Coastline automatic extraction method and system based on deep learning and satellite image

Abstract

The present application relates to the technical field of image processing, and particularly relates to a method and system for automatically extracting a coastline based on deep learning and satellite images. The method comprises the following steps: obtaining multi-temporal high-resolution image data of a target coastline, and performing preprocessing to obtain an enhanced image sequence; extracting a connected component of a saline erosion area, and generating a refined boundary pixel set; generating a distance field, calculating the minimum distance of each boundary pixel to the nearest boundary, generating an initial center point sequence, and optimizing the initial center point sequence by using a spline curve fitting algorithm; generating a polygon buffer zone, superimposing the polygon buffer zone with the enhanced image sequence, and verifying the polygon buffer zone to obtain a verified center point sequence; fitting a continuous vector path, extracting coastline vector data, performing multi-temporal change monitoring, obtaining high-salinity soil accumulation and erosion rate indicators, and updating pixel-level classification to generate a final coastline vector data sequence. The present application can efficiently and accurately extract a saline erosion coastline, and provide reliable data support for ecological environment monitoring.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for automatic coastline extraction based on deep learning and satellite imagery. Background Technology

[0002] Coastlines are among the most complex and dynamically changing natural interfaces on Earth's surface. With the ongoing impacts of global climate change and human activities, coastlines face various environmental challenges, including salt erosion, sea-level rise, and land subsidence. These changes pose significant challenges to the ecological environment, marine resource protection, and coastal urban planning. Traditional coastline monitoring methods rely on manual measurement and field surveys, which are not only extremely labor-intensive but also inefficient and costly, making it difficult to meet the demands of high-frequency, large-scale monitoring.

[0003] With the rapid development of satellite remote sensing technology, using satellite imagery for coastline extraction and monitoring has gradually become a feasible solution. However, automatic coastline extraction faces numerous challenges, especially against complex terrain backgrounds such as salt erosion. Factors such as image quality variations, blurred coastline boundaries, and environmental noise severely impact the accuracy and reliability of traditional image processing algorithms. Therefore, there is an urgent need for an efficient, accurate, and automated coastline extraction technology that can overcome environmental interference and achieve accurate monitoring of coastlines and their dynamic changes.

[0004] This invention presents an automatic coastline extraction method based on deep learning and satellite imagery. By introducing advanced deep learning models and remote sensing image processing technology, it can automatically and accurately extract coastlines in saline erosion areas and monitor their changes in real time. This method not only improves the accuracy and efficiency of coastline extraction but also enables multi-temporal dynamic monitoring, providing more scientific and precise data support for coastline management and environmental protection. Summary of the Invention

[0005] This invention provides a method and system for automatic coastline extraction based on deep learning and satellite imagery, which can be used to achieve high-precision coastline extraction and monitoring.

[0006] In a first aspect, the present invention provides an automatic coastline extraction method based on deep learning and satellite imagery, the method comprising:

[0007] Step S1: Acquire multi-temporal high-resolution image data of the target coastline using satellite remote sensing equipment and preprocess it to obtain an enhanced image sequence; perform pixel-level classification on the enhanced image sequence, extract connected components of the salt erosion area and refine the boundary to obtain a refined boundary pixel set; Step S2: Generate a distance field using the refined boundary pixel set, and calculate the minimum distance from each boundary pixel to the nearest boundary based on the Euclidean distance transform to generate an initial center point sequence; use a spline curve fitting algorithm to smooth the initial center point sequence to obtain an optimized center point sequence; Step S3: Generate a polygon buffer based on the optimized center point sequence and overlay it with the enhanced image sequence for verification, to obtain the verified center point sequence; Step S4: Generate a continuous vector path by fitting the verified center point sequence, extract the saline erosion coastline vector data based on the continuous vector path; monitor multi-temporal changes based on the saline erosion coastline vector data, obtain high salinity soil accumulation and erosion rate indicators, determine whether vegetation degradation and tidal influence exceed the preset threshold, optimize the monitoring path, and obtain a refined monitoring path. Step S5: Update the pixel-level classification through the refined monitoring path to generate the final coastline vector data sequence.

[0008] As a preferred embodiment of the present invention, step S1 includes: Multi-temporal high-resolution image data of the target coastline are acquired using satellite remote sensing equipment. Preliminary cropping and registration are performed on the coastal salt erosion areas of the target coastline to generate a standardized original image sequence. The gray value distribution of the standardized original image sequence is adjusted by corresponding preset illumination correction parameters and contrast enhancement parameters to obtain an enhanced image sequence, wherein the gray value of each pixel in the enhanced image sequence is normalized. The enhanced image sequence is input into a pre-trained deep learning U-Net semantic segmentation model to perform pixel-level classification, determining whether each pixel belongs to the category of saline water or eroded land, and generating a preliminary segmentation mask. Connected components of the saline erosion region are extracted from the preliminary segmentation mask, and curvature values ​​of vegetation degradation and tidal effects are calculated for each connected component. If the curvature value is higher than a preset threshold, morphological dilation operations of structuring element selection and boundary pixel expansion are used, combined with pixel set coordinate transformation and Euclidean distance metric formula, to refine the boundaries of connected components and generate a refined boundary pixel set.

[0009] As a preferred embodiment of the present invention, step S2, generating a distance field and constructing an initial center point sequence through the refined boundary pixel set, includes: Euclidean distance transformation is calculated based on the refined boundary pixel set and boundary pixels to generate a two-dimensional distance field. A mesh is constructed using this distance field, and the geometric center path is located by combining boundary pixel weight assignment and noise point distance filtering, generating an initial center point sequence. The continuity of the geometric center path and the degree of noise interference are verified based on the initial center point sequence. Noise points in the initial center point sequence are adjusted using preset weight assignment rules and distance filtering parameters to generate an initial center point sequence suitable for spline curve fitting, wherein each point in the initial center point sequence contains two-dimensional coordinate information and a distance field value.

[0010] As a preferred embodiment of the present invention, step S2, obtaining the optimized center point sequence, includes: Based on the initial center point sequence, a spline curve fitting algorithm is applied to smooth the noise interference caused by sea level rise and groundwater salinization, and the fitting residual is calculated to determine whether the fitting residual is lower than a preset threshold. If the fitting residual is lower than the preset threshold, the smoothing result is obtained, and an optimized center point sequence is generated by combining a grayscale diffusion model and kernel size adaptive scaling. Based on the coordinates and grayscale values ​​of the optimized center point sequence, the continuity and stability of the smoothed sequence are verified to obtain the optimized center point sequence. Each point in the optimized center point sequence is adjusted by kernel size to adapt to boundary changes at different scales.

