Island shoreline identification method and system based on artificial intelligence

Abstract

The invention discloses an island shoreline identification method and system based on artificial intelligence, and relates to the technical field of shoreline change monitoring. According to the invention, through fusion modeling of the optical image, the SAR image and the topographic data, the limitation of a single data source is broken through; a vegetation coverage area and a non-vegetation shoreline can be accurately distinguished through the multispectral characteristics of the optical image, and the problem of monitoring blind areas in scenes such as cloud and mist and nighttime is solved through the all-weather penetrating power of the SAR image; a random forest classifier can effectively distinguish bedrock steep cliff, sandy sand beach and the like in combination with superpixel unit feature vectors constructed by topographic data, and accurate partitioning of a target island shoreline is completed; moreover, a dynamic global correction mechanism further eliminates space-time error accumulation, a standard deviation mean value is calculated through cross-season historical data, accidental errors such as typhoon and rainstorm are filtered, a future global correction coefficient is predicted in combination with a time sequence model, and the reliability of long-term monitoring is remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of coastline change monitoring, and specifically relates to an island coastline recognition method and system based on artificial intelligence. Background Art

[0002] Existing island coastline identification technologies are limited by a single data source and a fixed correction model, making it difficult to meet the needs of high-precision monitoring in complex geographical environments. Traditional methods rely heavily on optical remote sensing imagery to extract coastlines, but optical data is susceptible to interference from weather conditions such as clouds and fog, day and night, leading to misjudgment of water boundaries (e.g., the inability to identify intertidal zones during high tides). At the same time, spectral feature classification alone cannot effectively distinguish between coastline types with significant topographic differences. Furthermore, existing correction models generally use fixed global correction coefficients, which are unable to capture long-term coastline evolution trends (e.g., the average annual erosion rate caused by sea level rise) and high-frequency dynamic factors such as tides.

[0003] This application provides an island coastline identification method and system based on artificial intelligence to solve the above technical problems. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an island coastline identification method and system based on artificial intelligence, which breaks through the limitations of a single data source through the fusion modeling of optical images, SAR images and terrain data; the multispectral characteristics of optical images can accurately distinguish between vegetation-covered areas and non-vegetated coastlines, while the all-weather penetration capability of SAR images solves the problem of monitoring blind spots in scenes such as clouds and fog and at night; the random forest classifier can effectively distinguish between bedrock cliffs and sandy beaches, etc., by combining the superpixel unit feature vector constructed with terrain data, and complete the accurate zoning of the target island coastline. Moreover, the dynamic global correction mechanism further eliminates the accumulation of spatiotemporal errors, calculates the standard deviation mean through cross-seasonal historical data, filters out accidental errors such as typhoons and rainstorms, retains trend deviations such as sea level rise, and combines the time series model to predict the future global correction coefficient, significantly improving the reliability of long-term monitoring.

[0005] To achieve the above objectives, the first aspect of the present invention provides an island coastline recognition method based on artificial intelligence, comprising:

[0006] S100: Calculating a global correction coefficient for the target island; wherein the global correction coefficient is the mean of the standard deviations of the remote sensing distance values and the measured distance values of all shoreline points of the target island;

[0007] S200: partitioning the coastline of the target island using multi-source data to obtain a plurality of partitioned coastlines, wherein the multi-source data includes image data and terrain data, and the coastline types of the partitioned coastlines include coral reef areas, mangrove areas, bedrock cliff areas, sandy beach areas, and artificial coastline areas;

[0008] S300: Matching correction factors for a plurality of zoned shorelines; calculating a zoned correction factor for each zoned shoreline based on the correction factor and a global correction coefficient; wherein the correction factor is obtained based on actual measurement;

[0009] S400: Acquire a remote sensing image to be identified, extract a distance sequence to be corrected from the remote sensing image to be identified; correct the distance sequence to be corrected using a partition correction factor and a global correction coefficient to generate an actual coastline profile of the target island.

[0010] Preferably, the step of calculating the global correction coefficient includes:

[0011] S110: Collect historical remote sensing images and surface measured data of the target island; the surface measured data is the distance between the shoreline points and fixed features obtained by field measurement of typical shoreline points, and the field measurement data is matched with the time of the historical remote sensing images.

[0012] S120: uniformly select a number of shoreline points from the historical remote sensing image, calculate the remote sensing distance value between each shoreline point and its nearest fixed feature; integrate the remote sensing distance values of all shoreline points into a remote sensing distance sequence;

[0013] S130: Obtaining measured distance values between fixed features and corresponding shoreline points through surface measured data; integrating the measured distance values into a measured distance sequence;

[0014] S140: Calculate the standard deviation of the remote sensing distance value and the measured distance value corresponding to each shoreline point based on the remote sensing distance sequence and the measured distance sequence; the mean of the standard deviations of all shoreline points is used as the global correction coefficient.

[0015] Preferably, the step of calculating the global correction coefficient includes:

[0016] S110: Collect historical remote sensing images and surface measurement data of the target island for several years; the surface measurement data is the distance between the shoreline points and fixed features obtained by field measurement of typical shoreline points every quarter.

