Method and system for inter-ice channel identification based on synthetic aperture radar satellite

Abstract

The application discloses a method and system for identifying ice channel based on synthetic aperture radar satellite, the method comprises: after obtaining the first deformation rate according to the synthetic aperture radar satellite data, generating the ice channel image label according to the first deformation rate; obtaining the second training set by amplifying the first training set in the ice channel image label, collecting the third training set from the second training set by the effective data center point sampling method, and then processing the first deformation rate to obtain the second deformation rate; training the neural network by using the third training set and the second deformation rate to obtain the trained neural network model; inputting the to-be-tested data into the trained neural network model to obtain the skeleton graph, performing the connected region solving operation on the skeleton graph, and then displaying the ice channel position coordinates on the static map according to the marking result and the splitting result.

Description

Technical Field

[0001] This application relates to the fields of computer vision and polar marine science and technology, and in particular to a method and system for identifying interglacial waterways based on synthetic aperture radar satellites. Background Technology

[0002] Interglacial channels refer to linear fractures in sea ice formed by waves, wind, and ocean currents. They serve as crucial windows for heat and water exchange between the ocean and atmosphere, influencing the ocean's heat balance and sea ice mass balance in polar regions. Interglacial channels are vital navigation routes; the opening of Arctic shipping routes will significantly shorten shipping distances between Asia, Europe, and North America. Therefore, analyzing the characteristics and patterns of Arctic interglacial channels is of significant economic and strategic importance. To better monitor changes in Arctic sea ice interglacial channels, since the 1970s, a series of field observation projects have been implemented both domestically and internationally, focusing on the surface atmosphere, physical ocean, and ecological environment of areas involving Arctic sea ice and interglacial channels. However, due to the harsh conditions and difficulties of field observation in the polar regions, the obtained observational data is limited. Nevertheless, with the development of spaceborne optical, infrared, and microwave sensors, a large amount of long-term, high-resolution image data acquired by satellites has provided significant assistance to the study of polar sea ice and interglacial channels.

[0003] Currently, some progress has been made in the identification and labeling of interglacial channels based on satellite data. Most related research utilizes the narrow width and near-straight-line shape of interglacial channels to identify them using graphical approaches. Common methods include image filtering, skeleton extraction, and Hough transform. Moderate Resolution Imaging Spectroradiometer (MODIS) data and Visible Infrared Imaging Radiometer (VIIRS) data remain commonly used research subjects. Meanwhile, due to the advantages of radar satellites in dealing with clouds, fog, and shadows, research using Synthetic Aperture Radar (SAR) satellite data has significantly improved the recognition performance of interglacial channel identification algorithms. In terms of algorithms, most methods still employ classic computer graphics algorithms based on empirical assumptions and manually designed features. However, with the development of artificial intelligence in recent years, the exploration of deep learning applications in channel segmentation has gradually attracted the attention of research institutions both domestically and internationally.

[0004] However, due to the multifractal characteristics of sea ice deformation and fracturing, interglacial channels are morphologically complex. This increases the difficulty of manually designing features for traditional graphics algorithms based on empirical assumptions. Furthermore, the narrow width of these channels constitutes a fine-grained image segmentation problem, further challenging the accuracy of algorithmic recognition. In addition, most existing deep learning-based algorithms only identify channels without distinguishing between different channels, while subsequent research on related physical processes requires background information carried by the same channel. Simultaneously, interference from polar nights and clouds makes it difficult for algorithms based on optical and thermal infrared satellites to achieve all-weather, all-day observation. They also face issues such as missed identifications due to cloud cover or false deletions due to excessive cloud removal, affecting recognition accuracy. These unfavorable factors have become key constraints on the further application of interglacial channel recognition algorithms. Summary of the Invention

[0005] This application provides a method and system for identifying interglacial waterways based on synthetic aperture radar satellites, which at least solves the problems of low accuracy in identifying interglacial waterways and the fact that existing deep learning-based algorithms only identify waterways without classifying different waterways.

[0006] This invention provides a method for identifying interglacial channels based on synthetic aperture radar satellites, the method comprising:

[0007] Data preprocessing steps: After calculating the first deformation rate based on the synthetic aperture radar satellite data, generate interglacial waterway image tags based on the first deformation rate;

[0008] Dataset augmentation steps: Augment the first training set in the ice channel image label to obtain the second training set; after collecting the third training set from the second training set using the effective data point sampling method, process the first deformation rate to obtain the second deformation rate.

[0009] Neural network model training steps: The neural network is trained using the third training set and the second deformation rate to obtain the trained neural network model;

[0010] Visualization output steps: Input the test data into the trained neural network model to obtain a skeleton map. After performing connected component solving on the skeleton map, display the location coordinates of the interglacial waterway on a static map based on the labeling and splitting results.

[0011] The above-described method for identifying interglacial waterways based on synthetic aperture radar satellites includes the following data preprocessing steps:

[0012] The synthetic aperture radar satellite data is obtained through the server, and the first deformation rate is calculated based on the synthetic aperture radar satellite data using the first formula.