[0011] As a preferred embodiment of the present invention, step S3 includes: A polygonal buffer is generated from the optimized center point sequence; the polygonal buffer is superimposed on the enhanced image sequence to verify consistency; inconsistencies caused by ecosystem destruction and land subsidence trends are addressed through iterative fusion of propagation direction vector control and dilation operation; a verified center point sequence is generated through vector direction calculation and dilation kernel selection, combined with iterative fusion and boundary expansion control, wherein each point in the verified center point sequence undergoes consistency verification to ensure pixel-level matching with the enhanced image sequence; the boundary range of the polygonal buffer is adjusted according to the verified center point sequence.

[0012] As a preferred embodiment of the present invention, step S4, determining the vector data of the salt erosion coastline, includes: By applying the least squares method to fit the validated center point sequence, a continuous vector path is generated. Noise suppression mechanisms and path propagation constraints are incorporated, and the smoothness and continuity of the continuous vector path are adjusted by combining the fusion threshold setting and the number of operation iterations. The geometric consistency and noise suppression effect of the path are verified based on the coordinates and direction vectors of the continuous vector path. Finally, salt erosion coastline vector data is generated, wherein the salt erosion coastline vector data contains a two-dimensional coordinate sequence and path direction information.

[0013] As a preferred embodiment of the present invention, in step S4... Win Collect indicators of high-salinity soil accumulation and erosion rate, and determine whether vegetation degradation and tidal effects exceed preset thresholds and optimize monitoring pathways, including: Based on the saline erosion coastline vector data, indices of high salinity soil accumulation and erosion rate under multi-temporal variations are calculated; it is determined whether the vegetation degradation and tidal impact corresponding to the high salinity soil accumulation and erosion rate indices exceed preset thresholds; if they exceed the preset thresholds, the boundary expansion control and noise suppression mechanism are updated through vector direction calculation and expansion kernel selection, and the path propagation constraints are optimized; based on the optimized path propagation constraints, a multi-temporal variation monitoring sequence suitable for the refined monitoring path is generated, wherein the multi-temporal variation monitoring sequence includes time-series-based high salinity soil accumulation and erosion rate indices.

[0014] As a preferred embodiment of the present invention, in step S4, the refining monitoring path is obtained, including: The optimization path propagation constraints are used to extract adjustment parameters for the fusion threshold setting and the number of operation iterations. Based on the adjustment parameters, the monitoring sequence for sea-level rise and groundwater salinization is determined. The continuity of the ecosystem damage and land subsidence trends in the monitoring sequence is judged. If the continuity is lower than a preset threshold, a grayscale diffusion model and kernel size adaptive scaling are used, combined with spline curve fitting, to generate a refined monitoring path. Each point in the refined monitoring path contains smoothed coordinates and grayscale information. Based on the refined monitoring path, the continuity and stability of the monitoring sequence are verified, and the final refined monitoring path is generated.

[0015] As a preferred embodiment of the present invention, step S5 includes: The refined monitoring path is overlaid onto a standardized original image sequence, and pixel-level classification updates are performed. Based on the updated pixel-level classification results, a final vector data sequence for the salt erosion area of ​​a tropical island's concave coast is generated. The accuracy of the size adjustment threshold calibration and value propagation boundary constraints is verified using the final vector data sequence. Based on the verification results, the boundary range and coordinate information of the final vector data sequence are adjusted to generate a final coastline vector data sequence suitable for monitoring applications. The final coastline vector data sequence contains vector paths and classification labels for the salt erosion area under multi-temporal variations.

[0016] Secondly, the present invention also provides an automatic coastline extraction system based on deep learning and satellite imagery, for implementing the above-mentioned method, the system comprising: The image acquisition unit is used to acquire multi-temporal high-resolution image data of the target area through satellite remote sensing equipment and perform preprocessing to obtain enhanced image sequences. A pixel classification unit is used to perform pixel-level classification of the enhanced image sequence, extract connected components of the salt erosion region and refine the boundary to obtain a refined boundary pixel set. The computational fitting unit is used to generate a distance field through the refined boundary pixel set, and calculate the minimum distance from each boundary pixel to the nearest boundary based on the Euclidean distance transformation to generate an initial center point sequence; the initial center point sequence is smoothed by a spline curve fitting algorithm to obtain an optimized center point sequence. The overlay verification unit is used to generate a polygonal buffer based on the optimized center point sequence and overlay it with the enhanced image sequence for verification, thereby obtaining a verified center point sequence. The path fitting unit is used to generate a continuous vector path by fitting the verified center point sequence, and to extract vector data of salt erosion coastline based on the continuous vector path. The change monitoring unit is used to monitor multi-temporal changes based on the saline erosion coastline vector data, obtain indicators of high salinity soil accumulation and erosion rate, determine whether vegetation degradation and tidal influence exceed preset thresholds, optimize the monitoring path, and obtain a refined monitoring path. A classification update unit is used to update the pixel-level classification through the refined monitoring path to generate a final coastline vector data sequence. A fourth aspect of the invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described method.