[0017] S120: uniformly select a number of shoreline points from the historical remote sensing images of each season, calculate the remote sensing distance value between each shoreline point and its nearest fixed feature; integrate the remote sensing distance values of all shoreline points at the seasonal scale into a seasonal remote sensing distance sequence;

[0018] S130: Obtaining measured distance values between fixed features and corresponding shoreline points through surface measured data; integrating the measured distance values into a seasonally measured distance sequence according to a seasonal scale;

[0019] S140: Based on the seasonal scale remote sensing distance series and the seasonal scale measured distance series, the standard deviation of the remote sensing distance value and the measured distance value corresponding to each shoreline point is calculated across the quarters; the mean of the standard deviations of all shoreline points in the same quarter is taken as the global correction coefficient;

[0020] S150: All global correction coefficients are spliced and integrated into a global correction sequence in quarterly order; and the global correction coefficient of the target island during the set period is predicted based on the global correction sequence and the time series model.

[0021] Preferably, the standard deviation of the remote sensing distance value and the measured distance value corresponding to each shoreline point is calculated across seasons, and the calculation formula is: Where n is the number of historical quarters; d i,k is the shoreline point a in the kth quarter i,k With target object b i Remote sensing distance value, d′ i,q is the shoreline point a′ in the current quarter q i,q The measured distance value.

[0022] Preferably, the coastline of the target island is partitioned using multi-source data to obtain several partitioned coastlines, including:

[0023] S210: collecting image data of the target island coastline, and dividing the image data into a plurality of superpixel units; wherein each superpixel unit is associated with an ecological feature;

[0024] S220: Acquire terrain data of the target island, resample the terrain data based on the resolution of the image data, and obtain terrain raster data; wherein the terrain data includes slope, roughness, altitude, and intertidal zone level;

[0025] S230: constructing a unit feature vector for each superpixel unit based on the terrain data; wherein the unit feature vector includes ecological features, terrain features, and spatial coordinates;

[0026] S240: Input the unit feature vector into the trained classifier to divide the target island coastline into zones; wherein the zone types include coral reef area, mangrove area, bedrock cliff area, sandy beach area and artificial coastline area, and the classifier includes random forest or support vector machine.

[0027] Preferably, the correction formula for the distance sequence to be corrected is: corrected =d raw ×(1-S), S=S global ×α; where dcorrected is the distance between the corrected shoreline point and the fixed feature, d raw is the distance to be corrected, α is the corresponding partition correction factor, S global is the global correction factor.

[0028] Preferably, when correcting the distance sequence to be corrected, the tidal dynamic factor is introduced, and the partition correction coefficient calculation formula is: S i (t) = S global (t)×α j ×δ(t), δ(t)=1+q×(tide level(t)-average tide level), q is set according to the partition type; α j is the correction factor for the corresponding partition type, j is the partition type number; t is used to represent different time points, S global (t) refers to the global correction coefficient for the quarter corresponding to time t.

[0029] Preferably, when correcting the distance sequence to be corrected, a mixed correction factor is introduced, and the calculation formula of the partition correction coefficient is: S i (t) = S global (t)×[α mix (x) × γ(θ)] × δ(t);

[0030] Where, α mix (x) is the partition mixing correction factor, α mix (x) = α A ×ω A (x)+α B ×ω B (x), A and B are target points a i The two adjacent partitions at the location, α A and α B are the correction factors for partitions A and B respectively; ω B (x)=1-ω A (x), x is the target point a i The distance to the boundary of partition A, 0≤x≤D, where D is the total width of the overlap band;

[0031] γ(θ) is the directional sensitivity factor, γ(θ) = 1 + l × cosθ; θ is the angle between the shoreline normal vector and the north direction; l is the directional sensitivity coefficient, l∈[-0.1, 0.1].

[0032] Preferably, when correcting the distance sequence to be corrected, an overlapping band weight function is introduced, and the calculation formula of the partition correction coefficient is: S i (t) = S global (t)×[α mix(x)×γ(θ)]×δ(t)×ω(x); where ω(x) is the overlapping band weight function, y is the target point a i The distance along the shoreline normal direction.

[0033] A second aspect of the present invention provides an island shoreline identification system based on artificial intelligence, comprising a shoreline identification module and a data acquisition module in communication with the shoreline identification module;

[0034] Data acquisition module: used to extract the global correction coefficient of the target island from the database; the global correction coefficient is the mean of the standard deviation of the remote sensing distance value and the measured distance value of all shoreline points of the target island; and

[0035] Used to collect multi-source data through drones or remote sensing satellites; the multi-source data includes image data and terrain data;

[0036] Coastline identification module: used to partition the coastline of the target island using multi-source data to obtain a number of partitioned coastlines; wherein the coastline types of the partitioned coastlines include coral reef areas, mangrove areas, bedrock cliff areas, sandy beach areas and artificial coastline areas; and,

[0037] Used to match correction factors for several partitioned coastlines; calculate the partitioned correction factors of each partitioned coastline based on the correction factors and the global correction coefficient; obtain the remote sensing image to be identified, and extract the distance sequence to be corrected from the remote sensing image to be identified; use the partitioned correction factors and the global correction coefficient to correct the distance sequence to generate the actual coastline contour of the target island.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. This invention overcomes the limitations of a single data source by integrating optical imagery, SAR imagery, and terrain data. The multispectral characteristics of optical imagery can accurately distinguish between vegetated areas and non-vegetated coastlines, while the all-weather penetration of SAR imagery solves the problem of monitoring blind spots in scenes such as fog and at night. Combining the superpixel unit feature vectors constructed with terrain data, a random forest classifier can effectively distinguish between bedrock cliffs and sandy beaches, accurately zoning the target island coastline. Furthermore, a dynamic global correction mechanism further eliminates the accumulation of spatiotemporal errors. By calculating the mean standard deviation of cross-seasonal historical data, it filters out accidental errors such as typhoons and heavy rains, retains trend deviations such as sea level rise, and combines time series models to predict future global correction coefficients, significantly improving the reliability of long-term monitoring.

