensitive to water bodies. The calm water surface will reflect most of the radar waves, resulting in extremely weak VV signals (appearing as dark areas), while rough water bodies (such as waves) will enhance the scattering, which is used for comparison with natural shorelines. The VV echo intensity of mangroves or sandy beaches is between that of water bodies and artificial buildings. In the subsequent cascade integrated learning module, the support vector machine classifier uses the VV polarization data and the red light band for edge enhancement to obtain accurate artificial shoreline classification results.

[0043] The VH polarization data, that is, the radar transmits vertically and receives horizontally, is used to assist in distinguishing vegetation-covered areas. The vegetation index data is used to assist in distinguishing the vegetation-covered areas and bare surfaces in natural shorelines.

[0044] The vegetation index is normalized to obtain the Normalized Difference Vegetation Index (NDVI), which can be calculated using the following formula: ; RED represents the red light band. The Normalized Difference Vegetation Index (NDVI) can be used to distinguish vegetation-covered areas (such as mangroves, salt marshes) from bare surfaces to avoid misjudgment of the vegetation in natural shorelines.

[0045] Exemplarily, the process of concatenating multi-modal input data along the channel dimension is as follows: (1) Input alignment Resample the fused feature map and multi-source data to a unified resolution (e.g., 10m); (2) Channel-level concatenation ; where F multi represents the concatenated multi-modal input data. If the number of channels of F fused is C, then the number of channels after concatenation is C + 3 (NDWI, VV, and NDVI each occupy 1 channel); Concat represents the concatenation operation along the channel dimension (Channel Axis), stacking the feature maps of different modalities in the channel dimension while preserving the spatial resolution consistency (the height H and width W remain unchanged); F fused represents the fused feature map obtained through the dual-path U-Net, which is obtained by feature weighted fusion of the natural shoreline path and the artificial shoreline path, contains spatial context and edge detail information, and provides high-level semantic features (such as shoreline morphology and texture continuity); NDWI represents the Normalized Difference Water Index, VV represents VV polarization data, and NDVI represents the Normalized Difference Vegetation Index, which are respectively used to provide physical features (such as the spectral-scattering characteristics of water bodies, edges, and vegetation). NDWI suppresses the shadow misjudgment in optical images, VV compensates for the missing of optical data in cloudy and rainy weather, and NDVI avoids the confusion between vegetation-covered areas and shorelines.

[0046] Step S5A2: Input the multi-modal input data into the cascaded ensemble learning module to obtain the natural shoreline classification result and the artificial shoreline classification result; In some embodiments, in step S5A2, when inputting the multi-modal input data into the cascaded ensemble learning module to obtain the natural shoreline classification result and the artificial shoreline classification result, in order to input the multi-modal input data into the cascaded ensemble learning module, regional classification is performed, which specifically includes: For the natural shoreline area, use the random forest classifier in the cascaded ensemble learning module to perform threshold segmentation on the Normalized Difference Water Index and the near-infrared band to obtain the natural shoreline classification result; For the artificial shoreline area, use the support vector machine classifier in the cascaded ensemble learning module to perform edge enhancement on the VV polarization data and the red light band to obtain the artificial shoreline classification result.

[0047] Step S5A3: Perform spatial stitching on the natural shoreline classification result and the artificial shoreline classification result to generate the initial coastline.

[0048] Exemplarily, the classification result in step S5A2 is a binary mask. When performing spatial stitching, the initial coastline is generated through per-pixel logical operations. If a pixel is classified as both a natural shoreline and an artificial shoreline simultaneously, it can be marked according to a preset priority. The preset priority can be set in combination with the actual situation. For example, the priority of the artificial shoreline is higher than that of the natural shoreline to prevent artificial structures from being covered by the natural shoreline.

[0049] Through priority override and logical operations, the classification results of the two types of shorelines are merged into a conflict-free initial coastline, providing input for subsequent morphological optimization (denoising, closing) to ensure the integrity of artificial structures and the continuity of natural shorelines.