[0017] The beneficial effects of this invention are as follows: This invention acquires multi-temporal high-resolution image data of the target coastline using satellite remote sensing equipment. Enhanced image sequences are obtained through preprocessing to ensure consistent grayscale value distribution across different time points. The preprocessed image sequences effectively eliminate the influence of varying lighting conditions and weather factors. Deep learning models such as U-Net are used for pixel-level classification to accurately extract salt erosion areas, extract connected components, and refine boundaries, generating a refined boundary pixel set to ensure boundary accuracy and lay a solid foundation for subsequent steps. A distance field is generated based on the refined boundary pixel set, and the minimum distance from each boundary pixel to the nearest boundary is calculated using Euclidean distance transformation to further generate an initial center point sequence, effectively capturing key features of the coastline. In the optimization stage, a spline curve fitting algorithm is used to smooth the initial center point sequence, obtaining an optimized center point sequence that ensures the smoothness and continuity of the coastline. Based on the optimized... The center point sequence is used to generate a polygon buffer, which is then overlaid with an enhanced image sequence for verification. This further ensures the consistency between the generated center point sequence and the image data, avoiding errors caused by noise or ecosystem changes, and providing stable input data for subsequent path fitting. A continuous vector path is generated through fitting the verified center point sequence, and vector data of the saline erosion coastline is extracted. Combined with multi-temporal change monitoring, indicators of high-salinity soil accumulation and erosion rate are obtained, providing data support for ecological and environmental changes. If monitoring detects vegetation degradation or tidal impact exceeding a preset threshold, the monitoring path will be further optimized to ensure continuous monitoring accuracy. Pixel-level classification is updated by refining the monitoring path, generating the final coastline vector data sequence. Through the synergistic effect of the above steps, the accuracy and efficiency of coastline extraction are greatly improved, enabling it to adapt to monitoring needs in different regions and environments, and providing accurate data on saline erosion and ecological and environmental changes. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the automatic coastline extraction method based on deep learning and satellite imagery in the embodiment. Figure 2 This is a flowchart of the verified center point sequence acquisition method in an embodiment of the present invention; Figure 3 This is a structural diagram of the automatic coastline extraction system based on deep learning and satellite imagery in the embodiment. Detailed Implementation

[0020] This invention provides a method and system for automatic coastline extraction based on deep learning and satellite imagery. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] For ease of understanding, the specific process of the embodiments of the present invention will be described below, such as... Figure 1 As shown, the automatic coastline extraction method based on deep learning and satellite imagery in this embodiment of the invention includes: Step S1: Acquire multi-temporal high-resolution image data of the target coastline using satellite remote sensing equipment and preprocess it to obtain an enhanced image sequence; perform pixel-level classification on the enhanced image sequence, extract connected components of the salt erosion area and refine the boundaries to obtain a refined boundary pixel set; specifically including: Multi-temporal high-resolution image data of the target coastline are acquired using satellite remote sensing equipment. Preliminary cropping and registration are performed on the coastal salt erosion areas of the target coastline to generate a standardized original image sequence. The gray value distribution of the standardized original image sequence is adjusted by corresponding preset illumination correction parameters and contrast enhancement parameters to obtain an enhanced image sequence, wherein the gray value of each pixel in the enhanced image sequence is normalized. The enhanced image sequence is input into a pre-trained deep learning U-Net semantic segmentation model to perform pixel-level classification, determining whether each pixel belongs to the category of saline water or eroded land, and generating a preliminary segmentation mask. Connected components of the saline erosion region are extracted from the preliminary segmentation mask, and curvature values ​​of vegetation degradation and tidal effects are calculated for each connected component. If the curvature value is higher than a preset threshold, morphological dilation operations of structuring element selection and boundary pixel expansion are used, combined with pixel set coordinate transformation and Euclidean distance metric formula, to refine the boundaries of connected components and generate a refined boundary pixel set.

[0022] Specifically, multi-temporal high-resolution image data of the target coastline is acquired using satellite remote sensing equipment. Preliminary cropping and registration are performed on areas of coastal salt erosion to generate a standardized original image sequence. Illumination correction and contrast enhancement are applied to corresponding areas within the multi-temporal image sequence. This involves adjusting the grayscale distribution of the images using preset illumination correction and contrast enhancement parameters to obtain an enhanced image sequence. Furthermore, the grayscale value of each pixel is normalized to ensure consistency between different image sequences and reduce image differences caused by varying shooting conditions at different times. After obtaining the enhanced image sequence, the images are input into a pre-trained deep learning U-Net semantic segmentation model to perform pixel-level classification. The et model uses an encoder-decoder structure built through a convolutional neural network. It first downsamples the input image to extract features, then upsamples it to restore image resolution, ultimately outputting the classification result for each pixel, labeled as either saline water or eroded land, generating a preliminary segmentation mask. Further, it extracts connected components from the preliminary segmentation mask, where these components are adjacent pixel regions with similar or identical grayscale values. For each connected component, it calculates the curvature value of its boundary. Specifically, by calculating the local curvature of boundary pixels, it identifies the impact of vegetation degradation and tidal influences on coastline morphology. If the calculated curvature value exceeds a set threshold, it indicates that the boundary of that region is experiencing intensified erosion and requires further processing.

[0023] Morphological dilation is used to refine the boundaries, enhancing their continuity and accuracy. This operation selects appropriate structuring elements and adaptively adjusts their size based on curvature values. Combined with pixel set coordinate transformation and Euclidean distance metric, it further expands the boundary pixels, filling gaps caused by sea-level rise or groundwater salinization. This effectively reduces noise from high-salinity soil accumulation, thus optimizing the boundaries of connected components and generating a refined boundary pixel set. This refined boundary pixel set will serve as a basis for multi-temporal change monitoring and coastline extraction. Multi-temporal monitoring allows for the acquisition of indicators of saline soil accumulation and erosion rates. By comparing changes at different time points, it further determines whether vegetation degradation and tidal influence exceed preset thresholds, optimizing monitoring paths and improving monitoring accuracy. The refined monitoring path effectively updates pixel-level classifications and generates the final coastline data sequence.

[0024] Step S2: Generate a distance field using the refined boundary pixel set, and calculate the minimum distance from each boundary pixel to the nearest boundary based on the Euclidean distance transform to generate an initial center point sequence; use a spline curve fitting algorithm to smooth the initial center point sequence to obtain an optimized center point sequence; In step S2, generating a distance field and constructing an initial center point sequence using the refined boundary pixel set includes: Euclidean distance transformation is calculated based on the refined boundary pixel set and boundary pixels to generate a two-dimensional distance field. A mesh is constructed using this distance field, and the geometric center path is located by combining boundary pixel weight assignment and noise point distance filtering, generating an initial center point sequence. The continuity of the geometric center path and the degree of noise interference are verified based on the initial center point sequence. Noise points in the initial center point sequence are adjusted using preset weight assignment rules and distance filtering parameters to generate an initial center point sequence suitable for spline curve fitting, wherein each point in the initial center point sequence contains two-dimensional coordinate information and a distance field value.