[0040] 2. To address the high-frequency shoreline displacement caused by tides, the present invention introduces a tidal dynamic factor with differentiated partitions. The tidal impact is quantified through the q-value parameter, and the difference between the real-time tidal data and the average tidal level is converted into a correction coefficient adjustment amplitude. Moreover, through the application of a partitioned overlapping weighted mixer and a Gaussian kernel function, the recognition mutation of the hard boundary of the partition is eliminated. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 Schematic diagram of the method steps of the island coastline identification method of the present invention;

[0043] Figure 2 This is a schematic diagram of the method for obtaining the global correction coefficient in the present invention. Figure 1 ;

[0044] Figure 3 This is a schematic diagram of the method for obtaining the global correction coefficient in the present invention. Figure 2 ;

[0045] Figure 4 This is a schematic diagram of the system principle of the island coastline identification system in the present invention. DETAILED DESCRIPTION

[0046] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] Example 1:

[0048] See also Figure 1 The first embodiment of the present invention provides an island coastline recognition method based on artificial intelligence, comprising:

[0049] S100: Calculating a global correction coefficient for the target island; wherein the global correction coefficient is the mean of the standard deviations of the remote sensing distance values and the measured distance values of all shoreline points of the target island;

[0050] S200: partitioning the coastline of the target island using multi-source data to obtain a plurality of partitioned coastlines, wherein the multi-source data includes image data and terrain data, and the coastline types of the partitioned coastlines include coral reef areas, mangrove areas, bedrock cliff areas, sandy beach areas, and artificial coastline areas;

[0051] S300: Matching correction factors for a plurality of zoned shorelines; calculating a zoned correction factor for each zoned shoreline based on the correction factor and a global correction coefficient; wherein the correction factor is obtained based on actual measurement;

[0052] S400: Acquire a remote sensing image to be identified, extract a distance sequence to be corrected from the remote sensing image to be identified; correct the distance sequence to be corrected using a partition correction factor and a global correction coefficient to generate an actual coastline profile of the target island.

[0053] See also Figure 2 , in S100, the steps for calculating the global correction coefficient are as follows:

[0054] S110: Collect historical remote sensing images and surface measured data of the target island; the surface measured data is the distance between the shoreline points and fixed features obtained by field measurement of typical shoreline points. The field measurement data should match the time of the historical remote sensing images.

[0055] S120: uniformly select a number of shoreline points from the historical remote sensing image, calculate the remote sensing distance value between each shoreline point and its nearest fixed feature; integrate the remote sensing distance values of all shoreline points into a remote sensing distance sequence;

[0056] S130: Obtaining measured distance values between fixed features and corresponding shoreline points through surface measured data; integrating the measured distance values into a measured distance sequence;

[0057] S140: Calculate the standard deviation of the remote sensing distance value and the measured distance value corresponding to each shoreline point based on the remote sensing distance sequence and the measured distance sequence; the mean of the standard deviations of all shoreline points is used as the global correction coefficient.

[0058] For example, historical remote sensing imagery includes at least one optical image and one SAR image. Optical imagery, such as satellite optical imagery (e.g., Sentinel-2), is used to extract spectral characteristics of vegetation and ground objects; SAR imagery, acquired through SAR radar, is used for all-weather terrain contour extraction and water boundary identification. Before analyzing historical remote sensing imagery, radiometric calibration, atmospheric correction, and geometric correction are required.

[0059] Fixed features must be spatially stable and recognizable. Fixed bedrock points on the target island can be selected, as can fixed buildings on the target island. However, the position of fixed features must at least not change by more than 0.3 meters between consecutive quarters.

[0060] Select m coastline points uniformly (e.g., equally spaced grid sampling or stratified random sampling) from historical remote sensing images. These coastline points should cover a variety of categories, such as coral reefs, bedrock, artificial coastlines, etc. Mark the selected m coastline points as a i , i∈[1,m]. Before calculating the remote sensing distance value corresponding to the coastline point, the fixed object closest to the coastline point is taken as the target object and marked as b i ; then the shoreline point a i The remote sensing distance value is d i =|a i -b i |, integrate the remote sensing distance values of each coastline point into the remote sensing distance sequence D q ={d1,d2,…,d m}.

[0061] Calculate the shoreline point a′ based on the measured surface data i With target object b i The measured distance between the two is d′ i =|a′ i -b i |, the measured distance sequence is D′ q ={d′1,d′2,…,d′ m}.