[0050] The cascaded ensemble learning module achieves accurate regional classification through classifier combination. For the natural shoreline area, the random forest classifier utilizes the ensemble voting mechanism of multiple decision trees and processes the non-linear relationship between the normalized water index and the near-infrared band by constructing multiple threshold segmentation planes. This algorithm has good adaptability to irregular decision boundaries in the feature space and can effectively handle the spectral gradual change characteristics of natural landforms such as beaches and mudflats. For the artificial shoreline area, a support vector machine classifier is used to map the VV polarization data and the red light band to a high-dimensional space through a kernel function and construct an optimal separation hyperplane. The edge enhancement operation strengthens the contour features of artificial structures through a gradient operator, enabling the support vector machine to accurately identify the geometric features of linear features such as breakwaters and docks. This regional classification strategy gives full play to the advantages of different algorithms. The random forest deals with the fuzzy classification problem of natural shorelines, and the support vector machine is suitable for identifying the clear boundaries of artificial shorelines. The two form a complete coastline through spatial stitching.

[0051] In some embodiments, the morphological optimization includes: Step S5B1: For the initial coastline, perform denoising processing through opening operations; used to eliminate small noises and burrs in the initial coastline.

[0052] Exemplarily, the initial coastline is a binary image, where the land pixel value is recorded as 0 and the water pixel value is recorded as 1. The denoising includes: first performing an erosion operation to eliminate isolated land pixels with an area smaller than the first threshold; then performing a dilation operation to restore the original shape of the remaining land areas; Step S5B2: Perform a closing operation on the initial coastline that has undergone denoising processing to obtain an optimized coastline, which is used to fill small holes and connect adjacent regions.

[0053] Exemplarily, a binary image after noise removal is processed using a closing operation, where: first perform a dilation operation to fill water holes with an area smaller than the second threshold; then perform an erosion operation to restore the original boundary of the water area.

[0054] The first threshold and the second threshold are dynamically adjusted according to the regional characteristics of the classification results of each shoreline in the initial shoreline: For the natural shoreline area, a relatively large first threshold is set to retain small feature characteristics; For the artificial shoreline area, a relatively small second threshold is set to ensure edge continuity.

[0055] Step S5B3: Based on the optimized shoreline, perform boundary tracking processing, convert the obtained closed polygon into shoreline vector data, and obtain the final shoreline.

[0056] In shoreline segmentation, the initial shoreline may have inaccurate parts due to interference factors in remote sensing images (such as waves, clouds, sensor noise, etc.). For example, wave foam in the intertidal zone may form isolated white pixel points in the image, which will be regarded as noise. The structure of the artificial shoreline may have breaks or holes due to reflection or occlusion.

[0057] The role of denoising through opening operation is to remove these small noise points, such as isolated pixels or small protrusions, making the shoreline smoother and reducing incorrect segmentation areas. For example, sporadic non-shoreline points may appear in the sandy beach area due to reflection, and the opening operation can eliminate these points to avoid misjudgment as shorelines in subsequent processing.

[0058] The purpose of closing operation is to fill small holes inside the shoreline caused by discontinuous segmentation, or to connect parts that are broken due to occlusion. For example, cloud occlusion may cause a certain section of the shoreline to be interrupted during segmentation. The closing operation can expand the shoreline area through dilation operation, and then erode to restore the approximate shape, thereby filling these interruptions and making the shoreline continuous.

[0059] The sequential execution of opening and closing operations constructs a dual guarantee of noise suppression - structure enhancement: Noise sensitivity difference: The opening operation is sensitive to positive noise (pseudo-increased pixels), and the closing operation focuses on negative noise (pseudo-decreased pixels), and the two form a complementary filtering system.