[0025] Specifically, for each pixel in the refined boundary pixel set, the Euclidean distance to the nearest boundary pixel is calculated to generate a distance field. This distance field is a two-dimensional array, where each element represents the minimum distance value of the corresponding pixel. The distance field highlights the geometric features of the boundary structure, aiding in subsequent path localization. The image is divided into multiple grid cells using a grid construction method, and the distance values ​​within each cell are averaged to obtain a grid representative value. Boundary pixels are weighted, with higher weights assigned to boundary pixels with larger distance field values ​​to emphasize their role in path localization, and lower weights assigned to boundary pixels with smaller distance field values. Furthermore, a noise point distance filtering technique is applied, removing isolated noise points by setting a distance threshold to avoid interference from high-salinity soil accumulation, thereby determining the geometric center path and generating an initial center point sequence. Each center point records its two-dimensional coordinates and distance field value. This geometric center path is determined by calculating the geometric features of the boundary region, representing a path to the center of that region. After the point sequence is established, the continuity of the geometric center path and the degree of noise interference are further verified. The Euclidean distance between adjacent points is calculated and compared with a preset threshold to determine the continuity of the path. Points with an Euclidean distance greater than the preset threshold are considered discontinuous, while those with a distance less than the preset threshold are considered continuous. Simultaneously, the standard deviation of the distance field values ​​in the path is evaluated to measure the degree of noise interference. If there are many noisy points in the path, the initial center point sequence is adjusted according to preset weighting rules and distance filtering parameters to remove noisy points and smooth them. Specifically, outliers are replaced by neighborhood averaging to generate an initial center point sequence more suitable for spline curve fitting. The optimized center point sequence can be used for subsequent path fitting, reducing noise interference from factors such as sea-level rise and groundwater salinization, thereby improving the accuracy of coastline vector data. This technical solution, through precise image processing, enables efficient monitoring of saline erosion areas and accurate extraction of coastlines, providing reliable data support for subsequent ecological environment change analysis.

[0026] Further, in step S2, the optimized center point sequence is obtained, including: Based on the initial center point sequence, a spline curve fitting algorithm is used to smooth the noise interference caused by sea level rise and groundwater salinization, and the fitting residual is calculated to determine whether the fitting residual is lower than a preset threshold. If the fitting residual is lower than the preset threshold, the smoothing result is obtained, and an optimized center point sequence is generated by combining a grayscale diffusion model and kernel size adaptive scaling. Based on the coordinates and grayscale values ​​of the optimized center point sequence, the continuity and stability of the smoothed sequence are verified to obtain the optimized center point sequence. Each point in the optimized center point sequence is adjusted by kernel size to adapt to boundary changes at different scales.

[0027] Specifically, based on the initial center point sequence, a spline curve fitting algorithm is used to smooth noise interference caused by sea-level rise and groundwater salinization. Spline curve fitting is a curve interpolation method based on a polynomial piecewise function, which can smooth irregular data and reduce noise interference caused by ecological factors. By minimizing the deviation between the curve and the data points, the fitting result can accurately reflect the changing trend of the coastline, thereby reducing the interference of high-salinity soil accumulation on image quality. During the fitting process, the initial center point sequence is selected as control points, and the order of the B-spline is set to balance smoothness and... The fitting accuracy is set to, for example, third order, and the fitting residuals are calculated to determine the effectiveness of the fitting. If the fitting residuals are below a preset threshold, the fitting result is considered effective and the smoothing result is acceptable. Based on the smoothing result, an optimized center point sequence is generated by combining a grayscale diffusion model and an adaptive kernel size scaling technique. The grayscale diffusion model is an image smoothing technique based on anisotropic diffusion equations, which preserves edge details and reduces noise by iteratively propagating pixel grayscale values. The adaptive kernel size scaling dynamically adjusts the kernel size according to changes in local curvature, specifically reducing the kernel size in high-curvature regions to finely handle tidal effects. In the enhanced portion, the kernel size is expanded in low-curvature regions to smooth out interference areas caused by groundwater salinization. The optimized center point sequence can better adapt to boundary changes at different scales, improve the sequence's adaptability to ecosystem damage trends, and ensure the accuracy of subsequent vector transformations. Furthermore, the continuity and stability of the smoothed sequence are verified based on the coordinates and grayscale values ​​of the optimized center point sequence. Continuity verification is performed by calculating the Euclidean distance between adjacent points and their grayscale gradient change rate. If the change rate is lower than a preset threshold, the sequence's stability is confirmed. The optimized center point sequence, after this verification, maintains high accuracy when generating polygonal buffers and adapts to boundary changes at different scales. Specifically, a small kernel size is used for areas with accelerated local erosion rates, while a large kernel size is used for larger-scale overall land subsidence trends to meet different monitoring needs, such as those of tropical island depression coastlines. The optimized center point sequence improves the accuracy of erosion rate monitoring and provides reliable data support for subsequent multi-temporal change analysis. The above technical solution effectively improves the accuracy of coastline extraction, overcomes noise interference caused by ecological and environmental changes, and provides efficient and accurate technical support for automatic coastline extraction.

[0028] Step S3: Generate a polygon buffer based on the optimized center point sequence and overlay it with the enhanced image sequence for verification, to obtain the verified center point sequence; specifically including: A polygonal buffer is generated from the optimized center point sequence; the polygonal buffer is superimposed on the enhanced image sequence to verify consistency; inconsistencies caused by ecosystem destruction and land subsidence trends are addressed through iterative fusion of propagation direction vector control and dilation operation; a verified center point sequence is generated through vector direction calculation and dilation kernel selection, combined with iterative fusion and boundary expansion control, wherein each point in the verified center point sequence undergoes consistency verification to ensure pixel-level matching with the enhanced image sequence; the boundary range of the polygonal buffer is adjusted according to the verified center point sequence.

