A method and system for identifying and analyzing coastal erosion based on remote sensing shoreline classification

Abstract

The present application relates to the technical field of remote sensing monitoring of coastal geological environment, and particularly relates to a method and system for identifying and analyzing coastal erosion based on remote sensing shoreline classification. First, multiple remote sensing images of a target area taken by a remote sensing satellite are obtained and preprocessed as first target images. Then, a normalized water index is calculated, spectral, texture and geometric parameters are extracted, object patches are segmented based on the above features, and a coastal type distribution map is obtained using a classification algorithm. Next, a water edge line identification model based on a convolutional neural network is constructed and trained, the preprocessed first target images are input into the model, the instantaneous water edge line of the target area is obtained, and the instantaneous water edge line is corrected to obtain a shoreline image. The shoreline erosion rate is calculated based on multiple shoreline images, and the warning level is determined in combination with the corresponding grading threshold of the coastal type and sent to a management terminal for display. The present scheme can automatically and efficiently identify shoreline changes of different types of coast and determine erosion risks.

Description

Technical Field

[0001] This invention relates to the field of remote sensing monitoring technology for coastal geological environment, specifically to a method and system for identifying and analyzing coastal erosion based on remote sensing shoreline classification. Background Technology

[0002] The coastline is the boundary between land and sea, a crucial area with complex land-sea interactions and abundant resources. Under the combined influence of global and coastal regional environmental processes and human activities, the coastline undergoes drastic changes, and its impact on ecology, environment, and socio-economic development cannot be ignored. Therefore, research on coastline changes has received widespread attention.

[0003] In the fields of environmental monitoring and coastline management, coastal erosion is a common geological indicator. Coastal erosion has a profound impact on ecosystems, human settlements, and economic activities, especially on sandy coasts where changes in this geological element are particularly significant. Traditional coastal erosion monitoring methods rely on ground measurements and historical data analysis, but this approach is inefficient and lacks accuracy. Therefore, there is an urgent need for a more efficient and accurate coastal erosion monitoring technology. Summary of the Invention

[0004] The main objective of this invention is to provide a method and system for identifying and analyzing coastal erosion based on remote sensing shoreline classification, aiming to solve the current urgent need for a more efficient and accurate coastal erosion monitoring technology.

[0005] The technical solution proposed in this invention is as follows: A coastal erosion identification and analysis method based on remote sensing shoreline classification is applied to a coastal erosion identification and analysis system based on remote sensing shoreline classification; the system includes a cloud server; the method includes: The cloud server obtains multiple remote sensing images of the target area taken by the remote sensing satellite each time it revisits the target area through the network port, and uses them as the first target image; The cloud server preprocesses the acquired first target image; The cloud server builds and trains a waterline recognition model based on a convolutional neural network to obtain an optimal waterline recognition model; The cloud server inputs the preprocessed first target image into the optimized waterline recognition model to obtain the output waterline recognition result; The cloud server performs corrections based on the waterline recognition results to obtain a coastline image of the target area. The cloud server uses the coastline image to determine whether there is a risk of rapid coastline erosion in the target area.

[0006] Preferably, the cloud server preprocesses the acquired first target image, including: Radiometric calibration: The DN value of the first target image is converted into a physical radiance value to ensure an accurate correspondence between the image's radiometric measurements and the surface radiance value. Atmospheric correction processing; Image enhancement processing is performed on the first target image using histogram equalization. The first target image is denoised using a Gaussian filtering method.

[0007] Preferably, the cloud server constructs and trains a waterline recognition model based on a convolutional neural network to obtain a preferred waterline recognition model, including: The cloud server acquires remote sensing images for training, preprocesses the remote sensing images, and marks the preprocessed remote sensing images as training images. The cloud server performs pixel-level annotation on the training images to distinguish between land and water, and generates corresponding result labels. Water pixels in the training images are labeled as 0, and land pixels in the training images are labeled as 1. The cloud server performs block processing on the training images: the training images are cropped into several training sub-images; The cloud server divides all training sub-images and their corresponding result labels into training and validation sets according to their quantity ratio.

[0008] Preferably, the cloud server divides all training sub-images and corresponding result labels into training and validation sets proportionally, and then further includes: A cloud server constructs a waterline recognition model based on the U-Net architecture. The waterline recognition model structure consists of an encoder and a decoder. The encoder is used to: gradually extract high-level semantic features of the image through repeated convolution and pooling operations, while reducing the resolution of the feature map. The decoder is used to: gradually restore the resolution of the feature map through transposed convolution or upsampling operations. The cloud server uses the binary cross-entropy loss function as the loss function for the waterline recognition model. The cloud server uses intersection-union ratio, accuracy, and F1 score as evaluation metrics for the waterline recognition model. The cloud server inputs the training set into the waterline recognition model for training; During the training process of the waterline recognition model, the cloud server calculates the error of the waterline recognition model based on the loss function and updates the learnable weight parameters in the waterline recognition model through the backpropagation algorithm. The update process requires multiple iterations. The cloud server inputs the validation set into the trained waterline recognition model and determines whether the evaluation index of the trained waterline recognition model meets the first preset condition. The first preset condition is: the intersection-union ratio is greater than or equal to the first preset value, the accuracy is greater than or equal to the second preset value, and the F1 score is greater than or equal to the third preset value. If the first preset condition is met, the cloud server will use the trained waterline recognition model as the preferred waterline recognition model.

[0009] Preferably, the cloud server performs correction based on the waterline recognition results to obtain a coastline image of the target area, including: The cloud server converts the output waterline recognition result into a binary mask image based on a set segmentation threshold. The format of the waterline recognition result output by the waterline recognition model is a probability map. The cloud server performs opening and closing operations on the binary mask image sequentially to remove noise patches in the binary mask image and fill in tiny holes in land or water areas. The cloud server uses the Canny edge detection algorithm to extract the boundary lines of water and land from the binary mask image. The extracted boundary lines of water and land are the gridded water edge lines. The cloud server performs vectorization processing on the extracted raster boundary lines to obtain vector line files; The cloud server uses the Douglas-Peucker algorithm to smooth the waterline vector lines in the vector line file.

