A computer vision-based method for predicting island shoreline erosion risk

Abstract

The present invention provides a method for predicting the risk of island shoreline erosion based on computer vision, which belongs to the technical field of island shoreline erosion risk analysis. The present invention first obtains high-resolution remote sensing images and extracts island contour information, adopts a small grid approximation method to construct an initial shoreline model, generates a multi-level grid representation through grid merging operations and stores it as a shoreline feature matrix; then collects multi-time series remote sensing images, calculates the set of change matrices of adjacent time points, and analyzes the shoreline evolution pattern; establishes a coastal erosion dynamics equation model based on wave action intensity, tidal amplitude, sea level rise rate, geological structure coefficient and shoreline slope angle; applies a graph attention grid evolution model to analyze the law of shoreline change, and finally introduces an uncertainty analysis module to generate a multi-scale erosion risk assessment map, identify high-risk areas and output them as prediction results, thereby solving the technical problem of the difficulty in accurately predicting the risk of complex island shoreline erosion.

Description

Technical Field

[0001] The present invention belongs to the technical field of island shoreline erosion risk analysis, and specifically relates to an island shoreline erosion risk prediction method based on computer vision. Background Art

[0002] Coastal erosion on islands is a significant environmental issue in the context of global climate change. Traditional coastal erosion risk assessment relies primarily on field monitoring and mathematical modeling. Common methods include erosion rate calculation based on empirical formulas, shoreline change analysis using remote sensing imagery, and fluid dynamics modeling. These methods play a vital role in large-scale coastal management and protection, particularly in disaster prevention and mitigation efforts in island nations and densely populated coastal areas.

[0003] However, traditional technologies have significant drawbacks in their application. Existing remote sensing image analysis methods typically use simple linear representations of coastlines, making it difficult to accurately capture the details of complex, winding island coastlines. While physical models account for factors like waves and tides, they often overlook the influence of geological structure differences and local terrain features. Statistical models based on historical data lack an effective representation of multi-scale variations, resulting in inaccurate predictions in local areas.

[0004] Existing technologies struggle to simultaneously account for both large-scale, overall shoreline changes and small-scale, localized erosion details on islands with diverse coastline characteristics and patterns of change. This is particularly true in complex terrain, where the importance of global and local features cannot be effectively balanced, impacting prediction accuracy. This core technical challenge hinders accurate prediction of complex island coastline erosion risk, a significant constraint that severely hinders both the precise assessment of island coastline erosion risk and the implementation of effective protective measures. Summary of the Invention

[0005] In view of this, the present invention provides a computer vision-based island shoreline erosion risk prediction method, which can solve the technical problem in the existing technology that it is difficult to accurately predict the erosion risk of complex island shorelines.

[0006] The present invention is implemented as follows: The present invention provides a computer vision-based island shoreline erosion risk prediction method, which includes obtaining high-resolution island remote sensing images and extracting island contour information; constructing an initial island shoreline model based on a small grid approximation method; merging small grids to generate a multi-level grid representation; collecting multiple time series remote sensing images and calculating changes in the shoreline feature matrix; establishing a coastal erosion dynamics equation model, combining wave action intensity, tidal amplitude, sea level rise rate, island geological structure coefficient and shoreline slope angle to quantitatively calculate the erosion rate at different grid positions; applying a graph attention grid evolution model to analyze the shoreline change law, and the graph attention grid evolution model adjusts the attention allocation mechanism through the grid area weight coefficient to balance global changes and local details; introducing an uncertainty analysis module to calculate the confidence interval of the prediction result, generate a multi-scale erosion risk assessment map, identify high-risk areas and output the prediction results.

[0007] Among them, obtaining high-resolution island remote sensing images and extracting island contour information includes denoising, enhancing, registering and segmenting the images, and using machine vision algorithms to extract island contour information.

[0008] Among them, constructing the initial model of the island coastline based on the small grid approximation method includes dividing the island boundary into multiple square grids with side lengths of the first size value to form an island coastline grid set.

[0009] Among them, the small grid approximation method represents the island boundary as a set of square grids with equal side lengths. By adjusting the grid density, the complex coastline morphology can be accurately expressed. Each grid records the existence status and attribute information of the coastline in the region.

[0010] Among them, the first size value is the initial grid side length determined according to the complexity of the island coastline and the resolution of the remote sensing image, and the value range is 5 meters to 50 meters.

[0011] The merging operation of small grids to generate a multi-level grid representation includes finding the global largest square grid and storing all grid information as a shoreline feature matrix.

[0012] Among them, the shoreline feature matrix is ​​a two-dimensional data structure, whose element values ​​represent the attributes of the corresponding position grid, including whether it is a shoreline point, shoreline type, elevation value and stability index. The matrix dimension is determined by the grid division accuracy.

[0013] Among them, the coastal erosion dynamics equation is used to simulate the physical mechanism of coastal erosion. Based on the principle of conservation of mass and fluid dynamics theory, it considers the balance between the impact force of waves on the coastline and the resistance of the coastline material, and obtains the erosion rate distribution of grids at different locations through integral calculation.

[0014] Among them, the structure of the graph attention grid evolution model is a fusion architecture of a multi-layer graph convolutional neural network and a multi-head attention mechanism. The shoreline grid is represented as graph structure data, and the connection relationship between grids corresponds to the edge; the graph attention grid evolution model includes an encoding layer, a multi-layer graph convolution layer, a multi-head attention layer and a decoding layer.

[0015] Among them, the core of the atlas attention grid evolution model is to introduce a grid area adaptive attention allocation mechanism. The grid area adaptive attention allocation mechanism dynamically adjusts the attention distribution through three key parameters: grid density parameter, grid merging level and historical evolution rate of the change matrix in the shoreline feature matrix, so as to effectively capture the shoreline change characteristics of different scales.

[0016] This paper constructs an initial shoreline model using a small-grid approximation method and, combined with a graph-based attention grid evolution model, accurately predicts island shoreline erosion risk. This method not only accurately represents complex shoreline morphology but also adaptively allocates attention resources based on grid size, effectively addressing the inaccuracy inherent in traditional techniques due to simple linear representations.

[0017] This method comprehensively considers multiple factors, including wave intensity, tidal amplitude, sea level rise rate, geological structure coefficient, and shoreline slope angle, to establish a more comprehensive physical model using coastal erosion dynamics equations. In particular, the introduced atlas attention grid evolution model automatically balances global changes in large-area grids with local details in small-area grids. Dynamically adjusting attention distribution through a grid-area adaptive attention allocation mechanism overcomes the shortcomings of traditional methods in representing multi-scale changes.

[0018] Through multi-level grid representation and area weight adjustment mechanism, the present invention successfully solves the technical problem in the existing technology that it is difficult to accurately predict the erosion risk of complex island coastlines, and provides a more reliable scientific basis for island coastline protection and disaster risk management. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of the method of the present invention.

[0020] Figure 2 Schematic diagram of the multi-level grid representation in Example 2.

[0021] Figure 3 This is a relationship diagram between shoreline erosion rate and wave action intensity in different areas in Example 2. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0023] like Figure 1 FIG. 1 is a flowchart of a method for predicting island shoreline erosion risk based on computer vision provided by the present invention. The method includes the following steps:

[0024] S01. Acquire high-resolution island remote sensing images and perform image preprocessing, including denoising, enhancement, registration, and segmentation, and use machine vision algorithms to extract island contour information;

[0025] S02. constructing an initial island coastline model based on a small grid approximation method, dividing the island boundary into a plurality of square grids with a side length of a first size value, to form an island coastline grid set;

[0026] S03, merging the small grids to find the global largest square grid, generating a multi-level grid representation, and storing all grid information as a shoreline feature matrix;

[0027] S04, collecting multiple time series of island remote sensing images, repeating steps S01 to S03 for each time point, and obtaining a shoreline feature matrix at the corresponding time point;

[0028] S05. Calculate the change matrix of the shoreline characteristic matrix at adjacent time points, construct a time series change matrix set, and analyze the shoreline evolution pattern;

[0029] S06. Establish a coastal erosion dynamics equation model, combining wave intensity, tidal amplitude, sea level rise rate, island geological structure coefficient, and shoreline slope angle to quantitatively calculate the erosion rate at different grid locations;

[0030] S07. Applying the atlas attention grid evolution model to analyze the shoreline change pattern, wherein the atlas attention grid evolution model adjusts the attention allocation mechanism through the grid area weight coefficient to optimize the feature extraction efficiency;

[0031] S08. Introduce the uncertainty analysis module, calculate the confidence interval of the prediction results, generate a multi-scale erosion risk assessment map, identify high-risk areas and output them as the island coastline erosion risk prediction results.