[0029] Specifically, based on the optimized center point sequence, a polygonal buffer is first generated. The generation process utilizes a buffer algorithm to calculate the pixel region within a preset radius around each center point, forming the corresponding polygonal outline. Subsequently, this polygonal buffer is overlaid with the enhanced image sequence, and the pixel matching degree of the overlapping area is calculated to verify its consistency. If the matching degree is higher than a preset threshold, it is considered a consistent match; otherwise, it is marked as an inconsistent region. To further address inconsistencies caused by ecosystem destruction and land subsidence trends, an iterative fusion method combining propagation direction vector control and dilation operations is employed. This includes calculating the propagation direction vector of the inconsistent region, determined by the gradient direction of adjacent pixels, and used to guide the direction of the dilation operation. In the dilation operation, appropriate structural elements, such as a cross-shaped or circular kernel, are selected, and their bias is controlled by the direction vector, ensuring that dilation proceeds only along a specific direction. Through multiple iterations, pixels in inconsistent regions are gradually fused to achieve higher matching accuracy. The number of dilation iterations is set according to the severity of the land subsidence trend; the number of iterations increases in more severe cases to minimize interference with the original data during the fusion process and preserve the original data. Subtle erosion features are observed. Furthermore, after each dilation, consistency verification is performed, i.e., by recalculating the matching degree of the dilated region, ensuring that the pixel-level consistency between these regions and the enhanced image sequence is optimized. If the verification results still show inconsistencies, processing can continue by adjusting the dilation kernel size or increasing the number of iterations, thereby ensuring accurate matching of each point with the image data and improving the overall monitoring reliability. Through the above technical solution, the generated verified center point sequence has higher accuracy and can be further used to adjust the boundary range of the polygon buffer, ensuring that the buffer boundary matches the final generated center point sequence. The adjusted polygon buffer is fitted and optimized using the least squares method, further improving the smoothness and matching degree of the path, effectively improving the monitoring effect of salt erosion areas, and providing more accurate support for subsequent vector path fitting.

[0030] Step S4: Generate a continuous vector path by fitting the verified center point sequence, extract the saline erosion coastline vector data based on the continuous vector path; monitor multi-temporal changes based on the saline erosion coastline vector data, obtain high salinity soil accumulation and erosion rate indicators, determine whether vegetation degradation and tidal influence exceed the preset threshold, optimize the monitoring path, and obtain a refined monitoring path. In step S4, determining the vector data of the salt erosion coastline includes: By applying the least squares method to fit the validated center point sequence, a continuous vector path is generated. Noise suppression mechanisms and path propagation constraints are incorporated, and the smoothness and continuity of the continuous vector path are adjusted by combining the fusion threshold setting and the number of operation iterations. The geometric consistency and noise suppression effect of the path are verified based on the coordinates and direction vectors of the continuous vector path. Finally, salt erosion coastline vector data is generated, wherein the salt erosion coastline vector data contains a two-dimensional coordinate sequence and path direction information.

[0031] Specifically, a continuous vector path is generated by applying a least squares fitting algorithm. This fitting process transforms a discrete sequence of center points into a smooth, continuous path by minimizing the sum of squared residuals between each point and the fitted curve. To further optimize the accuracy and stability of the path, a noise suppression mechanism and path propagation constraints are incorporated. The noise suppression mechanism reduces interference from factors such as tidal influences and groundwater salinization by removing outliers that deviate from the path. The path propagation constraints limit excessive fluctuations in the path by setting appropriate fusion thresholds and iteration counts, ensuring the optimization of path smoothness and continuity. Based on path fitting and noise suppression, the fitting effect is further verified by calculating the geometric consistency of the path. Geometric consistency confirms the smoothness of the path by checking the Euclidean distance between adjacent points. A high degree of consistency in the spacing between adjacent points indicates good path consistency. Furthermore, the noise suppression effect is evaluated by calculating the change in the sum of squared residuals before and after path fitting. If the reduction in residuals exceeds a set threshold, the path is considered effectively optimized. These vector paths are used to generate vector data for salt-eroded coastlines. Specifically, two-dimensional coordinate sequences and path direction information are extracted from the validated paths. The path direction information is obtained by calculating the tangent vector at each point. This information collectively constitutes the vector data format for salt-eroded coastlines. This vector data not only accurately reflects the morphology of the coastline but can also be used in subsequent multi-temporal change monitoring, supporting the dynamic change analysis of salt-eroded areas in tropical island depression coastlines. This technical solution provides strong data support for coastline monitoring and ecological environmental protection.

[0032] Furthermore, in step S4, Obtain indicators of high-salinity soil accumulation and erosion rates.And determine whether vegetation degradation and tidal impact exceed preset thresholds and optimize monitoring paths, including: Based on the saline erosion coastline vector data, indices of high salinity soil accumulation and erosion rate under multi-temporal variations are calculated; it is determined whether the vegetation degradation and tidal impact corresponding to the high salinity soil accumulation and erosion rate indices exceed preset thresholds; if they exceed the preset thresholds, the boundary expansion control and noise suppression mechanism are updated through vector direction calculation and expansion kernel selection, and the path propagation constraints are optimized; based on the optimized path propagation constraints, a multi-temporal variation monitoring sequence suitable for the refined monitoring path is generated, wherein the multi-temporal variation monitoring sequence includes time-series-based high salinity soil accumulation and erosion rate indices.

[0033] Specifically, based on vector data of saline erosion coastlines, the accumulation and erosion rate indices of high-salinity soil are calculated using changes in multi-temporal imagery. The high-salinity soil accumulation index is quantified by calculating the differences in the lower boundary positions across different temporal phases, specifically by calculating the Euclidean distance between corresponding points in the center point sequence to determine the average distance the boundary moves inland, thus identifying the degree of high-salinity soil accumulation. Simultaneously, the erosion rate index is calculated by comparing changes in path length. Based on the proportion of path length change from its initial length to its current length, and considering time intervals, the change in erosion rate is derived. By calculating these high-salinity soil accumulation and erosion rate indices, the dynamics of saline erosion on the concave coastlines of tropical islands can be accurately captured, ensuring accurate quantification of multi-temporal data, reducing errors caused by simple pixel differences, and thus improving monitoring accuracy. After acquiring the high-salinity soil accumulation and erosion rate indices, it is further determined whether these indices exceed preset thresholds to identify ecological risks. For example, when the high salinity soil accumulation index exceeds a set threshold, it may indicate that vegetation degradation has exceeded the normal range; simultaneously, when the erosion rate index exceeds a preset value, it may indicate an increase in tidal influence. These judgments can help identify the intensification of salt erosion in a timely manner and provide a basis for subsequent optimization. If the index exceeds the preset threshold, the boundary expansion control and noise suppression mechanism are updated through vector direction calculation and expansion kernel selection, combined with an iterative fusion step, to optimize path propagation constraints. Specifically, the vector direction calculation is performed by calculating the direction vector through the coordinate difference between adjacent points, thereby determining the direction of boundary expansion. Based on this, a suitable expansion kernel is selected, and the size of the kernel is adjusted according to the intensity of tidal influence; for example, a larger kernel is used in areas of strong erosion. The kernel is used to cover a wider area, and then morphological expansion is performed. After each expansion operation, the extent of boundary expansion is adjusted by judging the residuals, and combined with a noise suppression mechanism, isolated points that are more than a preset threshold away from the geometric center path are removed, thereby reducing interference and improving the continuity and accuracy of the path. With each iteration, the path propagation constraints are continuously optimized, and finally a multi-temporal change monitoring sequence suitable for refining the monitoring path is generated. This sequence contains time-seriesed indicators of high salinity soil accumulation and erosion rate, providing accurate data support for subsequent monitoring. Through the above technical solution, the monitoring results can continuously track changes in the ecosystem, accurately identify the impact of factors such as sea-level rise and groundwater salinization, and improve the accuracy and robustness of coastline extraction.