[0010] Preferably, the cloud server uses the Douglas-Peucker algorithm to smooth the waterline vector lines in the vector line file, and then further includes: The cloud server selects two first target images that are continuously acquired by remote sensing satellites in the same revisit of the target area. The interval between the satellite imaging times of the two first target images must be less than or equal to a first preset time. The cloud server uses a preferred waterline recognition model to obtain vector line files corresponding to two consecutive first target images. The vector line file whose waterline vector is closer to the land is marked as the first vector line file, and the other vector line file is marked as the second vector line file. The cloud server marks the first target image corresponding to the first vector line file as the first original image, and marks the first target image corresponding to the second vector line file as the second original image; The cloud server acquires the digital elevation model of the target area; The cloud server uses ArcGIS software based on the digital elevation model to set a first target point every preset length along the waterline vector line of the first vector line file, and generates a coastline passing through the target point. The vertical plane of the coastline is perpendicular to the tangent between the first target point and the waterline vector line. The cloud server uses ArcGIS software based on the digital elevation model to obtain the intersection point of the coastline corresponding to the first target point and the waterline vector line of the second vector line file, and marks it as the second target point. The cloud server obtains the instantaneous tide height value of the target area at the time of satellite imaging of the first original image, as well as the average high tide level, and the instantaneous tide height value of the target area at the time of satellite imaging of the second original image. The cloud server calculates the shore slope of the first target point based on the distance between the first and second target points, the instantaneous tide level of the target area at the time of satellite imaging of the first original image, and the instantaneous tide level of the target area at the time of satellite imaging of the second original image. ; The cloud server is based on the shore slope of the first target point. Calculate the calibration distance between the coastline and the waterline at the first target point: The cloud server uses a digital elevation model of the target area to translate and calibrate each first target point along the vertical direction of the coastline towards the land, and then obtains the intersection point with the land. These intersection points are marked as correction points. The correction points are then connected sequentially to form a coastline vector line. The Douglas-Puk algorithm is used to smooth the coastline vector line, and the smoothed coastline vector line is superimposed on the first original image to obtain the final coastline image.

[0011] Preferably, the cloud server determines whether there is a risk of rapid coastline erosion in the target area based on the coastline image, including: The cloud server establishes a correspondence between the obtained coastline image and the satellite imaging time of the first target image corresponding to the coastline image, wherein each revisit period corresponds to one coastline image. The cloud server selects a first preset number of coastline images from the past and marks them as second target images. The last second target image is the coastline image closest to the current time, and the first second target image is the coastline image farthest from the current time. The interval between the satellite imaging times of any two adjacent second target images is greater than the second preset time. The cloud server identifies fixed landmarks in the second target image, wherein the fixed landmarks are located in the land portion of the second target image, and ensures that the same fixed landmark is included in all second target images; The cloud server obtains the center pixel of the fixed marker in the second target image, and determines the first reference line and the second reference line based on the center pixel. The first reference line is the line segment connecting the start and end points of the coastline vector line in the second target image, and the second reference line is the line segment that passes through the center pixel and runs through the second target image, and the second reference line is parallel to the first reference line. The cloud server selects a second preset number of target reference points from the coastline vector lines of each second target image; The cloud server obtains the distance values ​​between each target reference point and the second reference line in each second target image. ,in, Let be the distance between the j-th target reference point of the i-th second target image and the second reference line, in meters, where 1 ≤ i ≤ I, I is the first preset quantity, 1 ≤ j ≤ J, and J is the second preset quantity; cloud servers are based on Calculate the coastline erosion rate and determine whether there is a risk of excessively rapid erosion in the target area based on the coastline erosion rate.

[0012] Preferably, the cloud server is based on The formula for calculating the rate of coastal erosion is as follows: , , In the formula, The coastline erosion rate is the ratio of the (i+1)th second target image to the ith second target image. The interval between the satellite imaging time corresponding to the (i+1)th second target image and the satellite imaging time corresponding to the ith second target image is expressed in hours. The method of determining whether a target area is at risk of excessively rapid erosion based on the coastline erosion rate includes: The cloud server determines whether the second preset condition is met: traversing i from 1 to I-1, the coastline erosion rate of all (i+1)th second target images is greater than the preset threshold compared to the i-th second target image; If the second preset condition is met, the cloud server is based on Further assessment is needed to determine if any localized interference issues exist; If there are no interference issues, the cloud server determines that the target area is at risk of being eroded too quickly and generates an alarm message to send to the management terminal.

[0013] The present invention also proposes a coastal erosion identification and analysis system based on remote sensing shoreline classification, which applies a coastal erosion identification and analysis method based on remote sensing shoreline classification; the system includes a cloud server.

[0014] The above technical solution can achieve the following beneficial effects: This invention proposes a coastal erosion identification and analysis method based on remote sensing shoreline classification, providing a more efficient and accurate technical solution for monitoring coastal erosion. First, multiple remote sensing images of the target area taken by remote sensing satellites are acquired and used as the first target image, which is then preprocessed. Next, a waterline identification model based on a convolutional neural network is constructed and trained to obtain an optimized waterline identification model. The preprocessed first target image is then input into the optimized waterline identification model to obtain the output waterline identification result, thus obtaining the instantaneous waterline of the target area. The instantaneous waterline is then corrected to obtain the coastline image of the target area, which is then sent to a management terminal for display. This solution, based on remote sensing images and a convolutional neural network waterline identification model, can automatically and efficiently obtain coastline images of the target area within different revisit periods, enabling managers to monitor and observe coastline changes in the target area in a timely and convenient manner. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0016] Figure 1 This is a flowchart illustrating the first embodiment of a coastal erosion identification and analysis method based on remote sensing shoreline classification proposed in this invention. Figure 2 This is a flowchart illustrating the steps for determining the coastline erosion early warning level of a target area in the first embodiment of a coastal erosion identification and analysis method based on remote sensing coastline classification proposed in this invention. Figure 3 This is an example image of the first target image in the first embodiment of the coastal erosion identification and analysis method based on remote sensing shoreline classification proposed in this invention. Detailed Implementation

[0017] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0018] This invention proposes a method and system for identifying and analyzing coastal erosion based on remote sensing shoreline classification.

[0019] As attached Figure 1 - Appendix Figure 3As shown, in the first embodiment of the coastal erosion identification and analysis method based on remote sensing shoreline classification proposed in this invention, this method is applied to a coastal erosion identification and analysis system based on remote sensing shoreline classification; the system includes a cloud server and a management terminal that are communicatively connected to each other; this embodiment includes the following steps: Step S110: The cloud server obtains multiple remote sensing images of the target area taken by remote sensing satellites (such as the domestic Gaofen-1 satellite, Gaofen-2 satellite, Sentinel-2 satellite, etc.) each time they revisit the target area through a network port (such as a geospatial data cloud), and uses them as the first target image.

[0020] Specifically, the requirements for the first target image are: cloud cover less than or equal to 5%; each time the remote sensing satellite revisits the target area, it will continuously capture multiple remote sensing images. In this embodiment, the Gaofen-1 satellite is used as the remote sensing satellite. Gaofen-1 is a sun-synchronous orbit satellite. With its ±25° side-swing capability, Gaofen-1 can achieve a rapid revisit every 4 days (i.e., a revisit cycle of 5 days). By forming a constellation of multiple satellites, this time can be further shortened to 1 day, or even every few hours in emergency situations; this means that a higher frequency of monitoring can be carried out on specific target areas.