[0062] It should be noted that each shoreline point a i Corresponding to the only target object b i , establish an association table to store the coordinates and type information of the two. i It is a shoreline point selected from the remote sensing image, and is the position determined after analyzing and processing the remote sensing image. i It is a shoreline point determined during field measurement. i and shoreline point a′ i Represents the same shoreline point, but the acquisition method is different. For example, shoreline point a i and shoreline point a′ i If they are in different seasons, the spatial position error of the shoreline points across seasons can be reduced through image registration technology and homonymous point matching technology. For example, a registration algorithm based on feature points (such as SIFT+RANSAC) can be used to align images of different seasons to a unified geographic coordinate system. The geographic coordinates of each shoreline point are recorded when it is first marked, and the pixel values of the corresponding coordinates are directly extracted through the registered images in subsequent seasons to ensure that the position of the shoreline point remains unchanged.

[0063] In S200, the steps of partitioning the coastline of the target island to obtain a plurality of partitioned coastlines are as follows:

[0064] S210: collecting image data of the target island coastline and dividing the image data into a plurality of superpixel units; wherein each superpixel unit is associated with ecological characteristics such as an average spectral value and an NDVI value;

[0065] S220: Acquire terrain data of the target island, resample the terrain data based on the resolution of the image data, and obtain terrain raster data; wherein the terrain data includes slope, roughness, altitude, and intertidal zone level;

[0066] S230: constructing a unit feature vector for each superpixel unit based on the terrain data; wherein the unit feature vector includes ecological features, terrain features, and spatial coordinates;

[0067] S240: Input the unit feature vector into the trained classifier to divide the target island coastline into zones; wherein the zone types include coral reef area, mangrove area, bedrock cliff area, sandy beach area and artificial coastline area, and the classifier includes random forest or support vector machine.

[0068] In S210, the image data is primarily used to extract ecological characteristics such as vegetation types and spectral characteristics of the target island coastline. Therefore, the image data includes multispectral and multispectral images. Dividing the image data into several superpixel units can be achieved using a simple linear iterative algorithm, with the following parameters set: segmentation scale: 10-20 pixels (corresponding to 0.5-1 meter in real-world terms) and compactness: 0.1-0.5 (the smaller the value, the closer the segmentation unit fits the boundary of the feature).

[0069] In S220, terrain data is primarily used to obtain slope, roughness, elevation, and intertidal zone levels for each superpixel unit. Therefore, terrain data can be obtained through LiDAR point cloud modeling acquired by airborne LiDAR. During resampling, bilinear interpolation is used for continuous factors such as slope and roughness, and nearest neighbor interpolation is used for coastal zone types.

[0070] The classifier includes one of the following: random forest, support vector machine, gradient boosting tree, and convolutional neural network. Taking random forest as an example, its training process is as follows:

[0071] 1. Data collection and processing:

[0072] 1) Through manual interpretation or field surveys, annotate 500-2000 superpixel samples, covering five sub-regions, such as coral reefs and mangroves. Each sample includes a feature vector (average spectrum, NDVI, slope, altitude, etc.) and a corresponding label (e.g., "coral reef area").

[0073] 2) Feature extraction: Normalization: Z-score normalization is performed on continuous features (such as slope and elevation); One-hot encoding: One-hot processing is performed on discrete features (such as intertidal zone level);

[0074] 2. Model training:

[0075] The samples were divided into training set and validation set in an 8:2 ratio. A random forest was selected, and the number of trees and the maximum tree depth were preset. The training set feature vector was input into the classifier, and the mapping relationship between sample features and partition types was iteratively learned.

[0076] 3. Model evaluation and optimization:

[0077] 1) Calculate overall accuracy (OA) and category accuracy (e.g., accuracy for coral reef areas and mangrove areas) using the validation set, with a target accuracy of >90%. Output a confusion matrix to analyze misclassification cases (e.g., why a bedrock cliff was misclassified as a sandy beach).

[0078] 2) Grid search was used to adjust key parameters, testing different numbers of trees and maximum features in the random forest. The model was retrained using the optimized parameters and saved for subsequent shoreline zoning and classification.

[0079] In S300, the correction factors for each zone type corresponding to the target island coastline are different due to their own specific differences. The correction factor values can refer to the estimated range of relevant research and actual monitoring experience, or can be determined through field measurement and data analysis. In this embodiment, the correction factors for each zone type are referred to the following table:

[0080] Partition Type Correction factor estimation range coral reef area 1.5-2.0 Mangrove area 0.6-0.9 Bedrock cliff area 0.3-0.6 Sandy beach area 1.0-1.2 Artificial shoreline area 0.1-0.3

[0081] In step S400, a remote sensing image to be identified is obtained, and the remote sensing distance of the target island coastline is identified using the remote sensing image. This remote sensing distance is used as the distance to be corrected. The distance to be corrected is then corrected by region based on the correction factor and the global correction coefficient, and the corrected coastline is used as the actual coastline of the target island.

[0082] In the actual calculation process, the correction factor and the global correction coefficient are first multiplied to obtain the partition correction coefficient; then the distance to be corrected of each partition coastline is combined with the corresponding partition correction factor to achieve the correction of the distance to be corrected.