[0034] Further, in step S4, the refined monitoring path is obtained, including: The optimization path propagation constraints are used to extract adjustment parameters for the fusion threshold setting and the number of operation iterations. Based on the adjustment parameters, the monitoring sequence for sea-level rise and groundwater salinization is determined. The continuity of the ecosystem damage and land subsidence trends in the monitoring sequence is judged. If the continuity is lower than a preset threshold, a grayscale diffusion model and kernel size adaptive scaling are used, combined with spline curve fitting, to generate a refined monitoring path. Each point in the refined monitoring path contains smoothed coordinates and grayscale information. Based on the refined monitoring path, the continuity and stability of the monitoring sequence are verified, and the final refined monitoring path is generated.

[0035] Specifically, adjustment parameters for the fusion threshold setting and the number of operation iterations are extracted from the optimized path propagation constraints. Specifically, the fusion threshold setting, which is a floating-point list, is identified by traversing the threshold array and iteration counter in the path propagation constraints. This threshold setting is calculated by accumulating boundary expansion control values ​​from previous iterations and is used to adjust the sensitivity of the monitoring sequence. Based on these adjustment parameters, monitoring sequences for sea-level rise and groundwater salinization are determined. These sequences include timestamps and quantified values ​​of the corresponding processes, such as the rate of sea-level rise and the degree of soil salinization, effectively reflecting changes in the ecological environment. The continuity of ecosystem damage and land subsidence trends in the monitoring sequences is judged by calculating the difference in damage indicators between adjacent time points. If the variance of this difference is less than the set continuity threshold, it is considered continuous; if it is greater than the threshold, it is marked as discontinuous, thus ensuring the stability and reliability of the monitoring sequence. If the detected continuity is below the threshold, a grayscale diffusion model and kernel size adaptive scaling technology, combined with spline curve fitting, are used to generate a refined monitoring path. The grayscale diffusion model is based on image gradient calculation, spreading from the center point to surrounding pixels to smooth noise interference areas and reduce noise. The system minimizes uneven illumination caused by environmental changes. Simultaneously, adaptive kernel scaling adjusts the kernel size based on local pixel density, determining the scaling factor by calculating the Euclidean distance between pixels within the region to adapt to the feature requirements of different areas. Furthermore, cubic spline interpolation is used to fit the diffused and scaled point sequence, generating a smooth monitoring path. Each fitted point contains smoothed coordinates and grayscale values, further improving the stability and continuity of the monitoring path. To ensure path accuracy, the continuity and stability of the refined monitoring path are verified to guarantee pixel-level matching with the image sequence. During path verification, the stability of the path is confirmed by calculating the consistency of distances between path points and comparing the path fitting residuals. If the consistency and stability meet preset requirements, the final refined monitoring path is output for updating pixel-level classification, thereby supporting the monitoring of coastal salt erosion areas. This technical solution, through the combined use of grayscale diffusion models, adaptive kernel scaling, and spline curve fitting, optimizes the continuity and accuracy of the path and effectively handles noise interference in salt erosion monitoring, providing an efficient and accurate automatic coastline extraction solution.

[0036] Step S5: Update the pixel-level classification through the refined monitoring path to generate the final coastline data sequence; specifically including: The refined monitoring path is overlaid onto a standardized original image sequence, and pixel-level classification updates are performed. Based on the updated pixel-level classification results, a final vector data sequence for the salt erosion area of ​​a tropical island's concave coast is generated. The accuracy of the size adjustment threshold calibration and value propagation boundary constraints is verified using the final vector data sequence. Based on the verification results, the boundary range and coordinate information of the final vector data sequence are adjusted to generate a final vector data sequence suitable for monitoring applications. The final coastline vector data sequence contains the coastline vector path and classification labels of the salt erosion area under multi-temporal variations.

[0037] Specifically, a refined monitoring path is superimposed onto a standardized original image sequence, and pixel-level classification updates are performed to update the classification label of each pixel in the original image sequence. This involves masking the image sequence using the pixel coordinates along the path to update the classification label of each pixel. The refined monitoring path originates from the smoothing results of the spline curve fitting and grayscale diffusion model mentioned above. The superposition of the refined monitoring path ensures that the image update takes into account the influence of factors such as sea-level rise and groundwater salinization on the boundaries, thereby improving the classification accuracy of saline erosion areas. For example, in a multi-temporal image sequence, the refined monitoring path covers boundary changes under the influence of sea-level rise. The superimposed classification update effectively filters noise interference, resulting in higher monitoring consistency and benefiting subsequent vector generation. Based on the updated pixel-level classification results, a final coastline vector data sequence for coastal saline erosion areas is further generated. This sequence contains multi-temporal change information extracted from the optimized monitoring path, such as the path changes of the coastline and the classification labels of the erosion areas. This process, by refining boundaries and eliminating noise, generates smooth and accurate vector data, ensuring high-precision coastline extraction.