[0021] Step S120: The cloud server preprocesses the acquired first target image.

[0022] Specifically, ENVI 5.3 software was used as the processing software to perform orthorectification, radiometric calibration, and atmospheric correction in sequence. ENVI (Environment for Visualizing Images) is an industrial-grade platform for remote sensing and geographic image analysis, proficient in processing multi / hyperspectral, LiDAR, radar, and multi-temporal data, and also possesses visualization, algorithm toolchain, and secondary development capabilities.

[0023] Step S121: Calculate the normalized water index that highlights the features of different types of coastlines for the first target image after preprocessing, and extract the spectral, texture parameters and geometric parameters of the first target image. The texture parameter is the gray-level co-occurrence matrix, and the geometric parameters include shape and size.

[0024] Specifically, different types of coastlines correspond to different characteristics; for example: Bedrock coastlines are composed of bedrock, with a winding coastline and numerous headlands and bays. They are characterized by a clear boundary between land and water, often with shadows cast by rock formations. The texture is coarse, the shapes are irregular, and they are frequently accompanied by mountains and vegetation.

[0025] Sandy coastlines consist of loose sand and gravel, with relatively straight shorelines and wide beaches. They are characterized by bright colors (especially on dry beaches), fine and uniform texture, and smooth, bright stripes along the waterline.

[0026] Muddy coastlines consist of silt and mud, with straight shorelines and wide, gently sloping mudflats. They are characterized by a dark, uniform color tone, smooth texture, blurred water-land boundaries, and the frequent development of tidal channels.

[0027] Biotidal coastlines are constructed or modified by organisms such as mangroves and coral reefs. They are characterized by dark green, coarse-textured patches of mangroves and bright or dark patches of coral reefs in clear, shallow waters.

[0028] Artificial coastlines consist of man-made structures such as breakwaters and wharves. Their characteristics include a regular, straight shoreline with a bright white color, clearly distinguishing them from natural coastlines.

[0029] Step S121: Segment the first target image into object patches with similar features based on the normalized water index, spectrum, texture parameters, and geometric parameters.

[0030] Specifically, segmenting into object patches rather than analyzing them on a per-pixel basis allows for better utilization of geometric features such as shape and context.

[0031] Step S122: Classify the segmented object patches using supervised classification (Support Vector Machine SVM) or unsupervised classification (such as K-means algorithm), and perform patch merging and smoothing post-processing on the classification results to obtain the final coastline type distribution map.

[0032] Step S130: The cloud server builds and trains a waterline recognition model based on a convolutional neural network to obtain an optimal waterline recognition model.

[0033] Specifically, a well-trained convolutional neural network model can automatically identify coastlines from remote sensing images of the coast. The core principle is to treat coastline identification as a "semantic segmentation" task: the convolutional neural network model determines for each pixel in the coastline image whether it is land, seawater, or other land features (such as beaches, vegetation, etc.), thereby accurately delineating the boundary line between land and sea (i.e., the waterline) at the pixel level.

[0034] Step S140: The cloud server inputs the preprocessed first target image into the preferred waterline recognition model to obtain the output waterline recognition result.

[0035] Step S150: The cloud server performs correction based on the waterline recognition results to obtain the coastline image of the target area, and sends it to the management terminal for display.

[0036] Step S160: The cloud server determines whether there is a risk of rapid coastline erosion in the target area based on the coastline image.

[0037] Specifically, in this embodiment, the preferred waterline recognition model outputs an instantaneous land-sea boundary line, which is obtained based on semantic segmentation of remote sensing images and corresponds to the instantaneous land-sea boundary during satellite imaging. The coastline in the geological exploration sense usually refers to the "multi-year average high tide line", which needs to be corrected for tide level based on the output waterline line to obtain the final coastline image of the target area. Subsequently, the coastline image can be used to determine whether there is a risk of rapid coastline erosion in the target area, thereby facilitating managers to conveniently monitor and observe changes in the coastline of the target area.

[0038] This invention proposes a coastal erosion identification and analysis method based on remote sensing shoreline classification, providing a more efficient and accurate technical solution for monitoring coastal erosion. First, multiple remote sensing images of the target area taken by remote sensing satellites are acquired and used as the first target image, which is then preprocessed. Next, a waterline identification model based on a convolutional neural network is constructed and trained to obtain an optimized waterline identification model. The preprocessed first target image is then input into the optimized waterline identification model to obtain the output waterline identification result, thus obtaining the instantaneous waterline of the target area. The instantaneous waterline is then corrected to obtain the coastline image of the target area, which is then sent to a management terminal for display. This solution, based on remote sensing images and a convolutional neural network waterline identification model, can automatically and efficiently obtain coastline images of the target area within different revisit periods, enabling managers to monitor and observe coastline changes in the target area in a timely and convenient manner.

[0039] As attached Figure 2 As shown, step S160 above specifically includes the following steps: S161: The cloud server calculates the coastline erosion rate based on the coastline image.

[0040] S162: The cloud server determines the coastline erosion warning level of the target area based on the coastline erosion rate and the coastline type of the target area.

[0041] Specifically, different coastal types correspond to different erosion warning levels. Therefore, this scheme determines the coastal erosion warning level of the target area based on the coastline erosion rate and the coastal type of the target area. The erosion warning levels include severe erosion, strong erosion, ordinary erosion, slight erosion, and stable erosion.

[0042] 1. Bedrock Coast: Bedrock coasts are composed of highly resistant rocks, with an extremely low erosion rate, primarily manifested in the retreat of sea cliffs. Their classification thresholds are the most stringent. Severe erosion: Erosion rate greater than 0.1 meters / year; Strong erosion: erosion rate greater than 0.01 and less than or equal to 0.1 m / year; Ordinary erosion: erosion rate greater than 0.001 and less than or equal to 0.01 m / year; Micro-erosion: Erosion rate greater than 0 and less than or equal to 0.001 m / year; Stable: Erosion rate is approximately 0 meters per year.

[0043] 2. Sandy coast: Sandy coasts are composed of loose sand and gravel. They are the most actively eroded and most extensively studied type of coast, with the most mature classification standards. Severe erosion: Erosion rate greater than 3.0 meters / year; Strong erosion: erosion rate greater than 2.0 and less than or equal to 3.0 meters / year; Ordinary erosion: erosion rate greater than 1.0 and less than or equal to 2.0 meters / year; Micro-erosion: Erosion rate greater than 0.5 and less than or equal to 1.0 m / year; Stable: Erosion rate less than or equal to 0.5 m / year.