[0060] Edge conformal constraint: Compared with single erosion or dilation, the combination of opening and closing operations can maintain the shoreline curvature characteristics to the greatest extent during the denoising and filling processes, avoiding distortion phenomena such as straightening artificial shorelines or over-smoothing natural shorelines.

[0061] A method for segmenting and extracting coastline based on remote sensing data provided by the present invention constructs a dual-path U-Net model to process the differential features of natural coastline and artificial coastline respectively, and combines a dynamic weight fusion module to achieve adaptive feature fusion of multi-source remote sensing data, finally effectively solving the technical problem that it is difficult to balance the weak boundary fuzziness of natural coastline and the strong boundary features of artificial coastline in coastline segmentation. Through the independently designed natural coastline coding path and artificial coastline coding path, it can extract differential features for the gradual fuzzy characteristics of natural coastline and the regular hardening structure of artificial coastline, overcoming the defect of poor adaptability of traditional single models to different types of coastline features; the dynamic weight fusion module is used to adaptively weight and fuse the dual-path features, which not only retains the context continuity features of natural coastline, but also enhances the edge sharpness features of artificial coastline, significantly improving the overall segmentation accuracy of mixed coastline in complex coastal zone environment; through the end-to-end land-sea segmentation and morphological optimization process, it effectively suppresses the intertidal zone shadow interference and multi-source data noise, realizing the automatic and efficient extraction from multi-source remote sensing data to the vectorized result of coastline, providing high-precision technical support for coastal zone monitoring.

[0062] On the other hand, the present invention also provides a coastline segmentation and extraction system based on remote sensing data, which executes a method for segmenting and extracting coastline based on remote sensing data according to any one of the above, as Figure 2 shown, the system includes: a data acquisition module, a model construction module, a fused feature map generation module and a result output module; The data acquisition module is used to acquire multi-source remote sensing image data of the target coastal zone area; The model construction module is used to construct a dual-path U-Net model, and the dual-path U-Net model includes a natural coastline coding path and an artificial coastline coding path; The fused feature map generation module is used to fuse the output features of the natural coastline coding path and the artificial coastline coding path through a dynamic weight fusion module to generate a fused feature map; The result output module is used to perform land-sea segmentation and morphological optimization based on the fused feature map to segment and extract the coastline.

[0063] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A coastline segmentation and extraction method based on remote sensing data, characterized in that, It includes the following steps: Step S1: Obtain multi-source remote sensing image data of the target coastal zone area; Step S2: Construct a dual-path U-Net model, and the dual-path U-Net model includes a natural shoreline encoding path and an artificial shoreline encoding path; Step S3: The natural shoreline encoding path and the artificial shoreline encoding path obtain a natural shoreline feature map and an artificial shoreline feature map according to the multi-source remote sensing image data; Step S4: Use a dynamic weight fusion module to fuse the natural shoreline feature map and the artificial shoreline feature map to generate a fused feature map; Step S5: Based on the fused feature map, perform land-sea segmentation and morphological optimization to segment and extract the coastline.

2. The method for segmenting and extracting a coastline based on remote sensing data according to claim 1, wherein, The dual-path U-Net model in step S2 includes: The natural shoreline encoding path uses a convolutional kernel to extract the edge features of the natural shoreline to obtain a natural shoreline feature map; The artificial shoreline encoding path uses depthwise separable convolution to extract the edge features of the artificial shoreline to obtain an artificial shoreline feature map; The natural shoreline feature map and the artificial shoreline feature map are transmitted to the decoder through skip connections.

3. A method for segmenting and extracting a coastline based on remote sensing data according to claim 1, characterized in that, In step S4, using a dynamic weight fusion module to fuse the natural shoreline feature map and the artificial shoreline feature map to generate a fused feature map includes: Concatenate the natural shoreline feature map output by the natural shoreline encoding path and the artificial shoreline feature map output by the artificial shoreline encoding path along the channel dimension; Perform spatial attention calculation and channel attention calculation on the concatenated feature map respectively to obtain spatial attention weights and channel attention weights; Generate a dynamic weight α based on the spatial attention weights and channel attention weights; Perform weighted fusion on the natural shoreline feature map and the artificial shoreline feature map according to the dynamic weight α to generate the fused feature map.