[0032] Among them, the small grid approximation method specifically represents the island boundary as a set of square grids with equal side lengths, and accurately expresses the complex coastline morphology by adjusting the grid density. Each grid records the existence status and attribute information of the coastline in the area.

[0033] Among them, the first size value is the initial grid side length determined based on the complexity of the island coastline and the resolution of the remote sensing image. The value range is 5 meters to 50 meters, which is used to ensure that the grid division can capture the coastline details while maintaining computational efficiency.

[0034] Among them, large-area grids specifically refer to grid units with an area greater than 625 square meters formed by merging multiple adjacent small grids. They are usually used to represent areas with relatively straight coastlines and obtain higher global change weights in the atlas attention grid evolution model.

[0035] Among them, small-area grids specifically refer to basic grid units with an area of ​​less than 100 square meters, which are usually distributed in areas with large coastline curvature or complex shapes. In the atlas attention grid evolution model, they are mainly responsible for the accurate expression of local detail changes.

[0036] Among them, the shoreline feature matrix is ​​specifically a two-dimensional data structure, whose element values ​​represent the attributes of the corresponding position grid, including whether it is a shoreline point, shoreline type, elevation value and stability index. The matrix dimension is determined by the grid division accuracy.

[0037] Among them, wave intensity is a physical quantity that characterizes the impact energy of waves on the shoreline. Its unit is kilowatts per meter. It is calculated through wave height, period and incident angle, and is an important input parameter of the coastal erosion dynamics equation model.

[0038] Among them, tidal amplitude specifically refers to the maximum value of water level change during the tidal cycle, measured in meters, obtained from measured data or tidal forecast models, and affects the vertical distribution range of the eroded area of ​​the coastline.

[0039] Among them, the sea level rise rate specifically refers to the height of the regional sea level rise each year, measured in millimeters per year. It is derived from long-term sea level monitoring data and is a key parameter for predicting long-term coastline changes.

[0040] Among them, the island geological structure coefficient is a dimensionless parameter that quantifies the anti-erosion ability of island coastline materials. It is determined comprehensively based on rock hardness, soil structure and vegetation coverage. The value range is 0 to 1. The larger the value, the stronger the anti-erosion ability.

[0041] Among them, the shoreline slope angle specifically refers to the angle formed by the shoreline and the horizontal plane, measured in degrees. It is obtained through calculation of elevation data and affects the distribution of wave energy on the shoreline and the erosion efficiency.

[0042] Among them, the coastal erosion dynamics equation is used to simulate the physical mechanism of coastal erosion. The input includes wave intensity, tidal amplitude, sea level rise rate, island geological structure coefficient and coastline slope angle, and the output is the coastline retreat distance and eroded material volume per unit time. The coastal erosion dynamics equation is based on the principle of conservation of mass and fluid dynamics theory, considering the balance between the impact force of waves on the coastline and the resistance of coastline materials, and obtains the erosion rate distribution of grids at different locations through integral calculation.

[0043] Among them, the grid area weight coefficient specifically refers to the weight parameter for allocating attention resources according to the grid area size in the atlas attention grid evolution model. It is calculated by the ratio of the grid area to the total area, and is used to balance the influence of global changes and local details in the prediction.

[0044] Among them, the specific structure of the atlas attention grid evolution model is a fusion architecture of a multi-layer graph convolutional neural network and a multi-head attention mechanism, which represents the shoreline grid as graph structure data, and the connection relationship between grids corresponds to the edge; the atlas attention grid evolution model includes an encoding layer, a multi-layer graph convolution layer, a multi-head attention layer and a decoding layer, wherein the multi-head attention mechanism assigns different weights according to the grid area, large-area grids obtain higher attention weights in global change prediction, and small-area grids maintain accurate expression in local detail change prediction; the core of the atlas attention grid evolution model is to introduce a grid area adaptive attention allocation mechanism, which dynamically adjusts the attention distribution through three key parameters: grid density parameters, grid merging levels and historical evolution rate of the change matrix in the shoreline feature matrix, so as to effectively capture the shoreline change characteristics of different scales.

[0045] Among them, the grid density parameter specifically refers to the number of grids per unit area, with the unit being per square kilometer. It is calculated from the first size value and is the basic parameter for adjusting the grid area adaptive attention allocation mechanism.

[0046] Among them, the grid merging level specifically refers to the number of levels at which small grids are gradually merged to form a large grid. The value range is 1 to 10, which is used to characterize the complexity of the grid structure and affect the attention allocation strategy.

[0047] Among them, the historical evolution rate of the change matrix specifically refers to the speed at which the shoreline characteristic matrix changes at adjacent time points. The unit is square meters per year. It is obtained through historical data analysis and is an important basis for predicting future change trends.

[0048] The steps for establishing the training dataset for the atlas attention grid evolution model include collecting historical remote sensing image sequences of typical island coastlines around the world and annotating the changes; preprocessing the original images and constructing a grid representation according to a unified standard; calculating the change matrix between different time points and annotating the influencing factors; constructing supervised learning labels based on measured erosion results; and dividing the data into training, validation, and test sets to ensure a balanced data distribution and coverage of a variety of island types and climatic conditions.

[0049] Among them, the training steps of the graph attention grid evolution model specifically include initializing model parameters; optimizing model parameters according to the mini-batch stochastic gradient descent method; using the cross-entropy loss function to evaluate the difference between the predicted results and the true labels; introducing regularization technology to prevent overfitting; implementing the early stopping strategy to avoid overtraining; setting weight coefficients for different island types; improving the generalization ability of the model in new island scenarios through transfer learning methods; freezing the feature extraction layer parameters after training, and retaining the key layer parameters for fine-tuning.

[0050] Among them, the multi-scale erosion risk assessment map is specifically a risk distribution map generated based on different spatial scales and time spans. It uses color gradients to represent the degree of erosion risk and combines geographic information system technology to achieve intuitive display and spatial analysis of risk areas.

[0051] The specific implementation of the above steps is described in detail below.

[0052] Step S01 is implemented by acquiring high-resolution island remote sensing images and performing a series of image preprocessing operations to extract island contour information. First, the remote sensing image is denoised using a wavelet transform denoising algorithm to eliminate sensor noise and atmospheric interference. Next, histogram equalization is applied to enhance image contrast, clarifying the boundary between the island and the sea. A feature point matching algorithm is then used to perform multi-temporal image registration, ensuring precise spatial alignment of images at different time points. Finally, the threshold-based OTSU segmentation algorithm is used to separate the island from the sea, with a threshold selected between 0.4 and 0.6. Morphological opening and closing operations are then used to optimize the segmentation results. Finally, the Canny edge detection algorithm is used to extract the island contours, with a high-low threshold ratio of 2:1. The recommended high threshold is 1.5 times the image gradient mean. This step aims to provide accurate island boundary information for subsequent shoreline meshing, ensuring the accuracy and reliability of shoreline extraction.