[0044] 3. Muddy coasts: Muddy coasts are composed of fine-grained sediments (silt, mud), which are extremely sensitive to hydrodynamic changes. Once erosion occurs, the rate is usually very fast, so their classification threshold is much higher than that of sandy coasts. Severe erosion: Erosion rate greater than 10.0 meters / year; Strong erosion: erosion rate greater than 5.0 and less than or equal to 10.0 meters / year; Ordinary erosion: erosion rate greater than 1.0 and less than or equal to 5.0 meters / year; Micro-erosion: Erosion rate greater than 0.5 and less than or equal to 1.0 m / year; Stable: Erosion rate less than or equal to 0.5 m / year.

[0045] 4. Biotidal Coasts: Erosion classification of biotidal coasts (such as mangrove and coral reef coasts) is rarely seen in published literature, and their stability is highly dependent on the health of the biological community. Considering their high ecological value and high vulnerability, this scheme adopts the approach of ecological vulnerability assessment and sets relatively strict classification standards: Severe erosion: Erosion rate greater than 5.0 meters / year; Strong erosion: erosion rate greater than 2.0 and less than or equal to 5.0 meters / year; Ordinary erosion: erosion rate greater than 0.5 and less than or equal to 2.0 meters / year; Micro-erosion: Erosion rate greater than 0.1 and less than or equal to 0.5 meters per year; Stable: Erosion rate less than or equal to 0.1 m / year.

[0046] 5. Stability assessment of artificial coastlines (such as seawalls and breakwaters) relies more on engineering structure monitoring. Currently, there is a lack of unified erosion rate classification standards. Erosion typically refers to the damage to structures or the erosion of the foundation. This scheme provides a reference framework: Severe erosion: Erosion rate greater than 5.0 meters / year; Strong erosion: erosion rate greater than 2.0 and less than or equal to 5.0 meters / year; Ordinary erosion: erosion rate greater than 0.5 and less than or equal to 2.0 meters / year; Micro-erosion: Erosion rate greater than 0.1 and less than or equal to 0.5 meters per year; Stable: Erosion rate less than or equal to 0.1 m / year.

[0047] In the second embodiment of the coastal erosion identification and analysis method based on remote sensing shoreline classification proposed in this invention, based on the first embodiment, step S120 includes the following steps: Step S211: Radiometric calibration processing: Convert the DN value (Digital Number) of the first target image into a physical radiance value to ensure an accurate correspondence between the image's radiometric measurements and the surface radiance values.

[0048] Specifically, radiometric calibration is performed using ENVI 5.3 software to eliminate the effects of atmospheric scattering and absorption.

[0049] Step S212: Atmospheric correction processing.

[0050] Specifically, atmospheric correction is performed using ENVI 5.3 software to obtain the true surface reflectance.

[0051] Step S213: Perform image enhancement processing on the first target image using the histogram equalization method.

[0052] Specifically, histogram equalization improves global contrast by redistributing pixel gray values ​​evenly, thereby enhancing the image quality of the first target image; edge sharpening can also be used to enhance high-frequency edge information.

[0053] Step S214: Perform denoising processing on the first target image using the Gaussian filtering method.

[0054] In the third embodiment of the coastal erosion identification and analysis method based on remote sensing shoreline classification proposed in this invention, based on the first embodiment, step S130 includes the following steps: Step S310: The cloud server acquires remote sensing images for training, preprocesses the remote sensing images, and marks the preprocessed remote sensing images as training images.

[0055] Specifically, the remote sensing images used for training can be commonly used free medium-resolution data (such as Sentinel-2, with a resolution of 10m) or commercial high-resolution data (such as the Gaofen series).

[0056] Sentinel-2 is a high-resolution optical remote sensing satellite under the European Space Agency's (ESA) Copernicus program. It is a three-satellite network with global coverage, a revisit period of 5 days, and its data is freely available. The Gaofen series is China's independent high-resolution Earth observation system, launched successively from GF-1 to GF-14 starting in 2013, covering optical, SAR, hyperspectral, and stereo mapping.

[0057] Step S320: The cloud server performs pixel-level annotation on the training images to distinguish between land and water, and generates corresponding result labels (the result labels are binary images and are consistent with the pixels of the corresponding training images). Among them, the water pixels in the training images are labeled as 0 (i.e., black), and the land pixels in the training images are labeled as 1 (i.e., white).

[0058] Specifically, in this embodiment, ArcGIS software tools are used to perform pixel-level annotation on the training images (i.e., each training image is annotated) to distinguish between the background (land) and the target (water). For artificial coastlines, the waterline can be directly labeled as the coastline. The convolutional neural network waterline recognition model learns the mapping from pixels in the original remote sensing image to 0 / 1 labels. The generated label is a binary image, and the pixel size of the binary image is consistent with the pixel size of the corresponding training image.

[0059] Step S330: The cloud server performs block processing on the training images: the training images are cropped into several training sub-images, wherein the pixel size of the training sub-images is 256x256 or 512x512.

[0060] Specifically, in this step, after cropping the training image into several training sub-images, the corresponding result labels (i.e., binary images) of the training images will also be cropped so that each training sub-image corresponds to one result label.

[0061] Specifically, due to the memory limitations of the convolutional neural network waterline recognition model and the characteristics of high-resolution images, the training images need to be cropped into several training sub-images. This step requires simultaneous block processing of the original image and the corresponding label image.

[0062] Specifically, in the subsequent training steps, the training sub-images are used as input parameters for the waterline recognition model, and the corresponding result labels of the training sub-images are used as output parameters to train the waterline recognition model.

[0063] Furthermore, to prevent overfitting of the waterline recognition model and improve its generalization ability, each input patch can be randomly transformed in real time during subsequent model training to simulate images under different conditions, allowing the waterline recognition model to learn more robust features.

[0064] Step S340: The cloud server divides all training sub-images and their corresponding result labels into training sets and validation sets according to the quantity ratio.

[0065] Specifically, the ratio of the training set to the validation set is 85%:15%.

[0066] Specifically, in this partitioning step, the training sub-images and their corresponding result labels are treated as a whole (which can be understood as a data packet), and then partitioned proportionally based on the number of data packets. For example, if there are a total of 100 data packets, 85 of them are divided into the training set and the other 15 data packets are divided into the validation set.

[0067] In the fourth embodiment of the coastal erosion identification and analysis method based on remote sensing shoreline classification proposed in this invention, based on the third embodiment, after step S350, the following steps are further included: Step S410: The cloud server constructs a waterline recognition model based on the U-Net architecture. The waterline recognition model structure consists of an encoder and a decoder, enabling it to capture both global context and fine local details simultaneously. The encoder (downsampling path) is used to: gradually extract high-level semantic features of the image through repeated convolution and pooling operations, while reducing the resolution of the feature map. The decoder (upsampling path) is used to: gradually restore the resolution of the feature map through transposed convolution or upsampling operations. The input layer of the waterline recognition model structure is a single channel (grayscale image).