[0083] For example, the distance to be corrected is marked as d raw , the corresponding partition correction factor is α, and the global correction coefficient is S global , the correction formula is d corrected =d raw ×(1-S), S=Sglobal ×α;d corrected is the distance between the corrected shoreline point and the fixed feature, and S is the partition correction coefficient.

[0084] This embodiment achieves multi-dimensional feature modeling of the coastline through multi-source data fusion and fixed global correction; optical images extract vegetation and spectral characteristics, SAR images break through weather restrictions to identify water boundaries, and terrain data (slope, intertidal zone) assists in distinguishing land feature types; statistical methods are combined to calculate the global correction coefficient and filter single sample noise; the global correction coefficient is combined with the correction factor of the partitioned coastline, fully considering the diversity of island coastlines, improving the accuracy of coastline correction, and solving the limitations of a single data source.

[0085] Example 2: Compared with the fixed global correction coefficient in Example 1, this example dynamically obtains the global correction coefficient, that is, calculates the global correction sequence based on the historical data of the target island, inputs the global correction sequence into the time series model, and predicts the global correction coefficient of the target island in the future period. Figure 3 .

[0086] S110: Collect historical remote sensing images and surface measurement data of the target island for several years; the surface measurement data is the distance between the shoreline points and fixed features obtained by field measurement of typical shoreline points every quarter.

[0087] S120: uniformly select a number of shoreline points from the historical remote sensing images of each season, calculate the remote sensing distance value between each shoreline point and its nearest fixed feature; integrate the remote sensing distance values of all shoreline points at the seasonal scale into a seasonal remote sensing distance sequence;

[0088] S130: Obtaining measured distance values between fixed features and corresponding shoreline points through surface measured data; integrating the measured distance values into a seasonally measured distance sequence according to a seasonal scale;

[0089] S140: Based on the seasonal scale remote sensing distance series and the seasonal scale measured distance series, the standard deviation of the remote sensing distance value and the measured distance value corresponding to each shoreline point is calculated across the quarters; the mean of the standard deviations of all shoreline points in the same quarter is taken as the global correction coefficient;

[0090] S150: All global correction coefficients are spliced and integrated into a global correction sequence in quarterly order; and the global correction coefficient of the target island during the set period is predicted based on the global correction sequence and the time series model.

[0091] In S110, the time range of the historical remote sensing images and surface measured data corresponding to the target island can be the past 3-5 years. Generally speaking, a larger time range helps improve the prediction accuracy of the global correction coefficient.

[0092] Each quarter's historical remote sensing imagery should include at least one optical image and one SAR image. The acquisition time of the historical remote sensing images and the surface measured data should be kept consistent to minimize errors in distance calculations. Before analyzing historical remote sensing images, they must undergo radiometric calibration, atmospheric correction, and geometric correction.

[0093] Fixed features can be selected from fixed bedrock points on the target island, or fixed buildings on the target island. However, the position change of fixed features between consecutive seasons should be less than 0.3 meters.

[0094] In S120, m coastline points are uniformly selected from the remote sensing images of each season. These coastline points should cover a variety of categories, such as coral reefs, bedrock, artificial coastlines, etc. The selected m coastline points are marked as a i,q , i∈[1,m], q is the quarter number, q∈[1,Q].

[0095] Before calculating the remote sensing distance value corresponding to the shoreline point, the fixed object closest to the shoreline point is taken as the target object and marked as b. i ; then the shoreline point a i,q The remote sensing distance value is d i,q =|a i,q -b i |, integrate the remote sensing distance values of each coastline point into the seasonal scale remote sensing distance series D q ={d 1,q ,d 2,q ,…,d m,q}.

[0096] In S130, the shoreline point a′ is calculated based on the measured surface data. i,q With target object b i The measured distance between the two is d′ i,q =|a′ i,q -b i |, the measured distance sequence is D' q ={q′ 1,q ,d′ 2,q ,…,d′ m,q}.

[0097] In S140, the formula for calculating the standard deviation of the remote sensing distance value and the measured distance value corresponding to each shoreline point across the quarter is: Where n is the number of historical quarters. For example, taking the first three quarters (k = q-3, q-2, q-1), n = 3. The measured value of the current quarter (q) is used to calculate the deviation between each historical remote sensing value and the measured value. The total sample size is n = 3. i,k is the shoreline point a in the kth quarter i,k With target object bi Remote sensing distance value, d′ i,q is the shoreline point a′ in the current quarter (q) i,q The measured distance value.

[0098] It should be noted that in island coastline monitoring, calculating standard deviations across quarters is intended to extract long-term, stable error characteristics from the time series, filter out short-term random interference, and ensure that the global correction coefficient reflects the trend deviation of remote sensing imagery. Single-quarter standard deviations, on the other hand, are suitable for real-time emergency monitoring, where errors may include a large number of random factors.

[0099] Take the mean standard deviation of all shoreline points in the same quarter as the global correction coefficient S global,q , Then the global correction sequence is [S global,1 , S global,2 ,…,S global,Q ].