[0038] The calibration effect of the size adjustment threshold and the accuracy of the value propagation boundary constraints were further verified through the final vector data sequence. The verification process included size adjustment and boundary consistency checks on the generated vector data to ensure that the generated coastline data maintained high accuracy and stability under multi-temporal changes. If deviations were found to exceed the preset range, the boundary range and coordinate information of the vector data were adjusted according to the actual verification results to adapt to different monitoring needs. The above technical solution, by combining boundary expansion control and dilatation kernel selection, optimizes path propagation constraints during iterative processing, further improving the accuracy of the data. The generated coastline vector data sequence contains coastline vector paths and classification labels for saline erosion areas under multi-temporal changes, thus effectively addressing dynamically changing coastal areas, improving the accuracy and robustness of coastline monitoring and protection, and ensuring real-time detection and response to ecological risks.

[0039] This invention also provides an automatic coastline extraction system based on deep learning and satellite imagery, used to implement the above-mentioned method, such as... Figure 3 As shown, the system includes: The image acquisition unit is used to acquire multi-temporal high-resolution image data of the target coastline through satellite remote sensing equipment and perform preprocessing to obtain an enhanced image sequence. A pixel classification unit is used to perform pixel-level classification of the enhanced image sequence, extract connected components of the salt erosion region and refine the boundary to obtain a refined boundary pixel set. The computational fitting unit is used to generate a distance field through the refined boundary pixel set, and calculate the minimum distance from each boundary pixel to the nearest boundary based on the Euclidean distance transformation to generate an initial center point sequence; the initial center point sequence is smoothed by a spline curve fitting algorithm to obtain an optimized center point sequence. The overlay verification unit is used to generate a polygonal buffer based on the optimized center point sequence and overlay it with the enhanced image sequence for verification, thereby obtaining a verified center point sequence. The path fitting unit is used to generate a continuous vector path by fitting the verified center point sequence, and to extract vector data of salt erosion coastline based on the continuous vector path. The change monitoring unit is used to monitor multi-temporal changes based on the saline erosion coastline vector data, obtain indicators of high salinity soil accumulation and erosion rate, determine whether vegetation degradation and tidal influence exceed preset thresholds, optimize the monitoring path, and obtain a refined monitoring path. The classification update unit is used to update the pixel-level classification through the refined monitoring path to generate the final coastline vector data sequence.

[0040] In summary, this invention acquires multi-temporal high-resolution image data of the target coastline using satellite remote sensing equipment, and obtains enhanced image sequences through preprocessing to ensure the consistency of grayscale value distribution at different time points. The preprocessed image sequences can effectively eliminate the influence of different lighting conditions and weather factors on the images. Furthermore, deep learning models such as U-Net are used for pixel-level classification to accurately extract salt erosion areas, extract connected components, and refine boundaries, generating a refined boundary pixel set to ensure boundary accuracy and lay a solid foundation for subsequent steps. A distance field is generated based on the refined boundary pixel set, and the minimum distance from each boundary pixel to the nearest boundary is calculated using Euclidean distance transformation to further generate an initial center point sequence, thereby effectively capturing the key features of the coastline. In the optimization stage, a spline curve fitting algorithm is used to flatten the initial center point sequence. The process involves smoothing the image to obtain an optimized center point sequence, ensuring the smoothness and continuity of the coastline. A polygon buffer is generated based on this optimized center point sequence and verified by overlaying it with an enhanced image sequence. This further ensures the consistency between the generated center point sequence and the image data, avoiding errors caused by noise or ecosystem changes, and providing stable input data for subsequent path fitting. A continuous vector path is generated through fitting the verified center point sequence, and vector data of the saline erosion coastline is extracted. Combined with multi-temporal change monitoring, indicators of high-salinity soil accumulation and erosion rate are obtained, providing data support for ecological and environmental change analysis. If monitoring detects vegetation degradation or tidal impact exceeding a preset threshold, the monitoring path will be further optimized to ensure continuous monitoring accuracy. Pixel-level classification is updated by refining the monitoring path, generating the final coastline vector data sequence. Through the synergistic effect of these steps, the accuracy and efficiency of coastline extraction are greatly improved, enabling the system to adapt to monitoring needs in different regions and environments, and providing accurate data on saline erosion and ecological and environmental changes.

[0041] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0042] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0043] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatic coastline extraction based on deep learning and satellite imagery, characterized in that, The method includes: Step S1: Acquire multi-temporal high-resolution image data of the target coastline using satellite remote sensing equipment and preprocess it to obtain an enhanced image sequence; perform pixel-level classification on the enhanced image sequence, extract the connected components of the salt erosion area and refine the boundaries to obtain a refined boundary pixel set; Step S2: Generate a distance field using the refined boundary pixel set, and calculate the minimum distance from each boundary pixel to the nearest boundary based on the Euclidean distance transform to generate an initial center point sequence; use a spline curve fitting algorithm to smooth the initial center point sequence to obtain an optimized center point sequence; Step S3: Generate a polygon buffer based on the optimized center point sequence and verify it by overlaying it with the enhanced image sequence to obtain a verified center point sequence; including: generating a polygon buffer from the optimized center point sequence; overlaying the polygon buffer with the enhanced image sequence to verify consistency; processing inconsistencies caused by ecosystem destruction and land subsidence trends through propagation direction vector control and dilation operation iterative fusion; generating a verified center point sequence by vector direction calculation and dilation kernel selection, combined with iterative fusion and boundary expansion control, wherein each point in the verified center point sequence undergoes consistency verification to ensure pixel-level matching with the enhanced image sequence; adjusting the boundary range of the polygon buffer according to the verified center point sequence; Step S4: Generate a continuous vector path by fitting the verified center point sequence, extract the saline erosion coastline vector data based on the continuous vector path; monitor multi-temporal changes based on the saline erosion coastline vector data, obtain high salinity soil accumulation and erosion rate indicators, determine whether vegetation degradation and tidal influence exceed the preset threshold, optimize the monitoring path, and obtain a refined monitoring path. Step S5: Update the pixel-level classification through the refined monitoring path to generate the final coastline vector data sequence.

2. The method as described in claim 1, characterized in that, Step S1 includes: Multi-temporal high-resolution image data of the target coastline are acquired using satellite remote sensing equipment. Preliminary cropping and registration are performed on the coastal salt erosion areas in the target coastline to generate a standardized original image sequence. The gray value distribution of the standardized original image sequence is adjusted by corresponding preset illumination correction parameters and contrast enhancement parameters to obtain an enhanced image sequence, wherein the gray value of each pixel in the enhanced image sequence is normalized. The enhanced image sequence is input into a pre-trained deep learning U-Net semantic segmentation model to perform pixel-level classification, determining whether each pixel belongs to the category of saline water or eroded land, and generating a preliminary segmentation mask. Connected components of the saline erosion region are extracted from the preliminary segmentation mask, and curvature values ​​of vegetation degradation and tidal effects are calculated for each connected component. If the curvature value is higher than a preset threshold, morphological dilation operations of structuring element selection and boundary pixel expansion are used, combined with pixel set coordinate transformation and Euclidean distance metric formula, to refine the boundaries of connected components and generate a refined boundary pixel set.