[0068] Specifically, U-Net (Convolutional Networks for Biomedical Image Segmentation) is a classic neural network model for image segmentation, widely used in remote sensing, semantic segmentation, defect detection and other scenarios, and performs excellently for segmenting small-sample, high-resolution images.

[0069] The core feature of the U-Net architecture is skip connections, which stitch the high-resolution feature maps from each stage of the encoder onto the corresponding upsampling results of the decoder, providing the decoder with rich spatial and detailed information, thereby achieving accurate pixel-level localization.

[0070] Step S420: The cloud server uses the binary cross-entropy loss function as the loss function of the waterline recognition model.

[0071] Specifically, the loss function is used to measure the difference between the model's prediction and the true label. The binary cross-entropy (BCE) loss function is a commonly used loss function for binary classification and segmentation tasks.

[0072] Step S430: The cloud server uses the AdamW algorithm as the optimization algorithm for the waterline recognition model.

[0073] Specifically, AdamW, with its adaptive learning rate feature, makes the training process more stable, converges faster, and is more robust to hyperparameters.

[0074] Step S440: The cloud server uses the intersection-over-union (IoU), accuracy, and F1 score as evaluation metrics for the waterline recognition model.

[0075] Specifically, the evaluation metrics are used to monitor the performance of the waterline recognition model on the validation set and do not participate in gradient backpropagation. The core evaluation metric is IoU, which calculates the ratio of the overlap between the predicted and ground truth regions to the total area and is the gold standard for measuring segmentation accuracy.

[0076] Precision means: how many of the samples predicted as "positive" by the waterline recognition model are actually "positive", that is, whether the model's recognition result is correct.

[0077] The F1 score is also an important evaluation metric. It is a comprehensive metric used to evaluate model performance in classification problems. It is the harmonic mean of precision and recall.

[0078] Step S450: The cloud server inputs the training set into the waterline recognition model for training.

[0079] Step S460: During the training process of the waterline recognition model, the cloud server calculates the error of the waterline recognition model based on the loss function and updates the learnable weight parameters in the waterline recognition model through the backpropagation algorithm. The update process requires multiple iterations (generally 100-200).

[0080] Specifically, the aforementioned weighting parameters include: 1. Convolutional layer parameters: These are the most important and numerous learnable parameters in the U-Net architecture, responsible for extracting image features.

[0081] In the encoder and decoder of the U-Net architecture, 3×3 and 1×1 convolutional kernels (or filters) are widely used. Each convolutional kernel has corresponding weights and an optional bias term.

[0082] 2. Weights: A convolutional kernel is essentially a weight matrix (for example, a 3×3 convolutional kernel contains 9 weights). These weights determine how the convolutional kernel performs a weighted summation on the input image or feature map to extract specific features (such as edges, textures, etc.).

[0083] 3. Bias: Each convolution kernel is usually appended with a bias term. Adding this bias after the weighted sum of the convolution calculation can increase the flexibility of the model and help it fit the data better.

[0084] 4. Batch Normalization (BN) layer parameters: In this embodiment, the U-Net architecture introduces batch normalization layers between convolutional layers to accelerate training and improve model stability. The batch normalization layers themselves contain two types of learnable parameters used to restore or transform the data after normalization, allowing the U-Net architecture to learn the data distribution best suited for the current task. Specifically, these include: a scaling factor for scaling the normalized data; and an offset factor for shifting the normalized data.

[0085] 5. Transposed convolutional layer parameters: In the decoder section of the U-Net architecture in this embodiment, an upsampling operation is required to restore the feature map to the size of the original input image. A common and learnable method is the transposed convolution operation; similar to standard convolution, the transposed convolution also has its own learnable weights and bias parameters, enabling the network to learn the optimal upsampling method.

[0086] After each iteration, the cloud server uses the validation set to evaluate the performance of the waterline recognition model and adjusts the learning rate based on the evaluation results (e.g., validation set IoU no longer increases).

[0087] Step S470: The cloud server inputs the validation set into the trained waterline recognition model and determines whether the evaluation index of the trained waterline recognition model meets the first preset condition. The first preset condition is: the intersection-union ratio is greater than or equal to the first preset value (e.g., 0.9), the accuracy is greater than or equal to the second preset value (98%), and the F1 score is greater than or equal to the third preset value (e.g., 0.58).

[0088] Step S480: If the first preset condition is met, the cloud server will use the trained waterline recognition model as the preferred waterline recognition model.

[0089] In the fifth embodiment of the coastal erosion identification and analysis method based on remote sensing shoreline classification proposed in this invention, based on the first embodiment, step S150 includes the following steps: Step S510: The cloud server converts the output waterline recognition result into a binary mask image based on a set segmentation threshold (e.g., 0.5), wherein the format of the waterline recognition result output by the waterline recognition model is a probability map.

[0090] Specifically, the above steps involve binarization: setting a threshold (such as 0.5) to convert the probability map into a binary mask image containing only 0 (water) and 1 (land).

[0091] Step S520: The cloud server performs opening and closing operations on the binary mask image in sequence to remove the "salt and pepper" noise patches in the binary mask image and fill the tiny holes in the land or water areas to improve the smoothness of the final coastline.

[0092] Specifically, the principle of opening operation is to erode first and then expand, while the principle of closing operation is to expand first and then erode.

[0093] Step S530: The cloud server uses the Canny edge detection algorithm to extract the boundary lines of water bodies and land from the binary mask image. The extracted boundary lines of water bodies and land are the gridded water edge lines.

[0094] Specifically, the Canny edge detection algorithm is a classic and optimal edge detection algorithm, widely used in remote sensing images, image segmentation, U-Net preprocessing, and target contour extraction. It has the advantages of low false detection, high localization, and single-pixel edge detection.

[0095] Step S540: The cloud server performs vectorization processing on the extracted raster boundary lines to obtain a vector line file in GeoJSON format.

[0096] Step S550: The cloud server uses the Douglas-Peucker algorithm to smooth the waterline vector lines in the vector line file.

[0097] Specifically, the Douglas-Peucker algorithm is a curve / contour simplification algorithm. Its core uses recursive divide-and-conquer combined with distance thresholding to significantly reduce the number of points while preserving the shape as much as possible. It is widely used in GIS, remote sensing contour simplification, map simplification, trajectory compression, U-Net contour preprocessing, etc. This step performs smoothing to remove unnecessary jagged edges and nodes, making the waterline vector line more consistent with the shape of the natural coastline.