[0053] The specific implementation of step S02 is to construct an initial model of the island coastline using a small grid approximation method. First, determine an appropriate first size value as the grid side length. Based on the complexity of the island coastline and the resolution of the remote sensing image, select an appropriate value within the range of 5 to 50 meters. Complex coastlines should be 5 to 15 meters, while straight coastlines can be 20 to 50 meters. Then, based on the island outline extracted in step S01, set sampling points at fixed intervals along the coastline, with the sampling interval equal to 1 / 2 of the first size value. Next, generate a square grid with a side length of the first size value, centered at each sampling point. Determine whether the grid contains the coastline point. If so, include the grid in the coastline grid set. Finally, sort and number the grids, establish a topological relationship between the grids, and record the spatial position and adjacent grid information of each grid. The purpose of this step is to discretize the continuous island coastline into a grid representation that is easy for computer processing, laying the foundation for subsequent multi-level grid construction.

[0054] The specific implementation of step S03 involves merging small grids to generate a multi-level grid representation. First, a maximum square detection algorithm is applied to search for adjacent grids that can be merged from the initial grid set. The criteria for this is that the adjacent grid attributes have a similarity greater than 0.85 and that the merged grids form a larger square. Then, grids are merged using a bottom-up strategy, with small grids preferentially merged into medium grids, which are then merged into large grids, ensuring that the side lengths of the merged grids are integer multiples of the initial grid lengths. Next, a hierarchical grid tree structure is constructed to record the inclusion relationships between grids, with large grids serving as parent nodes and the contained small grids as child nodes. Finally, all grid information is converted into a shoreline feature matrix. The matrix dimensions correspond to the size of the original remote sensing image, and the matrix elements record the attribute information of the grid at the corresponding location, including grid type (shoreline / non-shoreline), grid size (small / medium / large), elevation value, and stability index (ranging from 0 to 1). The purpose of this step is to reduce the redundancy of the shoreline representation while preserving detailed information in key areas and improving subsequent computational efficiency.

[0055] The specific implementation method of step S04 is to collect multiple time series of island remote sensing images and repeat the operation process of steps S01 to S03 for each time point. First, collect a series of island remote sensing images spanning at least 5 years, with a time interval of 3 months to 1 year recommended to ensure that seasonal and interannual changes are captured; then, apply the preprocessing method of step S01 to each time series image to ensure that all image processing standards are consistent; then, perform the small grid approximation method of step S02 on each processed image to construct the initial coastline model of the corresponding time point; finally, apply the grid merging strategy of step S03 to generate a multi-level grid representation and coastline feature matrix for each time point, and keep the matrix dimension consistent to facilitate subsequent comparative analysis. The purpose of this step is to construct a time series coastline feature dataset to provide basic data for analyzing the laws of coastline evolution.

[0056] The specific implementation of step S05 is to calculate the change matrix of the shoreline characteristic matrix at adjacent time points and construct a set of time series change matrices. First, a matrix difference operation is used to calculate the difference between the shoreline characteristic matrices at adjacent time points to obtain an initial change matrix. Then, a spatial filtering algorithm is applied to remove noise from the change matrix. The filter window size is set to 3×3 to 5×5, and change signals with a confidence level exceeding 0.75 are retained. Next, the change rate of each grid location is calculated, including the area change rate and morphological change rate, and significant change areas are marked. The change rate threshold is set to ±10% of the previous moment value. Finally, the change matrices corresponding to all time points are organized into a time series data structure, and the shoreline evolution pattern is analyzed to identify erosion acceleration zones, erosion slowdown zones, and stable zones. The judgment criteria are an average annual retreat distance greater than 1 meter, an average annual retreat distance less than 0.2 meter, and an average annual change distance less than 0.05 meter. The purpose of this step is to quantify the spatiotemporal change characteristics of the shoreline and provide a dynamic change basis for subsequent erosion risk prediction.

[0057] The specific implementation of step S06 involves establishing a coastal erosion dynamics model to quantitatively calculate erosion rates at different grid locations. First, model input parameters are collected, including wave intensity (unit: kilowatts per meter, typical range: 1 to 15), tidal amplitude (unit: meter, typical range: 0.5 to 5), sea level rise rate (unit: millimeters per year, typical range: 2 to 10), island geological structure coefficient (dimensionless, range: 0 to 1), and shoreline slope angle (unit: degree, typical range: 5 to 45). Then, based on the principle of mass conservation and fluid dynamics theory, a dynamic equation describing the shoreline erosion process is established. This equation considers the balance between wave energy input and geological structural resistance. Next, the erosion equation is applied to calculate erosion rates for different grid types, using a simplified calculation mode for large grids and a refined calculation mode for small grids. The calculation accuracy threshold is set to 0.1 meters per year. Finally, a shoreline grid erosion rate distribution map is generated, visually representing the erosion risk level at different locations. The purpose of this step is to quantify the erosion process through a physical model and provide theoretical support for risk prediction.

[0058] Step S07 is implemented by applying a graph-attention grid evolution model to analyze shoreline evolution patterns. First, shoreline grid data is converted into a graph structure, with grid centers serving as nodes and connections between adjacent grids as edges. Next, a graph-attention network model is constructed, comprising an encoding layer, multiple graph convolution layers (3 to 5 layers), a multi-head attention layer (4 to 8 heads), and a decoding layer. Next, an area-based attention allocation mechanism is introduced, with large grids (greater than 625 square meters) receiving a higher global attention weight (weight coefficient 0.6 to 0.8), while small grids (less than 100 square meters) are focused on local details (weight coefficient 0.2 to 0.4). Finally, the trained model is used to predict future shoreline evolution trends, outputting three core metrics: grid position change, morphological change, and rate change. This step aims to integrate data-driven and physical model approaches to improve prediction accuracy and adaptability.

[0059] The specific implementation of step S08 involves introducing an uncertainty analysis module to calculate confidence intervals for the prediction results and generate a multi-scale erosion risk assessment map. First, using a Monte Carlo simulation method, key input parameters are randomly perturbed within a range of ±2 times the parameter standard deviation, generating 1,000 to 5,000 sets of simulation parameters. Next, a graph-based attention grid evolution model is run to obtain multiple sets of prediction results, calculating the probability distribution of the predicted value for each grid location. Next, a 90% confidence interval is determined based on the probability distribution, and the results are visualized as a risk level map. A five-level risk scale (extremely high risk, high risk, moderate risk, low risk, and very low risk) is used, with risk thresholds of annual erosion rates greater than 2 meters, 1 to 2 meters, 0.5 to 1 meter, 0.1 to 0.5 meters, and less than 0.1 meter, respectively. Finally, a multi-scale erosion risk assessment map is generated, including short-term (1 to 3 years), medium-term (3 to 10 years), and long-term (10 to 30 years) risk assessments, identifying high-risk areas requiring priority protection. The purpose of this step is to quantify the uncertainty of the predictions and provide support for risk-based decision-making.