[0100] In S150, the global correction coefficient of the target island set period is predicted based on the global correction sequence and the time series model, including:

[0101] S151: performing data conversion on the global correction coefficient and time in the global correction sequence to ensure that they can be input into the time series model;

[0102] S152: Determine a set period for prediction, select several sets of training data from the global correction sequence according to the set period for prediction; and train a time series model using the several sets of training data;

[0103] S153: Determine model input data according to the set period of prediction, and input the model input data into the time series model to obtain a global correction coefficient for prediction.

[0104] In S152, if the time period is set to 3, the time series model needs to output the global correction coefficient for the next three quarters. Starting from the first data in the global correction sequence, samples are selected by sliding according to a length of 15 (or greater than 15) data to ensure that the input data covers a complete year. The sliding interval is 1, and several samples are obtained. The last three data in each sample are used as the model output, and the remaining data are used as the model input. The time series model is trained using these samples. The time series model can be a Prophet model or an LSTM model.

[0105] It should be noted that when identifying the shoreline of the target island using the remote sensing image to be identified, the acquisition time of the remote sensing image to be identified is determined, and the corresponding global correction coefficient is matched based on the acquisition time. At the same time, factors such as the tidal dynamic factor that are used for image shoreline identification are calculated based on the acquisition time.

[0106] This embodiment introduces dynamic global correction and time series prediction, constructs a correction sequence based on 3-5 years of historical data, and uses the Prophet / LSTM model to capture the long-term trend of errors; the dynamic correction mechanism adapts to the interannual changes in shoreline erosion / siltation, replacing manual periodic calibration, and significantly improving monitoring efficiency.

[0107] Example 3: Based on Example 1 or Example 2, this example introduces a tidal dynamic factor when performing zone correction on the target island coastline, and adjusts the correction formula to: S i (t) = S global (t)×α j ×δ(t), δ(t)=1+q×(tide level(t)-mean tide level), the value of q is related to the partition type; α j is the correction factor for the corresponding partition type, j is the partition type number; t is used to represent different time points, S global (t) refers to the global correction coefficient for the quarter corresponding to time t, S i (t) is the partition correction coefficient.

[0108] The tidal dynamic factor is introduced in this embodiment because tide is a high-frequency dynamic factor that affects the position of island coastlines. Tidal changes directly lead to periodic displacement of the coastline. For example, at high tide, seawater floods the intertidal zone, and the coastline in remote sensing images shifts landward. At low tide, the intertidal zone is exposed, and the coastline extends seaward. By introducing the tidal dynamic factor, shoreline identification errors caused by tidal level differences can be corrected in real time, in accordance with the principles of coastal zone dynamics. The average tide level is the arithmetic mean of the tide levels over a certain period of time (such as a day, month, or year), reflecting the average state of the tide.

[0109] The value of q can be determined according to the partition type. The average value can be used in the specific calculation. Please refer to the following table for details:

[0110] Partition Type q value range Value meaning coral reef area 0.015-0.025 For every 1 meter change in tide level, the correction factor is adjusted by 1.5%-2.5% Sandy beach area 0.010-0.020 For every 1 meter change in tide level, the correction factor is adjusted by 1.0%-2.0% Mangrove area -0.005--0.010 For every 1 meter change in tide level, the correction factor is adjusted by -0.5% to 1.0% Bedrock cliff area 0.001-0.005 For every 1 meter change in tide level, the correction factor is adjusted by 0.1%-0.5% Artificial shoreline area 0.000-0.001 The effect of tide level change on the correction coefficient is close to 0

[0111] This embodiment introduces a tidal dynamic factor and quantifies the influence of tide level through a zone differentiation parameter (q value); the correction coefficient is adjusted in real time in combination with tide level data, which effectively solves the problem of high-frequency shoreline displacement caused by tides. It is particularly suitable for semi-diurnal and diurnal tide sea areas, and significantly improves the recognition accuracy of sensitive areas such as intertidal wetlands.

[0112] Example 4: Based on Example 3, in order to ensure a smooth transition of the correction coefficient at the partition boundary and avoid sudden changes in shoreline identification caused by hard partition boundaries, a partition overlap band weighted mixing mechanism and directional sensitivity correction can be introduced.

[0113] Adjust the correction formula to: S i (t) = S global (t)×[αmix (x) × γ(θ)] × δ(t) × ω(x);

[0114] Where, α mix (x) is the partition mixing correction factor, α mix (x) = α A ×ω A (x)+α B ×ω B (x), A and B are target points a i The two adjacent partitions at the location, α A and α B are the correction factors for partitions A and B respectively; ω B (x)=1-ω A (x), x is the target point a i The distance to the boundary of partition A, and 0≤x≤D, D is the total width of the overlapping band, which can be 40 meters, that is, 20 meters on each side), and W is the width of the Gaussian kernel, which can be 10 meters, used to control the transition rate.

[0115] γ(θ) is the directional sensitivity factor, γ(θ) = 1 + l × cosθ; θ is the angle between the shoreline normal vector and the north direction; l is the directional sensitivity coefficient, l∈[-0.1, 0.1], which is determined by the zoning type. For example, the coral reef area has strong erosion toward the sea, and l = 0.1.

[0116] ω(x) is the overlapping band weight function, y is the target point a i Distance along the shoreline normal direction, W y =20 meters.