[0098] In the sixth embodiment of the coastal erosion identification and analysis method based on remote sensing shoreline classification proposed in this invention, based on the fifth embodiment, after step S550, the following steps are further included: Step S601: The cloud server selects two first target images continuously acquired by the remote sensing satellite in the same revisit of the target area. The interval between the satellite imaging times of the two first target images must be less than or equal to a first preset time (e.g., 2 minutes; in this embodiment, the interval between the satellite imaging times of the two first target images is 1 minute).

[0099] Specifically, the satellite imaging time is accurate to the minute in UTC, which is used to determine the instantaneous tide level. The UTC time can be obtained from the metadata of the corresponding remote sensing image.

[0100] Step S602: The cloud server obtains the vector line files corresponding to two consecutive first target images based on the preferred waterline recognition model, and marks the vector line file whose waterline vector line is closer to the land in the two vector line files corresponding to the first target images as the first vector line file, and marks the other vector line file as the second vector line file.

[0101] Specifically, this method acquires two images of the first target, primarily for subsequent calculation of the beach slope in the target area. Therefore, it is necessary to ensure that there is a relatively obvious distance (e.g., greater than 2m) between the waterline vector lines corresponding to the two first target images. Thus, if the waterline vector lines in the vector line files corresponding to the two first target images are relatively close and it is impossible to determine which one is closer to the land, then two new first target images are selected to more easily distinguish the first vector line file and the second vector line file.

[0102] Step S603: The cloud server marks the first target image corresponding to the first vector line file as the first original image, and marks the first target image corresponding to the second vector line file as the second original image; Step S604: The cloud server obtains the digital elevation model of the target area.

[0103] Specifically, the digital elevation model (DEM) here has high resolution and can cover the intertidal zone of the target area, with the coverage extending from the low tide line to above the high tide line. The digital elevation model in this embodiment samples the coastal tidal flat topography dataset of China with a resolution of 100 meters (elevation accuracy of about 0.4 meters).

[0104] Step S605: The cloud server, based on the digital elevation model, uses ArcGIS software to set a first target point every preset length (e.g., 10 meters) along the waterline vector line of the first vector line file, and generates a coastline passing through the target point. The vertical plane of the coastline is perpendicular to the tangent between the first target point and the waterline vector line.

[0105] Specifically, the coastal profile here is the spatial carrier for which subsequent correction calculations need to be performed.

[0106] Step S606: The cloud server, based on the digital elevation model, uses ArcGIS software to obtain the intersection point of the coastline corresponding to the first target point and the waterline vector line of the second vector line file, and marks it as the second target point.

[0107] Specifically, in layman's terms, the second target point is the intersection of the waterline vector line of the second vector line file as the waves move from the first target point towards the land.

[0108] Step S607: The cloud server obtains the instantaneous tide height value of the target area at the time of satellite imaging of the first original image, the average high tide value, and the instantaneous tide height value of the target area at the time of satellite imaging of the second original image.

[0109] Specifically, in this embodiment, an open-source global tide model (such as the CoastSat toolbox) is used to query the tide level at any location and time. The CoastSat toolbox has integrated the FES2022 global tide model, so it is possible to obtain the tide level (including instantaneous tide level and mean spring tide high tide level) of the target area at the time of remote sensing image imaging based on the CoastSat toolbox.

[0110] Step S608: The cloud server calculates the shore slope of the first target point based on the distance between the first target point and the second target point, the instantaneous tide height of the target area at the time of satellite imaging of the first original image, and the instantaneous tide height of the target area at the time of satellite imaging of the second original image. : , In the formula, The instantaneous tidal height value, in meters, corresponds to the satellite imaging time of the target area in the first original image. The instantaneous tidal height value, in meters, corresponds to the satellite imaging time of the target area in the second original image. This represents the distance between the first target point and the second target point, in meters (m).

[0111] Step S609: The cloud server uses the beach slope based on the first target point. Calculate the calibration distance between the coastline and the waterline at the first target point: , In the formula, L is the calibration distance, which is the distance the waterline moves toward the land side, in meters; The mean high tide level of the target area at the time of satellite imaging in the first original image, in meters; The instantaneous tidal height value of the target area at the time of satellite imaging in the first original image, in meters.

[0112] Step S610: Based on the digital elevation model of the target area, the cloud server translates each first target point along the vertical direction of the coastline towards the land by a calibration distance to obtain the intersection with the land and marks it as a correction point. The correction points are connected in sequence to form a coastline vector line. The Douglas-Puk algorithm is used to smooth the coastline vector line, and the smoothed coastline vector line is superimposed on the first original image to obtain the final coastline image.

[0113] Specifically, the Douglas-Puk algorithm is used to thin out and simplify polylines / contours composed of ordered discrete points, reducing the number of vertices while preserving the overall shape and key inflection points.

[0114] Connecting all the calculated correction points sequentially generates a preliminarily corrected coastline vector line. However, due to calculation errors or terrain noise, the coastline may appear jagged. Therefore, the Douglas-Peucker algorithm is used for smoothing to reduce complexity while preserving the main shape.

[0115] In the seventh embodiment of the coastal erosion identification and analysis method based on remote sensing shoreline classification proposed in this invention, based on the first embodiment, step S160 includes the following steps: Step S710: The cloud server establishes a correspondence between the obtained coastline image and the satellite imaging time of the first target image corresponding to the coastline image, wherein each revisit period corresponds to one coastline image.

[0116] Specifically, in this embodiment, each rework cycle is set to generate only one coastline image to facilitate subsequent comparison of coastline positions in the images.

[0117] Step S720: The cloud server selects a first preset number (e.g., 3) of coastline images from the past and marks them as second target images. The last second target image is the coastline image closest to the current time, and the first second target image is the coastline image farthest from the current time. The interval between the satellite imaging times of any two adjacent second target images is greater than the second preset time (e.g., 1 month).

[0118] Specifically, the interval is set to be greater than one month to obtain more obvious coastline erosion data. In this embodiment, the interval between the satellite imaging times of two adjacent second target images is uniformly set to two months.

[0119] Step S730: The cloud server determines fixed landmarks (such as lighthouses or other fixed structures) in the second target image, wherein the fixed landmarks are located in the land portion of the second target image, and ensures that the same fixed landmark is included in all second target images.

[0120] Step S740: The cloud server obtains the center pixel of the fixed marker in the second target image, and determines the first reference line and the second reference line based on the center pixel. The first reference line is the line segment connecting the start and end points of the coastline vector line in the second target image, and the second reference line is the line segment passing through the center pixel and traversing the second target image, and the second reference line is parallel to the first reference line.