[0060] The detailed structure of the graph attention grid evolution model and the specific implementation method of its training dataset are as follows: The core structure of the graph attention grid evolution model adopts an architecture that combines graph neural network and attention mechanism, including four main components. First, the encoding layer is responsible for converting the input shoreline feature matrix into a graph structure representation. Each grid corresponds to a graph node, and the node features include grid type, area, location coordinates and historical change rate. Adjacent grids are connected by edges, and the edge weights are calculated based on the similarity between grids, with the similarity threshold set to 0.7; then, the multi-layer graph convolution layer consists of a three-layer graph convolutional network, each layer contains 64 convolution kernels with a size of 3×3, and the activation function uses ReLU. Batch normalization and residual connections are added after each layer to capture the spatial dependencies between grids; then, the multi-head attention layer contains 8 attention heads, each of which independently learns the importance distribution of different feature subspaces. The attention weight is adaptively adjusted by the grid area. The adjustment coefficient is the logarithmic ratio of the area to the baseline area (100 square meters) to balance global changes and local details; finally, the decoding layer converts the graph representation back to a grid matrix format and outputs the predicted changes of each grid position, including position offset and morphological changes. The specific implementation of the training dataset includes five core steps: First, a series of historical remote sensing images of 50 to 100 representative island coastlines worldwide is collected, spanning at least 10 years, with a spatial resolution better than 5 meters, covering diverse climatic conditions and geological types. Second, all images are standardized and preprocessed, including atmospheric, geometric, and radiometric corrections, to ensure consistent data quality. Third, a unified grid representation is constructed, with grid sizes comparable across islands, and grid attribute information is recorded at each time point. Fourth, a change matrix is ​​calculated for adjacent time points, and environmental factor data, including wave, tidal, and meteorological data, is collected for the corresponding time periods. Fifth, supervised learning labels are annotated based on the measured coastline changes, and the training, validation, and test sets are divided into training, validation, and test sets in a ratio of 7:2:1 to ensure balanced and representative data distribution. The model is trained using mini-batch stochastic gradient descent with batch sizes ranging from 32 to 64, an initial learning rate of 0.001, and a dynamically adjusted cosine annealing strategy. The number of training epochs ranges from 100 to 200, with an early stopping patience parameter set to 20 epochs.

[0061] The mathematical model or calculation process involved in the present invention is described in detail below.

[0062] In step S05, the change matrix of the shoreline characteristic matrix at adjacent time points is calculated, which is specifically expressed as follows:

[0063] ;

[0064] Where, For the moment The corresponding change matrix; For the moment The shoreline characteristic matrix; For the moment The shoreline characteristic matrix.

[0065] The calculation formula of the rate of change matrix is ​​as follows:

[0066] ;

[0067] Where, is the rate of change matrix; is the time interval in years; The zero elements in are represented by a constant smaller than the smallest non-zero value. (The recommended value is ) instead to avoid division by zero errors.

[0068] In step S06, the core formula of the coastal erosion dynamics equation model is expressed as follows:

[0069] ;

[0070] Where, Grid position Erosion rate in meters per year; is the wave action intensity, in kilowatts / meter; is the tidal amplitude in meters; is the rate of sea level rise in millimeters per year; is the island geological structure coefficient, dimensionless, ranging from 0 to 1; is the shoreline slope angle, in degrees; is the erosion coefficient, the default value is 0.25, calibrated by historical data; is the geological resistance coefficient, the default value is 0.75, obtained through experiments; It is a stabilization factor with a default value of 0.05, which is used to avoid the denominator being zero.

[0071] The method for obtaining the parameters in the erosion equation is:

[0072] Obtained through wave observation data or wave model, the calculation formula is:

[0073] ;

[0074] Where, is the density of seawater, the default value is 1025 kg / m3; is the acceleration due to gravity, the default value is 9.8 m / s²; is the significant wave height in meters; is the wave group velocity in meters per second; is the wave incident angle, in degrees.

[0075] It can be obtained through tidal observation station data or tidal forecast models, and can be obtained by spatial interpolation of the nearest tidal station data. The calculation formula is:

[0076] ;

[0077] Where, For the Tidal amplitude at each tide station; is the spatial weight, and the calculation formula is:

[0078] ;

[0079] Where, Grid position To The distance between tide stations in kilometers.

[0080] Obtained through sea level monitoring data, the calculation formula is:

[0081] ;

[0082] Where, is the global average sea level rise rate, with a default value of 3.3 mm / year; It is the regional sea level change, obtained through satellite altimetry data; The local sea level changes are obtained by land measurement and GPS data, taking into account the vertical movement of the crust and the impact of human activities.

[0083] Obtained through geological surveys and rock hardness tests, the calculation formula is:

[0084] ;

[0085] Where, is the rock hardness factor, ranging from 0 to 1; is the soil structure factor, ranging from 0 to 1; is the vegetation cover factor, ranging from 0 to 1; is the weight coefficient, satisfying , the default values ​​are 0.5, 0.3, and 0.2 respectively.

[0086] Calculated through elevation data, the formula is:

[0087] ;

[0088] Where, is the elevation difference along the coastline of the grid, in meters; is the corresponding horizontal distance in meters; The coefficient for converting radians to degrees.

[0089] In step S07, the grid area weight coefficient calculation formula of the atlas attention grid evolution model is:

[0090] ;

[0091] Where, For Grid The area weight coefficient of is the grid area in square meters; is the minimum grid area, the default value is 25 square meters; is the maximum grid area, the default value is 625 square meters; is the minimum weight, the default value is 0.2; is the maximum weight, the default value is 0.8.

[0092] The attention distribution calculation formula of the grid area adaptive attention allocation mechanism is:

[0093] ;

[0094] Where, For Grid Attention score; is the grid density parameter weight, and the calculation formula is:

[0095] ;

[0096] Where, For Grid The grid density of the area, in units of grids per square kilometer; The default value is 400 per square kilometer for the optimal grid density. It is an adjustment parameter and its default value is 200.

[0097] is the historical change rate weight, and the calculation formula is:

[0098] ;

[0099] Where, For Grid The historical rate of change of , in square meters per year; is the weight enhancement coefficient, the default value is 0.5; It is a hyperbolic tangent function, which is used to limit the range of weight change; is the scaling factor, the default value is 0.1.

[0100] In step S08, the risk score is calculated as follows:

[0101] ;

[0102] Where, For Grid Comprehensive risk score; Represents short term ( ), mid-term ( ) and long term ( ) Risk score; is the time scale weight, satisfying , the default values ​​are 0.5, 0.3, and 0.2 respectively.

[0103] The formula for calculating the risk score at each time scale is:

[0104] ;

[0105] Where, is the probability of occurrence, ranging from 0 to 1; is the impact intensity, ranging from 0 to 10; is vulnerability, ranging from 0 to 1.

[0106] The confidence interval is calculated as:

[0107] ;

[0108] Where, For Grid confidence intervals for erosion rates; is the average predicted erosion rate in meters per year; is the standard deviation; The significance level is The critical value of the standard normal distribution, the corresponding 95% confidence interval It is 1.96.

[0109] The reason for using a power relationship in the coastal erosion dynamics equation is that wave energy is proportional to the square of the wave height, which is consistent with the basic principles of fluid dynamics; the 0.5 power of the tidal amplitude indicates that the tidal influence increases with the amplitude, but the growth rate slows down, which is consistent with actual observations; the 0.3 power of the sea level rise rate indicates that its promoting effect on erosion exists, but is not as significant as the wave effect, which is consistent with the results of coastal dynamics research. The cosine function is used in the denominator to describe the influence of the slope angle, reflecting the phenomenon that steep coastlines receive less wave energy and have relatively low erosion rates. Adding a stability factor This equation avoids numerical instability caused by the denominator approaching zero. It comprehensively considers ocean dynamics (waves, tides) and geological factors (geological structure, slope), and can more comprehensively describe the physical mechanism of coastal erosion.

[0110] The grid area weight coefficient adopts a logarithmic relationship because the range of grid area variation is large. A direct linear relationship would result in excessively high weights for large grid areas. A logarithmic transformation allows for a more rational allocation of attention resources, allowing the model to focus on large-scale trends while not neglecting local details. The grid density parameter weight adopts a Gaussian function, reflecting the fact that grid density has an optimal value. Excessively high or low grid density will reduce model performance. The historical change rate weight adopts a hyperbolic tangent function, which can pay more attention to areas with significant historical changes while keeping the function value bounded, thus helping the model capture key areas of change.

[0111] The risk scoring model uses the product of probability, impact intensity, and vulnerability, conforming to the basic framework of risk assessment and comprehensively considering the likelihood, severity, and tolerance of erosion events. Confidence intervals are calculated using the standard normal distribution assumption. Monte Carlo simulations are used to determine the distribution characteristics of the predicted results, providing quantitative uncertainty support for decision-making.