[0013] Based on the first deformation rate, image labels for the interglacial waterway are generated using a classical graphics algorithm.

[0014] The above-described method for identifying interglacial waterways based on synthetic aperture radar satellites includes the following data augmentation step:

[0015] The first training set in the ice channel image tags is augmented by random cropping, random left-right flipping, random up-down flipping, and random rotation angle.

[0016] The first deformation rate is logarithmically transformed and normalized to obtain the second deformation rate.

[0017] The above-described method for identifying interglacial waterways based on synthetic aperture radar satellites includes the following neural network model training steps:

[0018] After obtaining the loss function through the second formula, the first optimizer is set using gradient descent and cosine annealing hot restart methods to obtain the second optimizer;

[0019] After constructing the neural network model, the neural network is trained using the third training set and the second deformation rate, and the prediction result is output.

[0020] The interglacial channel identification method based on synthetic aperture radar satellites described above further includes the neural network model training step as follows:

[0021] Based on the prediction results, the training loss is calculated using the loss function.

[0022] Based on the training loss, the parameters of the neural network model are updated by the second optimizer to obtain the trained neural network model.

[0023] The above-described method for identifying interglacial waterways based on synthetic aperture radar satellites includes the following step: [Further details about the visualization output step would be needed here.]

[0024] After the test data is input into the trained neural network model, the probability output corresponding to the test data is obtained.

[0025] After binarizing the probability output according to a preset threshold to obtain a binarized image, the binarized image is then thinned to obtain the skeleton map.

[0026] The above-described method for identifying interglacial waterways based on synthetic aperture radar satellites further includes the visualization result output step as follows:

[0027] The first marked result of the interglacial waterway is obtained by performing a connected component solution operation on the skeleton graph.

[0028] Based on the first marking result, the ice channel is split with the intersection point as the center to obtain the splitting result.

[0029] The above-described method for identifying interglacial waterways based on synthetic aperture radar satellites further includes the visualization result output step as follows:

[0030] Based on the splitting results, the skeleton graph is again subjected to connected region solving to obtain the second labeling results of the split interglacial waterways;

[0031] Based on the second labeling result, the split interglacial waterway is segmented, and the cosine similarity between each segment of the same interglacial waterway is calculated to obtain the location coordinates of the split interglacial waterway.

[0032] The above-described method for identifying interglacial waterways based on synthetic aperture radar satellites further includes the visualization result output step as follows:

[0033] After converting the location coordinates to latitude and longitude coordinates using a file containing location coordinates and latitude and longitude coordinates of the same scale, the location coordinates of the interglacial waterway are displayed on the static map using the Basemap method in Python.

[0034] This invention also provides an interglacial channel identification system based on synthetic aperture radar (SAR) satellites, characterized in that it is applicable to the aforementioned interglacial channel identification method based on SAR satellites, wherein the interglacial channel identification system based on SAR satellites includes:

[0035] Data preprocessing unit: After calculating the first deformation rate based on synthetic aperture radar satellite data, it generates interglacial waterway image tags based on the first deformation rate;

[0036] Dataset augmentation unit: augments the first training set in the ice channel image label to obtain a second training set, and then collects a third training set from the second training set using an effective data point sampling method, and processes the first deformation rate to obtain a second deformation rate;

[0037] Neural network model training unit: The neural network is trained using the third training set and the second deformation rate to obtain the trained neural network model;

[0038] Visualization output unit: Input the test data into the trained neural network model to obtain a skeleton map, perform connected component solving on the skeleton map, and display the location coordinates of the interglacial waterway on a static map based on the labeling and splitting results.

[0039] Compared to related technologies, the interglacial waterway identification method and system proposed in this invention, based on synthetic aperture radar (SAR) satellites, preprocesses numerical model data and SAR satellite data, generates corresponding labels for images using classical graphics algorithms, divides the data into training, validation, and test sets, and analyzes the deformation rate calculated based on numerical models and SAR. This overcomes the limitations of visible light satellites in handling clouds, fog, and shadows, achieving all-weather, day-and-night interglacial waterway identification. Furthermore, data augmentation is applied to the training set data, automatically extracting features through neural network learning, reducing reliance on the accuracy of manually designed features. New samples continuously improve the model's recognition and generalization abilities, reduce class imbalance, and increase data diversity. A loss function is constructed to train the LadderNet-based interglacial waterway identification algorithm model, significantly reducing the severe class imbalance caused by invalid data during training due to limitations in satellite observation capabilities. Based on the segmentation results output by the neural network model, waterways are divided and labeled using connected component solving, and the final visualization results are output. This combines deep learning with traditional graphics methods, enabling the labeling and visualization of features such as the length, shape, and location of each waterway in the image, facilitating research and analysis.

[0040] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0041] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0042] Figure 1 This is a flowchart of an inter-ice waterway identification method based on synthetic aperture radar satellite according to an embodiment of this application;

[0043] Figure 2 This is a flowchart illustrating the overall process framework for interglacial waterway identification based on synthetic aperture radar satellites according to embodiments of this application.