[0121] Specifically, the first reference line reflects the overall direction of the coastline's extension, while the second reference line is a relatively fixed reference line used to compare distances with the first reference line to determine the erosion distance of the coastline.

[0122] Step S750: The cloud server selects a second preset number of target reference points from the coastline vector lines of each second target image.

[0123] Specifically, the second preset number is determined based on the total length of the coastline vector line. The longer the total length of the coastline vector line, the more target reference points are needed. In this embodiment, the second preset number is 20.

[0124] Furthermore, the specific method for determining the aforementioned target reference point is as follows: the first reference line of the second target image is divided into equal parts to obtain a second preset number of equal parts, with the starting point of the coastline vector line being the first equal part and the ending point of the coastline vector line being the last equal part. Each equal part is used to draw an extension line perpendicular to the first reference line, and the intersection of each extension line with the coastline vector line is the selected target reference point. Thus, it can be seen that the starting point of the coastline vector line is the first target reference point, and the ending point of the coastline vector line is the last target reference point.

[0125] Step S760: The cloud server obtains the distance values ​​between each target reference point and the second reference line in each second target image. ,in, Let be the distance between the j-th target reference point of the i-th second target image and the second reference line, in meters, 1≤i≤I, where I is the first preset quantity (3), 1≤j≤J, where J is the second preset quantity (20).

[0126] Step S770: Cloud server based Calculate the coastline erosion rate and determine whether there is a risk of excessively rapid erosion in the target area based on the coastline erosion rate.

[0127] In the eighth embodiment of the coastal erosion identification and analysis method based on remote sensing shoreline classification proposed in this invention, based on the seventh embodiment, the cloud server is based on... The formula for calculating the rate of coastal erosion is as follows: , , In the formula, The coastline erosion rate is the ratio of the (i+1)th second target image to the ith second target image. The interval between the satellite imaging time corresponding to the (i+1)th second target image and the satellite imaging time corresponding to the ith second target image is 1 minute.

[0128] Specifically, the unit of the coastline erosion rate of the second target image obtained through the above calculation formula is m / mon.

[0129] Step S770, which describes determining whether there is a risk of excessively rapid erosion in a target area based on the coastline erosion rate, includes the following steps: Step S810: The cloud server determines whether the second preset condition is met: Traverse i from 1 to I-1, and the coastline erosion rate of all (i+1)th second target images is greater than the preset threshold compared to the i-th second target image.

[0130] Specifically, different types of coastlines have different preset thresholds. For sandy coastlines, the preset threshold is 0.17 m / mon; for silty coastlines, the preset threshold is 0.83 m / mon.

[0131] Step S820: If the second preset condition is met, the cloud server based on Further assessment is needed to determine if any localized interference issues exist.

[0132] The specific steps to determine whether there are local interference issues are as follows: Cloud server judgment Does the third preset condition meet? For the following formula, traversing i from 1 to I-1 and j from 1 to J, the calculated result is... All are less than the fourth preset value (e.g., 5%). , Specifically, in the above technical formula The meaning is: for any j, the distance between the j-th target reference point and the second reference line in the (i+1)-th second target image is the difference between the distance between the same (j-th) target reference point and the second reference line in the ith second target image, minus the average erosion difference. The absolute value of ) is then divided by the average erosion difference; It reflects whether the target reference points on the coastline are synchronously and uniformly eroded towards the land.

[0133] Under normal circumstances, erosion along the coastline towards the land is synchronous and uniform, that is... The calculated value is relatively small (less than 5%); if the calculated value is... A larger value indicates that the corresponding j-th target reference point has a significantly larger or smaller erosion amount compared to other target reference points (i.e., there are local disturbances, such as changes in the coastline caused by human activities (e.g., sand dredging, artificial reclamation, etc.)). It cannot be simply and directly identified as a risk of excessively rapid erosion and requires manual verification to rule out this anomaly.

[0134] If the third preset condition is met, the cloud server determines that there is no local interference; if the third preset condition is not met, the cloud server determines that there is local interference.

[0135] Step S830: If there are no interference issues, the cloud server determines that the target area is at risk of being eroded too quickly and generates an alarm message to send to the management terminal.

[0136] The present invention also proposes a coastal erosion identification and analysis system based on remote sensing shoreline classification, which applies a coastal erosion identification and analysis method based on remote sensing shoreline classification; the system includes a cloud server.

[0137] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0138] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A coastal erosion identification and analysis method based on remote sensing shoreline classification, characterized in that, A coastal erosion identification and analysis system based on remote sensing shoreline classification; the system includes a cloud server; the method includes: The cloud server obtains multiple remote sensing images of the target area taken by the remote sensing satellite each time it revisits the target area through the network port, and uses them as the first target image; The cloud server preprocesses the acquired first target image; The cloud server builds and trains a waterline recognition model based on a convolutional neural network to obtain an optimal waterline recognition model; The cloud server inputs the preprocessed first target image into the optimized waterline recognition model to obtain the output waterline recognition result; The cloud server performs corrections based on the waterline recognition results to obtain a coastline image of the target area. The cloud server uses the coastline image to determine whether there is a risk of rapid coastline erosion in the target area.

2. The coastal erosion identification and analysis method based on remote sensing shoreline classification according to claim 1, characterized in that, The cloud server preprocesses the acquired first target image, including: Radiometric calibration: The DN value of the first target image is converted into a physical radiance value to ensure an accurate correspondence between the image's radiometric measurements and the surface radiance value. Atmospheric correction processing; Image enhancement processing is performed on the first target image using histogram equalization. The first target image is denoised using a Gaussian filtering method.

3. The coastal erosion identification and analysis method based on remote sensing shoreline classification according to claim 1, characterized in that, The cloud server constructs and trains a waterline recognition model based on a convolutional neural network to obtain an optimal waterline recognition model, including: The cloud server acquires remote sensing images for training, preprocesses the remote sensing images, and marks the preprocessed remote sensing images as training images. The cloud server performs pixel-level annotation on the training images to distinguish between land and water, and generates corresponding result labels. Water pixels in the training images are labeled as 0, and land pixels in the training images are labeled as 1. The cloud server performs block processing on the training images: the training images are cropped into several training sub-images; The cloud server divides all training sub-images and their corresponding result labels into training and validation sets according to their quantity ratio.