[0112] Shoreline characteristic matrix The elements of are represented as follows:

[0113] ;

[0114] Where, For the moment In position The eigenvector at ; The grid type is 0 (non-shoreline) or 1 (shoreline); is the grid size, which can be 1 (small), 2 (medium) or 3 (large); is the elevation value in meters; is a stability index ranging from 0 to 1.

[0115] It's important to note that the coastal erosion dynamics equation is based on physical mechanisms, taking into account the balance between wave impact and coastal resistance. The wave intensity term is raised to the power of 1.5, emphasizing the significant impact of extreme wave events on erosion; the tidal amplitude term is raised to the power of 0.5, reflecting the nonlinear contribution of tides to erosion; and the sea level rise rate term is raised to the power of 0.3, reflecting the cumulative effect of long-term sea level change on erosion. These three dynamic factors together constitute the driving force of erosion, while the geological structure coefficient and slope angle represent the resistance. The choice of these power relationships in the equation is based on extensive field data fitting and theoretical analysis, and can better reflect the relative contribution of each factor to the erosion process. Compared to existing technologies, this equation considers both short-term (waves, tides) and long-term (sea level rise) dynamics, and incorporates regional characteristics (geological structure, slope), resulting in greater adaptability and predictive power.

[0116] It is important to note that the grid area adaptive attention allocation mechanism determines the attention allocation of each grid by multiplying three key weight factors, effectively capturing the characteristics of changes at different scales. The grid density parameter weight adopts a Gaussian function, which ensures computational efficiency while ensuring accuracy. The historical change rate weight uses a hyperbolic tangent function to enhance attention to areas with significant historical changes, improving the model's sensitivity to key regions. Compared to traditional fixed-weight models, this mechanism can dynamically adjust attention distribution based on the characteristics of different regions, improving the model's adaptability and predictive accuracy.

[0117] Specifically, the principle of the present invention is as follows: The core technical principle of the present invention is to combine complex coastline representation with a multi-scale attention mechanism, and to achieve accurate prediction through a small grid approximation method and a graph attention grid evolution model. First, the small grid approximation method breaks through the traditional linear representation mode and represents the island coastline as a set of square grids with equal sides. This representation method can flexibly adjust the grid density according to the complexity of the coastline, thereby accurately capturing the detailed features of the coastline. By merging small grids, a multi-level grid representation is formed, which not only retains local detail information but also extracts global structural features.

[0118] Secondly, the coastal erosion dynamics equation model established in this paper, based on the principle of mass conservation and fluid dynamics theory, quantifies the balance between wave impact force and shoreline material resistance into a mathematical model. This model not only considers conventional factors such as wave intensity, tidal amplitude, and sea level rise rate, but also incorporates island geological structure coefficients and shoreline slope angles, making the model more consistent with physical reality and providing a theoretical explanation for the differences in shoreline erosion rates in different regions.

[0119] The most innovative is the Graph Attention Grid Evolution Model, which represents shoreline grids as graph-structured data. Through a fusion architecture of a multi-layer graph convolutional neural network and a multi-head attention mechanism, it achieves deep learning of shoreline variation patterns. Its core is a grid-area adaptive attention allocation mechanism. This dynamically adjusts attention distribution through three key parameters: grid density, grid merging levels, and the historical evolution rate of the change matrix. This allows large grids to receive higher global change prediction weights, while small grids retain accurate representation of local details. This adaptive mechanism aligns with human visual attention allocation patterns and effectively balances global and local information, thereby improving prediction accuracy.

[0120] Furthermore, the present invention incorporates an uncertainty analysis module to calculate confidence intervals for prediction results and generate multi-scale erosion risk assessment maps, further enhancing the reliability and practicality of the predictions. This comprehensive technical solution fundamentally addresses the core technical issues in predicting erosion risk along complex island coastlines.

[0121] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0122] Step S01 is implemented by acquiring high-resolution island remote sensing images and performing a series of image preprocessing operations to extract island contour information. First, the remote sensing image is denoised using a wavelet transform denoising algorithm to eliminate sensor noise and atmospheric interference. Next, histogram equalization is applied to enhance image contrast, clarifying the boundary between the island and the sea. A feature point matching algorithm is then used to perform multi-temporal image registration, ensuring precise spatial alignment of images at different time points. Finally, the threshold-based OTSU segmentation algorithm is used to separate the island from the sea, with a threshold selected between 0.4 and 0.6. Morphological opening and closing operations are then used to optimize the segmentation results. Finally, the Canny edge detection algorithm is used to extract the island contours, with a high-low threshold ratio of 2:1. The recommended high threshold is 1.5 times the image gradient mean. This step aims to provide accurate island boundary information for subsequent shoreline meshing, ensuring the accuracy and reliability of shoreline extraction.

[0123] The specific implementation of step S02 is to construct an initial model of the island coastline using a small grid approximation method. First, determine an appropriate first size value as the grid side length. Based on the complexity of the island coastline and the resolution of the remote sensing image, select an appropriate value within the range of 5 to 50 meters. Complex coastlines should be 5 to 15 meters, while straight coastlines can be 20 to 50 meters. Then, based on the island outline extracted in step S01, set sampling points at fixed intervals along the coastline, with the sampling interval equal to 1 / 2 of the first size value. Next, generate a square grid with a side length of the first size value, centered at each sampling point. Determine whether the grid contains the coastline point. If so, include the grid in the coastline grid set. Finally, sort and number the grids, establish a topological relationship between the grids, and record the spatial position and adjacent grid information of each grid. The purpose of this step is to discretize the continuous island coastline into a grid representation that is easy for computer processing, laying the foundation for subsequent multi-level grid construction.

[0124] The specific implementation method of step S03 is to merge small grids to generate a multi-level grid representation. First, the maximum square detection algorithm is applied to find adjacent grids that can be merged from the initial grid set. The judgment condition is that the attribute similarity of adjacent grids is higher than 0.85 and a larger square is formed after merging; then, the grids are merged according to the bottom-up strategy, and small grids are preferentially merged into medium grids, and medium grids are merged into large grids, keeping the side length of the merged grid an integer multiple of the side length of the initial grid; then, a hierarchical grid tree structure is constructed to record the inclusion relationship between grids, with the large grid as the parent node and the contained small grids as the child nodes; finally, all grid information is converted into a shoreline feature matrix, and the matrix elements record the attribute information of the grids at the corresponding positions. Shoreline feature matrix The elements of are represented as follows:

[0125] Where, For the moment In position The eigenvector at ; The grid type is 0 (non-shoreline) or 1 (shoreline); is the grid size, which can be 1 (small), 2 (medium) or 3 (large); is the elevation value in meters; is the stability index, ranging from 0 to 1. The purpose of this step is to reduce the redundancy of shoreline representation while retaining the detailed information of key areas and improving the efficiency of subsequent calculations.

[0126] The specific implementation method of step S04 is to collect multiple time series of island remote sensing images and repeat the operation process of steps S01 to S03 for each time point. First, collect a series of island remote sensing images spanning at least 5 years, with a time interval of 3 months to 1 year recommended to ensure that seasonal and interannual changes are captured; then, apply the preprocessing method of step S01 to each time series image to ensure that all image processing standards are consistent; then, perform the small grid approximation method of step S02 on each processed image to construct the initial coastline model of the corresponding time point; finally, apply the grid merging strategy of step S03 to generate a multi-level grid representation and coastline feature matrix for each time point, and keep the matrix dimension consistent to facilitate subsequent comparative analysis. The purpose of this step is to construct a time series coastline feature dataset to provide basic data for analyzing the laws of coastline evolution.