4. A method for coastline segmentation and extraction based on remote sensing data according to claim 3, characterized in that The spatial attention calculation uses dilated convolution to extract context information to obtain spatial attention weights; The channel attention calculation generates channel attention weights through global average pooling and max pooling.

5. A method for coastline segmentation and extraction based on remote sensing data according to claim 3, characterized in that The calculation formula of the dynamic weight α is: ； In the formula, Ws(i,j) is the spatial attention weight at the spatial position (i,j) in the spliced feature map, Wc(i,j) is the channel attention weight at the spatial position (i,j) in the spliced feature map, and H and W are the height and width of the fused feature map respectively.

6. A method for coastline segmentation and extraction based on remote sensing data according to claim 1, characterized in that, In step S5, performing land-sea segmentation based on the fused feature map includes: Step S5A1: Perform multi-modal feature concatenation on the fused feature map and the multi-source remote sensing image data to obtain multi-modal input data; Step S5A2: Input the multi-modal input data into a cascaded ensemble learning module to obtain a natural shoreline classification result and an artificial shoreline classification result; Step S5A3: Perform spatial concatenation on the natural shoreline classification result and the artificial shoreline classification result to generate an initial coastline.

7. A method for coastline segmentation and extraction based on remote sensing data according to claim 6, characterized in that, In step S5A1, performing multi-modal feature concatenation on the fused feature map and the multi-source remote sensing image data to obtain multi-modal input data includes: Step S5A11: Obtain a normalized difference water index according to the green light band and the near-infrared band in the multi-source remote sensing image data; Step S5A12: Extract the VV polarization data of the synthetic aperture radar data in the multi-source remote sensing image data; Concatenate the normalized difference water index, the VV polarization data and the fused feature map along the channel dimension to obtain the multi-modal input data.

8. A method for segmenting and extracting a coastline based on remote sensing data according to claim 6, characterized in that, In step S5A2, the multi-modal input data is input into the cascaded ensemble learning module to obtain the natural shoreline classification result and the artificial shoreline classification result. To input the multi-modal input data into the cascaded ensemble learning module for region-based classification, it specifically includes: For the natural shoreline area, the normalized water index and the near-infrared band are threshold-segmented using the random forest classifier in the cascaded ensemble learning module to obtain the natural shoreline classification result; For the artificial shoreline area, the VV polarization data and the red light band are edge-enhanced using the support vector machine classifier in the cascaded ensemble learning module to obtain the artificial shoreline classification result.

9. A method for segmenting and extracting a coastline based on remote sensing data according to claim 1, characterized in that, The morphological optimization includes: Step S5B1: Denoise the initial coastline through opening operation; Step S5B2: Perform closing operation on the denoised initial coastline to obtain the optimized coastline; Step S5B3: Perform boundary tracking processing based on the optimized coastline to obtain the final coastline.

10. A coastline segmentation and extraction system based on remote sensing data, which executes a coastline segmentation and extraction method based on remote sensing data according to any one of claims 1-9, characterized in that, The system includes: a data acquisition module, a model construction module, a fused feature map generation module, and a result output module; The data acquisition module is used to acquire multi-source remote sensing image data of the target coastal zone area; The model construction module is used to construct a dual-path U-Net model, and the dual-path U-Net model includes a natural shoreline encoding path and an artificial shoreline encoding path; The fused feature map generation module is used to fuse the output features of the natural shoreline encoding path and the artificial shoreline encoding path through a dynamic weight fusion module to generate a fused feature map; The result output module is used to perform land-sea segmentation and morphological optimization based on the fused feature map to segment and extract the coastline.

Images