4. The coastal erosion identification and analysis method based on remote sensing shoreline classification according to claim 3, characterized in that, The cloud server divides all training sub-images and their corresponding result labels into training and validation sets proportionally, and then includes: A cloud server constructs a waterline recognition model based on the U-Net architecture. The waterline recognition model structure consists of an encoder and a decoder. The encoder is used to: gradually extract high-level semantic features of the image through repeated convolution and pooling operations, while reducing the resolution of the feature map. The decoder is used to: gradually restore the resolution of the feature map through transposed convolution or upsampling operations. The cloud server uses the binary cross-entropy loss function as the loss function for the waterline recognition model. The cloud server uses intersection-union ratio, accuracy, and F1 score as evaluation metrics for the waterline recognition model. The cloud server inputs the training set into the waterline recognition model for training; During the training process of the waterline recognition model, the cloud server calculates the error of the waterline recognition model based on the loss function and updates the learnable weight parameters in the waterline recognition model through the backpropagation algorithm. The update process requires multiple iterations. The cloud server inputs the validation set into the trained waterline recognition model and determines whether the evaluation index of the trained waterline recognition model meets the first preset condition. The first preset condition is: the intersection-union ratio is greater than or equal to the first preset value, the accuracy is greater than or equal to the second preset value, and the F1 score is greater than or equal to the third preset value. If the first preset condition is met, the cloud server will use the trained waterline recognition model as the preferred waterline recognition model.

5. The coastal erosion identification and analysis method based on remote sensing shoreline classification according to claim 1, characterized in that, The cloud server performs corrections based on the waterline recognition results to obtain a coastline image of the target area, including: The cloud server converts the output waterline recognition result into a binary mask image based on a set segmentation threshold. The format of the waterline recognition result output by the waterline recognition model is a probability map. The cloud server performs opening and closing operations on the binary mask image sequentially to remove noise patches in the binary mask image and fill in tiny holes in land or water areas. The cloud server uses the Canny edge detection algorithm to extract the boundary lines of water and land from the binary mask image. The extracted boundary lines of water and land are the gridded water edge lines. The cloud server performs vectorization processing on the extracted raster boundary lines to obtain vector line files; The cloud server uses the Douglas-Peucker algorithm to smooth the waterline vector lines in the vector line file.

6. The coastal erosion identification and analysis method based on remote sensing shoreline classification according to claim 5, characterized in that, The cloud server uses the Douglas-Peucker algorithm to smooth the waterline vector lines in the vector line file, and then includes: The cloud server selects two first target images that are continuously acquired by remote sensing satellites in the same revisit of the target area. The interval between the satellite imaging times of the two first target images must be less than or equal to a first preset time. The cloud server uses a preferred waterline recognition model to obtain vector line files corresponding to two consecutive first target images. The vector line file whose waterline vector is closer to the land is marked as the first vector line file, and the other vector line file is marked as the second vector line file. The cloud server marks the first target image corresponding to the first vector line file as the first original image, and marks the first target image corresponding to the second vector line file as the second original image; The cloud server acquires the digital elevation model of the target area; The cloud server uses ArcGIS software based on the digital elevation model to set a first target point every preset length along the waterline vector line of the first vector line file, and generates a coastline passing through the target point. The vertical plane of the coastline is perpendicular to the tangent between the first target point and the waterline vector line. The cloud server uses ArcGIS software based on the digital elevation model to obtain the intersection point of the coastline corresponding to the first target point and the waterline vector line of the second vector line file, and marks it as the second target point. The cloud server obtains the instantaneous tide height value of the target area at the time of satellite imaging of the first original image, as well as the average high tide level, and the instantaneous tide height value of the target area at the time of satellite imaging of the second original image. The cloud server calculates the shore slope of the first target point based on the distance between the first and second target points, the instantaneous tide level of the target area at the time of satellite imaging of the first original image, and the instantaneous tide level of the target area at the time of satellite imaging of the second original image. ; The cloud server is based on the shore slope of the first target point. Calculate the calibration distance between the coastline and the waterline at the first target point: The cloud server uses a digital elevation model of the target area to translate and calibrate each first target point along the vertical direction of the coastline towards the land, and then obtains the intersection point with the land. These intersection points are marked as correction points. The correction points are then connected sequentially to form a coastline vector line. The Douglas-Puk algorithm is used to smooth the coastline vector line, and the smoothed coastline vector line is superimposed on the first original image to obtain the final coastline image.

7. The coastal erosion identification and analysis method based on remote sensing shoreline classification according to claim 1, characterized in that, The cloud server determines whether there is a risk of rapid coastline erosion in the target area based on the coastline image, including: The cloud server establishes a correspondence between the obtained coastline image and the satellite imaging time of the first target image corresponding to the coastline image, wherein each revisit period corresponds to one coastline image. The cloud server selects a first preset number of coastline images from the past and marks them as second target images. The last second target image is the coastline image closest to the current time, and the first second target image is the coastline image farthest from the current time. The interval between the satellite imaging times of any two adjacent second target images is greater than the second preset time. The cloud server identifies fixed landmarks in the second target image, wherein the fixed landmarks are located in the land portion of the second target image, and ensures that the same fixed landmark is included in all second target images; The cloud server obtains the center pixel of the fixed marker in the second target image, and determines the first reference line and the second reference line based on the center pixel. The first reference line is the line segment connecting the start and end points of the coastline vector line in the second target image, and the second reference line is the line segment that passes through the center pixel and runs through the second target image, and the second reference line is parallel to the first reference line. The cloud server selects a second preset number of target reference points from the coastline vector lines of each second target image; The cloud server obtains the distance values ​​between each target reference point and the second reference line in each second target image. ,in, Let be the distance between the j-th target reference point of the i-th second target image and the second reference line, in meters, where 1 ≤ i ≤ I, I is the first preset quantity, 1 ≤ j ≤ J, and J is the second preset quantity; cloud servers are based on Calculate the coastline erosion rate and determine whether there is a risk of excessively rapid erosion in the target area based on the coastline erosion rate.

8. The coastal erosion identification and analysis method based on remote sensing shoreline classification according to claim 7, characterized in that, The cloud server is based on The formula for calculating the rate of coastal erosion is as follows: ， ， In the formula, The coastline erosion rate is the ratio of the (i+1)th second target image to the ith second target image. The interval between the satellite imaging time corresponding to the (i+1)th second target image and the satellite imaging time corresponding to the ith second target image is expressed in hours. The method of determining whether a target area is at risk of excessively rapid erosion based on the coastline erosion rate includes: The cloud server determines whether the second preset condition is met: traversing i from 1 to I-1, the coastline erosion rate of all (i+1)th second target images is greater than the preset threshold compared to the i-th second target image; If the second preset condition is met, the cloud server is based on Further assessment is needed to determine if any localized interference issues exist; If there are no interference issues, the cloud server determines that the target area is at risk of being eroded too quickly and generates an alarm message to send to the management terminal.

9. A coastal erosion identification and analysis system based on remote sensing shoreline classification, characterized in that, The coastal erosion identification and analysis method based on remote sensing shoreline classification as described in any one of claims 1-8 is applied; the system includes a cloud server.

Images