3. The method as described in claim 1, characterized in that, In step S2, generating a distance field and constructing an initial center point sequence using the refined boundary pixel set includes: Euclidean distance transformation is calculated based on the refined boundary pixel set and boundary pixels to generate a two-dimensional distance field. A mesh is constructed using this distance field, and the geometric center path is located by combining boundary pixel weight assignment and noise point distance filtering, generating an initial center point sequence. The continuity of the geometric center path and the degree of noise interference are verified based on the initial center point sequence. Noise points in the initial center point sequence are adjusted using preset weight assignment rules and distance filtering parameters to generate an initial center point sequence suitable for spline curve fitting, wherein each point in the initial center point sequence contains two-dimensional coordinate information and a distance field value.

4. The method as described in claim 3, characterized in that, In step S2, the optimized center point sequence is obtained, including: Based on the initial center point sequence, a spline curve fitting algorithm is applied to smooth the noise interference caused by sea level rise and groundwater salinization, and the fitting residual is calculated to determine whether the fitting residual is lower than a preset threshold. If the fitting residual is lower than the preset threshold, the smoothing result is obtained, and an optimized center point sequence is generated by combining a grayscale diffusion model and kernel size adaptive scaling. Based on the coordinates and grayscale values ​​of the optimized center point sequence, the continuity and stability of the smoothed sequence are verified to obtain the optimized center point sequence. Each point in the optimized center point sequence is adjusted by kernel size to adapt to boundary changes at different scales.

5. The method as described in claim 1, characterized in that, In step S4, the vector data of the salt erosion coastline is determined, including: By applying the least squares method to fit the validated center point sequence, a continuous vector path is generated. Noise suppression mechanisms and path propagation constraints are incorporated, and the smoothness and continuity of the continuous vector path are adjusted by combining the fusion threshold setting and the number of operation iterations. The geometric consistency and noise suppression effect of the path are verified based on the coordinates and direction vectors of the continuous vector path. Finally, salt erosion coastline vector data is generated, wherein the salt erosion coastline vector data contains a two-dimensional coordinate sequence and path direction information.

6. The method as described in claim 5, characterized in that, In step S4, indices of high-salinity soil accumulation and erosion rate are obtained, and it is determined whether vegetation degradation and tidal effects exceed preset thresholds and the monitoring path is optimized, including: Based on the saline erosion coastline vector data, indices of high salinity soil accumulation and erosion rate under multi-temporal variations are calculated; it is determined whether the vegetation degradation and tidal impact corresponding to the high salinity soil accumulation and erosion rate indices exceed preset thresholds; if they exceed the preset thresholds, the boundary expansion control and noise suppression mechanism are updated through vector direction calculation and expansion kernel selection, and the path propagation constraints are optimized; based on the optimized path propagation constraints, a multi-temporal variation monitoring sequence suitable for the refined monitoring path is generated, wherein the multi-temporal variation monitoring sequence includes time-series-based high salinity soil accumulation and erosion rate indices.

7. The method as described in claim 6, characterized in that, In step S4, the refined monitoring path is obtained, including: The optimization path propagation constraints are used to extract adjustment parameters for the fusion threshold setting and the number of operation iterations. Based on the adjustment parameters, the monitoring sequence for sea-level rise and groundwater salinization is determined. The continuity of the ecosystem damage and land subsidence trends in the monitoring sequence is judged. If the continuity is lower than a preset threshold, a grayscale diffusion model and kernel size adaptive scaling are used, combined with spline curve fitting, to generate a refined monitoring path. Each point in the refined monitoring path contains smoothed coordinates and grayscale information. Based on the refined monitoring path, the continuity and stability of the monitoring sequence are verified, and the final refined monitoring path is generated.

8. The method as described in claim 1, characterized in that, Step S5 includes: The refined monitoring path is overlaid onto a standardized original image sequence, and pixel-level classification updates are performed. Based on the updated pixel-level classification results, a final vector data sequence for the salt erosion area of ​​a tropical island's concave coast is generated. The accuracy of the size adjustment threshold calibration and value propagation boundary constraints is verified using the final vector data sequence. Based on the verification results, the boundary range and coordinate information of the final vector data sequence are adjusted to generate a final vector data sequence suitable for monitoring applications. The final coastline vector data sequence contains vector paths and classification labels for the salt erosion area under multi-temporal variations.

9. An automatic coastline extraction system based on deep learning and satellite imagery, used to implement the method as described in any one of claims 1-8, characterized in that, The system includes: The image acquisition unit is used to acquire multi-temporal high-resolution image data of the target area through satellite remote sensing equipment and perform preprocessing to obtain enhanced image sequences. A pixel classification unit is used to perform pixel-level classification of the enhanced image sequence, extract connected components of the salt erosion region and refine the boundary to obtain a refined boundary pixel set. The computational fitting unit is used to generate a distance field through the refined boundary pixel set, and calculate the minimum distance from each boundary pixel to the nearest boundary based on the Euclidean distance transformation to generate an initial center point sequence; the initial center point sequence is smoothed by a spline curve fitting algorithm to obtain an optimized center point sequence. The overlay verification unit is used to generate a polygonal buffer based on the optimized center point sequence and overlay it with the enhanced image sequence for verification, thereby obtaining a verified center point sequence. The path fitting unit is used to generate a continuous vector path by fitting the verified center point sequence, and to extract vector data of salt erosion coastline based on the continuous vector path. The change monitoring unit is used to monitor multi-temporal changes based on the saline erosion coastline vector data, obtain indicators of high salinity soil accumulation and erosion rate, determine whether vegetation degradation and tidal influence exceed preset thresholds, optimize the monitoring path, and obtain a refined monitoring path. The classification update unit is used to update the pixel-level classification through the refined monitoring path to generate the final coastline vector data sequence.

Images