[0127] The specific implementation of step S05 is to calculate the change matrix of the shoreline characteristic matrix at adjacent time points and construct a time series change matrix set. First, the difference between the shoreline characteristic matrices at adjacent time points is calculated using matrix difference operation to obtain the initial change matrix. The specific calculation formula is: Where, For the moment The corresponding change matrix; For the moment The shoreline characteristic matrix; For the moment The shoreline feature matrix is ​​then generated. A spatial filtering algorithm is then applied to remove noise from the change matrix. The filter window size is set to 3×3 to 5×5, and change signals with a confidence level exceeding 0.75 are retained. Next, the change rate of each grid position is calculated, including the area change rate and the morphological change rate. The calculation formula for the change rate matrix is: Where, is the rate of change matrix; is the time interval in years; The zero elements in are represented by a constant smaller than the smallest non-zero value. (The recommended value is ) to avoid division by zero errors. Finally, areas of significant change are marked, with the change rate threshold set at ±10% of the previous moment's value. The change matrix corresponding to all time points is organized into a time series data structure. The shoreline evolution pattern is analyzed to identify areas of accelerated erosion, areas of slowed erosion, and stable areas. The criteria for this are an average annual retreat distance greater than 1 meter, an average annual retreat distance less than 0.2 meter, and an average annual change distance less than 0.05 meter. The purpose of this step is to quantify the spatiotemporal characteristics of shoreline changes and provide a dynamic basis for subsequent erosion risk prediction.

[0128] The specific implementation of step S06 is to establish a coastal erosion dynamics equation model and quantitatively calculate the erosion rate at different grid locations. First, collect the model input parameters, including wave action intensity (unit: kilowatts per meter, typical value range 1 to 15), tidal amplitude (unit: meter, typical value range 0.5 to 5), sea level rise rate (unit: millimeter per year, typical value range 2 to 10), island geological structure coefficient (dimensionless, value range 0 to 1) and shoreline slope angle (unit: degree, typical value range 5 to 45); then, based on the principle of conservation of mass and fluid dynamics theory, establish a dynamic equation to describe the shoreline erosion process. This equation takes into account the balance between wave energy input and geological structure resistance. The core formula of the coastal erosion dynamics equation model is expressed as follows: Where, Grid position Erosion rate in meters per year; is the wave action intensity, in kilowatts / meter; is the tidal amplitude in meters; is the rate of sea level rise in millimeters per year; is the island geological structure coefficient, dimensionless, ranging from 0 to 1; is the shoreline slope angle, in degrees; is the erosion coefficient, the default value is 0.25, calibrated by historical data; is the geological resistance coefficient, the default value is 0.75, obtained through experiments; It is a stabilization factor with a default value of 0.05, which is used to avoid the denominator being zero. The method for obtaining the parameters in the erosion equation is: Obtained through wave observation data or wave model, the calculation formula is: Where, is the density of seawater, the default value is 1025 kg / m3; is the acceleration due to gravity, the default value is 9.8 m / s²; is the significant wave height in meters; is the wave group velocity in meters per second; is the wave incident angle, in degrees. It can be obtained through tidal observation station data or tidal forecast models, and can be obtained by spatial interpolation of the nearest tidal station data. The calculation formula is: Where, For the Tidal amplitude at each tide station; is the spatial weight, and the calculation formula is: Where, Grid position To The distance between tide stations in kilometers. Obtained through sea level monitoring data, the calculation formula is: Where, is the global average sea level rise rate, with a default value of 3.3 mm / year; It is the regional sea level change, obtained through satellite altimetry data; The local sea level changes are obtained by land measurement and GPS data, taking into account the vertical movement of the crust and the impact of human activities. Obtained through geological surveys and rock hardness tests, the calculation formula is: Where, is the rock hardness factor, ranging from 0 to 1; is the soil structure factor, ranging from 0 to 1; is the vegetation cover factor, ranging from 0 to 1; is the weight coefficient, satisfying The default values ​​are 0.5, 0.3, and 0.2 respectively. Calculated through elevation data, the formula is: Where, is the elevation difference along the coastline of the grid, in meters; is the corresponding horizontal distance in meters; is the coefficient for converting radians to degrees. Next, erosion equations are applied to different grid types to calculate erosion rates. Large grids use a simplified calculation mode, while small grids use a refined calculation mode. The calculation accuracy threshold is set to 0.1 meters per year. Finally, a shoreline grid erosion rate distribution map is generated, visually representing the erosion risk level at different locations. This step aims to quantify the erosion process through physical models and provide theoretical support for risk prediction.

[0129] The specific implementation method of step S07 is to apply the graph attention grid evolution model to analyze the law of shoreline changes. First, the shoreline grid data is converted into a graph structure representation, with the grid center point as the graph node and the connection relationship between adjacent grids as the graph edge; then, a graph attention network model is constructed, which includes an encoding layer, a multi-layer graph convolution layer (the number of layers is 3 to 5), a multi-head attention layer (the number of heads is 4 to 8) and a decoding layer; then, an attention allocation mechanism based on grid area is introduced, and large-area grids (greater than 625 square meters) are assigned higher global attention weights (weight coefficients are 0.6 to 0.8), and small-area grids (less than 100 square meters) focus on local detail expression (weight coefficients are 0.2 to 0.4). The grid area weight coefficient calculation formula of the graph attention grid evolution model is: Where, For Grid The area weight coefficient of is the grid area in square meters; is the minimum grid area, the default value is 25 square meters; is the maximum grid area, the default value is 625 square meters; is the minimum weight, the default value is 0.2; is the maximum weight, and the default value is 0.8. The attention distribution calculation formula of the grid area adaptive attention allocation mechanism is: Where, For Grid Attention score; is the grid density parameter weight, and the calculation formula is: Where, For Grid The grid density of the area, in units of grids per square kilometer; The default value is 400 per square kilometer for the optimal grid density. It is an adjustment parameter and its default value is 200. is the historical change rate weight, and the calculation formula is: Where, For Grid The historical rate of change of , in square meters per year; is the weight enhancement coefficient, the default value is 0.5; It is a hyperbolic tangent function, which is used to limit the range of weight change; is the scaling factor, with a default value of 0.1. Finally, the trained model is used to predict future shoreline evolution trends, outputting three core metrics: grid position change, morphological change, and rate change. This step aims to integrate data-driven and physical modeling approaches to improve prediction accuracy and adaptability.

[0130] The specific implementation of step S08 is to introduce an uncertainty analysis module, calculate the confidence interval of the prediction results, and generate a multi-scale erosion risk assessment map. First, using the Monte Carlo simulation method, the key input parameters are randomly perturbed within a range of ±2 times the parameter standard deviation, generating 1000 to 5000 sets of simulation parameters. Then, the atlas attention grid evolution model is run to obtain multiple sets of prediction results and calculate the probability distribution of the predicted value for each grid location. Then, based on the probability distribution, a 90% confidence interval is determined. The confidence interval calculation formula is: Where, For Grid confidence intervals for erosion rates; is the average predicted erosion rate in meters per year; is the standard deviation; The significance level is The critical value of the standard normal distribution, the corresponding 95% confidence interval The result is then visualized as a risk level graph, and the risk score is calculated as: Where, For Grid Comprehensive risk score; Represents short term ( ), mid-term ( ) and long term ( ) Risk score; is the time scale weight, satisfying The default values ​​are 0.5, 0.3, and 0.2 respectively. The risk score calculation formula for each time scale is: Where, is the probability of occurrence, ranging from 0 to 1; is the impact intensity, ranging from 0 to 10; Vulnerability is measured on a scale of 0 to 1. A five-level risk scale (extremely high risk, high risk, moderate risk, low risk, and very low risk) is used, with risk thresholds set at annual erosion rates greater than 2 meters, 1 to 2 meters, 0.5 to 1 meter, 0.1 to 0.5 meters, and less than 0.1 meters. Finally, a multi-scale erosion risk assessment map is generated, including short-term (1 to 3 years), medium-term (3 to 10 years), and long-term (10 to 30 years) risk assessments, identifying high-risk areas requiring priority protection. This step aims to quantify the uncertainty of the predictions and support risk-based decision-making.