[0117] Exemplary:

[0118] Assumption: Zone A (coral reef area): α A =1.8, directional sensitivity coefficient l A =0.1 (strong seaward erosion); Zone B (sandy beach area): α B =1.1, directional sensitivity coefficient l B = 0.01 (weak seaward erosion);

[0119] The target point position is: vertical distance x = 10 meters (10 meters from the boundary of the coral reef area, in the overlapping zone), parallel distance y = 0 meters (in the direction of the shoreline normal vector), and the angle θ between the shoreline normal vector and due north is 0° (completely facing the sea).

[0120] Partition mixing correction factor: ω B (10) = 1 - 0.952 = 0.048; α mix(10) = 1.8 × 0.952 + 1.1 × 0.048 = 1.7136 + 0.0528 = 1.7664 ≈ 1.77.

[0121] Directional sensitivity factor: γ(θ)γ(0°)=1+l A × cos(0°) = 1 + 0.1 × 1 = 1.1;

[0122] Overlapping band weight function:

[0123] This embodiment uses a Gaussian kernel function to achieve a smooth transition of correction coefficients of adjacent partitions through weighted blending of partition overlaps and directional sensitivity correction; the correction intensity is adjusted in combination with the direction of the shoreline normal vector, thereby improving the spatial continuity and dynamic response accuracy of complex shorelines, making the correction results more in line with natural geographical characteristics.

[0124] See also Figure 4 , a second embodiment of the present invention provides an island coastline identification system based on artificial intelligence, including a coastline identification module and a data acquisition module in communication with the coastline identification module;

[0125] Data acquisition module: used to extract the global correction coefficient of the target island from the database; the global correction coefficient is the mean of the standard deviation of the remote sensing distance value and the measured distance value of all shoreline points of the target island; and

[0126] Used to collect multi-source data through drones or remote sensing satellites; the multi-source data includes image data and terrain data;

[0127] Coastline identification module: used to partition the coastline of the target island using multi-source data to obtain a number of partitioned coastlines; wherein the coastline types of the partitioned coastlines include coral reef areas, mangrove areas, bedrock cliff areas, sandy beach areas and artificial coastline areas; and,

[0128] Used to match correction factors for several partitioned coastlines; calculate the partitioned correction factors of each partitioned coastline based on the correction factors and the global correction coefficient; obtain the remote sensing image to be identified, and extract the distance sequence to be corrected from the remote sensing image to be identified; use the partitioned correction factors and the global correction coefficient to correct the distance sequence to generate the actual coastline contour of the target island.

[0129] The data acquisition module is connected to a database and can directly extract required data from it. All non-real-time data in this invention is stored in the database. The data acquisition module can also acquire real-time data via drones and remote sensing satellites. The shoreline identification module primarily stores various types of data sent by the data acquisition module and simultaneously trains the classifier and artificial intelligence model. The database, data acquisition module, and shoreline identification module collaborate to identify the shoreline of the target island.

[0130] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An island coastline recognition method based on artificial intelligence, characterized in that: include: Calculate the global correction coefficient of the target island; the global correction coefficient is the mean of the standard deviations of the remote sensing distance values and the measured distance values of all shoreline points of the target island; The coastline of the target island is partitioned using multi-source data to obtain a plurality of partitioned coastlines, wherein the multi-source data includes image data and terrain data, and the coastline types of the partitioned coastlines include coral reef areas, mangrove areas, bedrock cliff areas, sandy beach areas, and artificial coastline areas; Matching correction factors for several zoned shorelines; calculating the zoned correction factors for each zoned shoreline based on the correction factors and the global correction coefficient; wherein the correction factors are obtained through actual measurements; Acquire the remote sensing image to be identified, extract the distance sequence to be corrected from the remote sensing image to be identified; use the partition correction factor and the global correction coefficient to correct the distance sequence to generate the actual coastline contour of the target island.

2. The island coastline identification method based on artificial intelligence according to claim 1, characterized in that: The calculation steps of the global correction coefficient include: Collect historical remote sensing images and surface measurement data of the target island; the surface measurement data is the distance between the shoreline points and fixed features obtained by field measurement of typical shoreline points. The field measurement data is matched with the time of the historical remote sensing images; Evenly select a number of shoreline points from historical remote sensing images, calculate the remote sensing distance value between each shoreline point and its nearest fixed feature; integrate the remote sensing distance values of all shoreline points into a remote sensing distance sequence; Obtain the measured distance value between the fixed ground object and the corresponding shoreline point through the surface measured data; integrate the measured distance value into a measured distance sequence; The standard deviation of the remote sensing distance value and the measured distance value corresponding to each shoreline point is calculated based on the remote sensing distance sequence and the measured distance sequence; the mean of the standard deviations of all shoreline points is used as the global correction coefficient.