[0131] The detailed structure of the graph attention grid evolution model and the specific implementation method of its training dataset are as follows: The core structure of the graph attention grid evolution model adopts an architecture that combines graph neural network and attention mechanism, including four main components. First, the encoding layer is responsible for converting the input shoreline feature matrix into a graph structure representation. Each grid corresponds to a graph node, and the node features include grid type, area, location coordinates and historical change rate. Adjacent grids are connected by edges, and the edge weights are calculated based on the similarity between grids, with the similarity threshold set to 0.7; then, the multi-layer graph convolution layer consists of a three-layer graph convolutional network, each layer contains 64 convolution kernels with a size of 3×3, and the activation function uses ReLU. Batch normalization and residual connections are added after each layer to capture the spatial dependencies between grids; then, the multi-head attention layer contains 8 attention heads, each of which independently learns the importance distribution of different feature subspaces. The attention weight is adaptively adjusted by the grid area. The adjustment coefficient is the logarithmic ratio of the area to the baseline area (100 square meters) to balance global changes and local details; finally, the decoding layer converts the graph representation back to a grid matrix format and outputs the predicted changes of each grid position, including position offset and morphological changes.

[0132] The specific implementation method for establishing the training dataset includes five core steps: first, collecting historical remote sensing image sequences of 50 to 100 typical island coastlines around the world, with a time span of no less than 10 years, a spatial resolution better than 5 meters, and covering different climatic conditions and geological types; second, performing standardized preprocessing on all images, including atmospheric correction, geometric correction, and radiation correction, to ensure data quality consistency; third, constructing a unified grid representation with comparable grid sizes across different islands and recording grid attribute information at each time point; fourth, calculating the change matrix of adjacent time points and collecting environmental factor data for the corresponding time period, including wave, tide, and meteorological data; fifth, annotating supervised learning labels based on the measured coastline change results, and dividing the training set, validation set, and test set into a ratio of 7:2:1 to ensure that the data distribution is balanced and representative.

[0133] The model was trained using mini-batch stochastic gradient descent, with batch sizes set to 32 to 64. The initial learning rate was 0.001, dynamically adjusted using a cosine annealing strategy. The number of training epochs ranged from 100 to 200, with an early stopping patience parameter set to 20. During training, the cross-entropy loss function was used to evaluate the difference between the predicted results and the true labels. L2 regularization was introduced to prevent overfitting, with the regularization coefficient set between 0.0001 and 0.001. Weighting coefficients were set for different island types, with a focus on erosion-prone types, such as a weight of 1.5 for sandy islands and 0.8 for rocky islands. Transfer learning was used to improve the model's generalization to new island scenarios. The pretrained model was trained on a global dataset and then fine-tuned on the target island data. After training, the feature extraction layer parameters were frozen, retaining the parameters of key layers for fine-tuning to adapt to the specific conditions of each island.

[0134] In this embodiment, the power relationship is used in the coastal erosion dynamics equation because wave energy is proportional to the square of the wave height, which is consistent with the basic principles of fluid dynamics; the 0.5 power of the tidal amplitude indicates that the tidal influence increases with the increase of the amplitude, but the growth rate slows down, which is consistent with the actual observation results; the 0.3 power of the sea level rise rate indicates that its promoting effect on erosion exists, but is not as significant as the wave effect, which is consistent with the results of coastal dynamics research. The cosine function is used in the denominator to describe the influence of the slope angle, reflecting the phenomenon that steep coastlines receive less wave energy and have a relatively low erosion rate. Adding a stability factor This equation avoids numerical instability caused by the denominator approaching zero. It comprehensively considers ocean dynamics (waves, tides) and geological factors (geological structure, slope), and can more comprehensively describe the physical mechanism of coastal erosion.

[0135] The reason why the grid area weight coefficient in this embodiment adopts a logarithmic relationship is that the range of grid area variation is large. A direct linear relationship will result in an excessively high weight for large-area grids. After logarithmic transformation, attention resources can be allocated more reasonably, so that the model can pay attention to large-scale change trends while not ignoring local details. The grid density parameter weight adopts the form of a Gaussian function, which reflects the characteristic that there is an optimal value for grid density. Too high or too low grid density will reduce model performance. The historical change rate weight adopts a hyperbolic tangent function, which can pay more attention to areas with significant historical changes while keeping the function value bounded, which helps the model capture key change areas.

[0136] The risk scoring model in this example uses the product of probability, impact intensity, and vulnerability, conforming to the basic framework of risk assessment and comprehensively considering the likelihood, severity, and tolerance of erosion events. Confidence intervals are calculated using the standard normal distribution assumption, and Monte Carlo simulations are used to determine the distribution characteristics of the predicted results, providing quantitative support for uncertainty in decision-making. The innovation of this assessment system lies in the introduction of a comprehensive risk assessment mechanism across multiple timescales, which balances short-term disaster prevention needs with long-term planning and management.

[0137] To better understand and implement the present invention, Example 2, a specific application scenario, is provided below. The research team selected a coral reef island in a certain sea area as the research object and conducted an experimental study using a computer vision-based method for predicting island shoreline erosion risk. This coral reef island covers an area of ​​approximately 2.7 square kilometers and has a coastline length of approximately 8.3 kilometers. Its geological structure is primarily coral reef limestone, with some sandy beaches. In recent years, due to climate change and human activities, shoreline erosion has become increasingly severe.

[0138] First, the team acquired high-resolution satellite remote sensing images of the island from 2015 to 2023, with a resolution better than 2 meters and an acquisition interval of 6 months. The wavelet transform denoising algorithm was used to eliminate image noise, increasing the signal-to-noise ratio from the original 16.5dB to 27.8dB. The histogram equalization method was then applied to enhance the image contrast. After enhancement, the boundary clarity between the island and the sea water increased by 42%. The multi-phase images were registered using a SIFT feature-based registration algorithm, with an average registration accuracy of 0.76 pixels. Finally, the OTSU segmentation algorithm was used to separate the island from the sea water, with a segmentation threshold of 0.47, and the Canny operator was used to extract the island contour. The high and low thresholds were set to 1.67 times and 0.83 times the image gradient mean, respectively.

[0139] An initial shoreline model was constructed using a small-grid approximation method. Considering the complexity of the shoreline and image resolution, an initial grid side length (first size value) of 12 meters was chosen. Sampling points were set at 6-meter intervals along the shoreline, resulting in a total of 1,384 sampling points. A square grid was generated around each sampling point, and 1,102 grids were selected for inclusion in the shoreline grid set.

[0140] like Figure 2 As shown in Figure 1, the small grids are merged to generate a multi-level grid representation. The attribute similarity threshold of adjacent grids is set to 0.87, and a bottom-up strategy is used to merge grids, ultimately forming a three-level grid structure, as shown in Table 1:

[0141] Table 1 Statistics of multi-level grids of island coastlines

[0142]

[0143] The team applied the above processing flow to the shoreline images at each time point, generating a shoreline feature matrix for that time point. They then calculated the change matrix between adjacent time points and applied a 5×5 spatial filter to remove noise, with a confidence threshold of 0.78. The research team analyzed island shoreline changes from 2015 to 2023 and identified three types of areas, as shown in Table 2:

[0144] Table 2. Classification statistics of coastline change regions

[0145]

[0146] A coastal erosion dynamics equation model was established, and environmental parameters of the study area were collected. The wave action intensity was calculated based on data from nearshore observation buoys, with an average value of 4.3 kW / m and a maximum value of 12.7 kW / m. The tidal amplitude was obtained from the Lingshui tide station data, with an average value of 1.2 meters. The sea level rise rate was determined to be 4.6 mm / year based on long-term monitoring data. The island geological structure coefficient was obtained through field surveys and the calculation formula is: , where the factor values ​​are shown in Table 3:

[0147] Table 3 Statistics of island geological structure parameters

[0148]

[0149] The shoreline slope angle was calculated based on the elevation data. The average slope of the eastern coral reef area is 12°, the average slope of the southern sandy beach is 6°, the average slope of the western mixed area is 15°, and the average slope of the northern steep cliff area is 36°.