3. The island coastline identification method based on artificial intelligence according to claim 1, characterized in that: The calculation steps of the global correction coefficient include: Collect several years of historical remote sensing images and surface measurement data of the target island; the surface measurement data is the distance between the shoreline points and fixed features obtained by field measurement of typical shoreline points every quarter; A number of shoreline points are evenly selected from the historical remote sensing images of each season, and the remote sensing distance value between each shoreline point and its nearest fixed feature is calculated; the remote sensing distance values of all shoreline points at the seasonal scale are integrated into a seasonal scale remote sensing distance sequence; The measured distance values between fixed features and corresponding shoreline points are obtained through surface measured data; the measured distance values are integrated into a seasonal scale measured distance sequence according to the seasonal scale; Based on the seasonal scale remote sensing distance series and seasonal scale measured distance series, the standard deviation of the remote sensing distance value and the measured distance value corresponding to each shoreline point is calculated across seasons; the mean of the standard deviation of all shoreline points in the same season is taken as the global correction coefficient; All global correction coefficients are spliced and integrated into a global correction sequence in quarterly order; the global correction coefficient of the target island set period is predicted based on the global correction sequence and time series model.

4. The island coastline identification method based on artificial intelligence according to claim 3 is characterized in that: The standard deviation of the remote sensing distance value and the measured distance value corresponding to each shoreline point is calculated across quarters. The calculation formula is: Where n is the number of historical quarters; d i,k is the shoreline point a in the kth quarter i,k With target object b i Remote sensing distance value, d′ i,q is the shoreline point a′ in the current quarter i,q The measured distance value.

5. The island coastline identification method based on artificial intelligence according to claim 1 is characterized in that: The coastline of the target island is partitioned using multi-source data to obtain several partitioned coastlines, including: Collecting image data of the target island coastline and segmenting the image data into a plurality of superpixel units; wherein each superpixel unit is associated with an ecological feature; Obtain the topographic data of the target island and resample it based on the resolution of the image data to obtain topographic raster data; the topographic data includes slope, roughness, altitude, and intertidal zone level; Constructing a unit feature vector for each superpixel unit based on terrain data; wherein the unit feature vector includes ecological characteristics, terrain characteristics and spatial coordinates; The unit feature vector is input into a pre-trained classifier to partition the target island coastline; the partition types include coral reef area, mangrove area, bedrock cliff area, sandy beach area and artificial coastline area, and the classifier includes random forest or support vector machine.

6. The island coastline identification method based on artificial intelligence according to claim 2, characterized in that: The correction formula for the distance sequence to be corrected is: corrected =d raw ×(1-S), S=S global ×α; where d corrected is the distance between the shoreline point and the fixed feature after correction, d raw is the distance to be corrected, α is the corresponding partition correction factor, S global is the global correction factor.

7. The island coastline identification method based on artificial intelligence according to claim 4 is characterized in that: When correcting the distance sequence to be corrected, the tidal dynamic factor is introduced and the partition correction coefficient is calculated as follows: S i (t) = S global (t)×α j ×δ(t), δ(t)=1+q×(tide level(t)-average tide level), q is set according to the partition type; α j is the correction factor for the corresponding partition type, j is the partition type number; t is used to represent different time points, S global (t) refers to the global correction coefficient for the quarter corresponding to time t.

8. The island coastline identification method based on artificial intelligence according to claim 7 is characterized in that: When correcting the distance sequence to be corrected, a mixed correction factor is introduced, and the calculation formula of the partition correction coefficient is: S i (t) = S global (t)×[α mix (x) × γ(θ)] × δ(t); Where, α mix (x) is the partition mixing correction factor, α mix (x) = α A ×ω A (x)+α B ×ω B (x), A and B are target points a i The two adjacent partitions at the location, α A and α B are the correction factors for partitions A and B respectively; ω B (x)=1-ω A (x), x is the target point a i The distance to the boundary of partition A, 0≤x≤D, where D is the total width of the overlap band; γ(θ) is the directional sensitivity factor, γ(θ) = 1 + l × cosθ; θ is the angle between the shoreline normal vector and the north direction; l is the directional sensitivity coefficient, l∈[-0.1, 0.1].

9. The island coastline identification method based on artificial intelligence according to claim 8, characterized in that: When correcting the distance sequence to be corrected, the overlapping band weight function is introduced, and the calculation formula of the partition correction coefficient is: S i (t) = S global (t)×[α mix (x)×γ(θ)]×δ(t)×ω(x); where ω(x) is the overlapping band weight function, y is the target point a i The distance along the shoreline normal direction.

10. An island coastline identification system based on artificial intelligence, used to implement the island coastline identification method based on artificial intelligence according to any one of claims 1 to 9, characterized in that: It includes a shoreline identification module and a data acquisition module connected to the shoreline identification module; Data acquisition module: used to extract the global correction coefficient of the target island from the database; the global correction coefficient is the mean of the standard deviation of the remote sensing distance value and the measured distance value of all shoreline points of the target island; and Used to collect multi-source data through drones or remote sensing satellites; the multi-source data includes image data and terrain data; Coastline identification module: used to partition the coastline of the target island using multi-source data to obtain a number of partitioned coastlines; wherein the coastline types of the partitioned coastlines include coral reef areas, mangrove areas, bedrock cliff areas, sandy beach areas and artificial coastline areas; and, Used to match correction factors for several partitioned coastlines; calculate the partitioned correction factors of each partitioned coastline based on the correction factors and the global correction coefficient; obtain the remote sensing image to be identified, and extract the distance sequence to be corrected from the remote sensing image to be identified; use the partitioned correction factors and the global correction coefficient to correct the distance sequence to generate the actual coastline contour of the target island.

Images