[0150] Figure 3The relationship between the shoreline erosion rate and wave intensity in four different areas of the island (the eastern coral reef area, the southern sandy beach, the western mixed area and the northern cliff area) is shown. The chart contains the fitting curve and the actual measured data points (with error bars) for each area, which clearly shows the impact of different geological structures on the erosion rate. The chart shows that the erosion rate in the southern sandy beach area increases fastest with increasing wave intensity, while the erosion rate in the northern cliff area is lower due to the higher hardness of the rock. The critical erosion threshold line (1.5 meters / year) is marked in the figure, and the points of change in physical mechanisms such as the increase point of wave breaking energy and the critical point of sediment transport are marked. The graph attention grid evolution model is applied to analyze the law of shoreline changes. The shoreline grid is converted into a graph structure, and a network model consisting of 4 layers of graph convolution layers and 6-head attention is constructed. Calculation of grid area weight coefficient Take it as 144 square meters, Take it as 2304 square meters, Take it as 0.25, Take it as 0.75. In the calculation of mesh density parameter weight, Take it as 420 / km2, Take it as 215. In the calculation of historical change rate weight, Take it as 0.45, Take it as 0.12.

[0151] The model was trained using the collected data using mini-batch stochastic gradient descent with a batch size of 48, an initial learning rate of 0.001, and a cosine annealing strategy. The model was trained for 150 epochs, achieving optimal performance at epoch 132, with an average prediction error of 0.076 m / year on the validation set. The model training results are shown in Table 4.

[0152] Table 4 Model training performance statistics

[0153]

[0154] The uncertainty analysis module is introduced to generate multi-scale erosion risk assessment maps. Using the Monte Carlo simulation method, 2500 sets of random perturbations are performed on key parameters and 95% confidence intervals are calculated. The value is 1.96. The risk score uses three time scales: short-term, medium-term, and long-term, with weights of 0.5, 0.3, and 0.2 respectively. Based on the assessment results, the island coastline is divided into five risk levels, as shown in Table 5:

[0155] Table 5 Distribution of shoreline erosion risk levels

[0156]

[0157] Traditional island shoreline erosion risk assessment relies primarily on historical satellite imagery comparisons and simplified physical models. These methods suffer from low shoreline extraction accuracy, coarse spatial resolution, and significant prediction uncertainty. Traditional methods typically rely on manual interpretation of island boundaries, resulting in errors exceeding 10 meters. Furthermore, due to a lack of representation methods adapted to complex shoreline morphologies, they are unable to accurately depict detailed shoreline changes. Furthermore, traditional physical models typically only consider a single or limited number of influencing factors, ignoring the effects of multiple factors. This leads to significant deviations between predictions and actual conditions, particularly in terms of their ability to predict shoreline changes under extreme events such as storm surges.

[0158] In contrast, the present invention's computer vision-based island shoreline erosion risk prediction method achieves technological advancements in the following areas: First, by employing a small-grid approximation and multi-level merging strategy, shoreline representation accuracy is improved by 3.7 times, adapting to a wide range of shoreline morphologies, from flat coasts to complex coral reefs. Second, by constructing a coastal erosion dynamics equation that considers multiple factors, such as waves, tides, and sea level rise, the model's prediction accuracy is improved by 42%. Third, by introducing a graph-attention grid evolution model, it achieves differentiated processing of large and small grids, capturing global trends while preserving local details, and improving the spatial accuracy of prediction results by 2.8 times. Fourth, through uncertainty analysis and multi-scale risk assessment, it provides a more comprehensive and reliable decision-making basis for risk management, increasing prediction confidence by 36%. Experimental results demonstrate that this method is particularly suitable for small islands with complex shoreline morphologies, providing strong technical support for island protection and sustainable development.

[0159] It should be noted that the variables involved in the present invention are explained in detail as shown in Tables 6 and 7 below.

[0160] Table 6 Variable Explanation Table (Part I)

[0161]

[0162] Table 7 Variable Explanation Table (Part II)

[0163]

[0164] The above description is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be covered by the scope of protection of the present invention.

Claims

1. A computer vision-based island shoreline erosion risk prediction method, characterized in that: The process includes acquiring high-resolution island remote sensing images and extracting island contour information; constructing an initial island coastline model based on a small grid approximation method; merging small grids to generate a multi-level grid representation; acquiring multiple time series remote sensing images and calculating changes in the coastline feature matrix; A coastal erosion dynamics equation model is established, and the erosion rate at different grid positions is quantitatively calculated by combining wave intensity, tidal amplitude, sea level rise rate, island geological structure coefficient and shoreline slope angle; the atlas attention grid evolution model is applied to analyze the law of shoreline change. The atlas attention grid evolution model adjusts the attention allocation mechanism through the grid area weight coefficient to balance global changes and local details; an uncertainty analysis module is introduced to calculate the confidence interval of the prediction results, generate a multi-scale erosion risk assessment map, identify high-risk areas and output the prediction results; wherein, the structure of the atlas attention grid evolution model is a fusion architecture of a multi-layer graph convolutional neural network and a multi-head attention mechanism, representing the shoreline grid as graph structure data, and the connection relationship between grids corresponds to the edge; the atlas attention grid evolution model includes an encoding layer, a multi-layer graph convolution layer, a multi-head attention layer and a decoding layer; the core of the atlas attention grid evolution model is to introduce a grid area adaptive attention allocation mechanism, which dynamically adjusts the attention distribution through three key parameters in the shoreline feature matrix: grid density parameter, grid merging level and historical evolution rate of the change matrix, to effectively capture the shoreline change characteristics of different scales.

2. The island shoreline erosion risk prediction method based on computer vision according to claim 1, characterized in that: Acquiring high-resolution island remote sensing images and extracting island contour information includes image denoising, enhancement, registration and segmentation, and using machine vision algorithms to extract island contour information.

3. The island shoreline erosion risk prediction method based on computer vision according to claim 2, characterized in that: Constructing an initial model of the island coastline based on the small grid approximation method includes dividing the island boundary into a plurality of square grids with a side length of a first size value to form an island coastline grid set.

4. The island shoreline erosion risk prediction method based on computer vision according to claim 3, characterized in that: The small grid approximation method represents the island boundary as a set of square grids with equal side lengths. By adjusting the grid density, the complex coastline morphology can be accurately expressed. Each grid records the existence status and attribute information of the coastline in the area.

5. The island shoreline erosion risk prediction method based on computer vision according to claim 4, characterized in that: The first size value is the initial grid side length determined based on the complexity of the island coastline and the resolution of the remote sensing image, and the value range is 5 meters to 50 meters.

6. The island shoreline erosion risk prediction method based on computer vision according to claim 5, characterized in that: The merging operation of small grids to generate a multi-level grid representation includes finding the global largest square grid and storing all grid information as a shoreline feature matrix.

7. The island shoreline erosion risk prediction method based on computer vision according to claim 6, characterized in that: The shoreline feature matrix is ​​a two-dimensional data structure whose element values ​​represent the attributes of the corresponding position grid, including whether it is a shoreline point, shoreline type, elevation value and stability index. The matrix dimension is determined by the grid division accuracy.

8. The island shoreline erosion risk prediction method based on computer vision according to claim 7, characterized in that: The coastal erosion dynamics equation is used to simulate the physical mechanism of coastal erosion. Based on the principle of conservation of mass and fluid dynamics theory, it considers the balance between the impact force of waves on the shoreline and the resistance of shoreline materials, and obtains the erosion rate distribution of grids at different locations through integral calculation.

Images