[0044] Figure 3 This is a schematic diagram of the LadderNet model and its training effect according to an embodiment of this application;

[0045] Figure 4This is a schematic diagram illustrating the effect of center point sampling according to an embodiment of this application;

[0046] Figure 5 This is a schematic diagram illustrating the post-processing and visualization process and effects according to embodiments of this application;

[0047] Figure 6 This is a schematic diagram of the intericy waterway identification system based on synthetic aperture radar satellite of the present invention.

[0048] The attached figures are labeled as follows:

[0049] Data preprocessing unit: 51;

[0050] Dataset augmentation units: 52;

[0051] Neural network model training units: 53;

[0052] Visualization results output unit: 54. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0054] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0055] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0056] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0057] This invention proposes a method and system for identifying interglacial waterways based on synthetic aperture radar (SAR) satellites. Using SAR satellite data provided by the Technical University of Denmark (DTU Space) as the research object, it leverages microwave remote sensing, which is unaffected by polar nights and fog, enabling all-weather, day-and-night observation of interglacial waterways, overcoming the limitations of visible light satellite observation capabilities. For fine-grained segmentation of interglacial waterways, a LadderNet-based interglacial waterway identification algorithm is provided. This algorithm utilizes calculated polar region deformation rate data for learning and automatically extracts and segments features from the image using a trained neural network model. This reduces the reliance of traditional graphics algorithms on feature selection based on empirical assumptions, effectively improving identification robustness and accuracy. A post-processing algorithm based on connected component solving is used to divide the segmented binary image, enabling the labeling of different waterways, thus facilitating the research and application of interglacial waterway identification results.

[0058] The embodiments of this application will be described below using the identification of interglacial waterways based on synthetic aperture radar satellites as an example.

[0059] Example 1

[0060] This embodiment provides a method for identifying interglacial waterways based on synthetic aperture radar satellites. Please refer to... Figure 1 , Figure 1This is a flowchart of an inter-ice waterway identification method based on synthetic aperture radar satellite according to an embodiment of this application; Figure 2 This is a flowchart illustrating the overall process framework for interglacial waterway identification based on synthetic aperture radar satellites according to embodiments of this application. Figure 3 This is a schematic diagram of the LadderNet model and its training effect according to an embodiment of this application; Figure 4 This is a schematic diagram illustrating the effect of center point sampling according to an embodiment of this application; Figure 5 This is a schematic diagram illustrating the post-processing and visualization process and effects according to embodiments of this application, such as... Figures 1 to 5 As shown, the method for identifying interglacial waterways includes:

[0061] Data preprocessing step S1: After calculating the first deformation rate based on the synthetic aperture radar satellite data, generate interglacial waterway image tags based on the first deformation rate;

[0062] Dataset augmentation step S2: Augment the first training set in the ice channel image label to obtain the second training set. After collecting the third training set from the second training set through the effective data center point sampling method, process the first deformation rate to obtain the second deformation rate.

[0063] Neural network model training step S3: The neural network is trained using the third training set and the second deformation rate to obtain the trained neural network model;

[0064] Visualization result output step S4: Input the test data into the trained neural network model to obtain the skeleton map, perform connected component solving operation on the skeleton map, and display the location coordinates of the interglacial waterway on the static map according to the labeling results and splitting results.

[0065] In this embodiment, the data preprocessing step S1 includes:

[0066] The synthetic aperture radar satellite data is obtained through the server, and the first deformation rate is calculated based on the synthetic aperture radar satellite data using the first formula.

[0067] Based on the first deformation rate, image labels for the interglacial waterway are generated using a classical graphics algorithm.

[0068] In specific implementation, such as Figures 1-2 As shown, data preprocessing is performed on synthetic aperture radar (SAR) satellite data. Specifically, firstly, SAR satellite data for the Arctic region containing shear and divergence information is obtained from the corresponding data server. Then, the deformation rate corresponding to each coordinate position is calculated using this data, as shown in the following formula:

[0069]

[0070] Secondly, corresponding labels are generated for the images; finally, the training set, validation set, and test set are divided into three data sets, with the ratio of the training set to the validation set being 9:1.

[0071] In this embodiment, the dataset augmentation step S2 includes:

[0072] The first training set in the ice channel image tags is augmented by random cropping, random left-right flipping, random up-down flipping, and random rotation angle.

[0073] The first deformation rate is logarithmically transformed and normalized to obtain the second deformation rate.

[0074] In specific implementation, such as Figure 4 As shown, the dataset augmentation is detailed as follows: First, the original input data is augmented through random cropping, random left-right flipping, random up-down flipping, and random rotation angles to alleviate overfitting, improve the model's generalization ability, and expand the sample space; second, as... Figure 4 As shown in the left figure, due to the limitations of the satellite's observation capabilities, a large proportion of invalid data exists in the original input data. To avoid class imbalance caused by invalid data, the sampling range is determined by detecting valid data in the original input data. During the loading of training data, the sampling position is judged, and only the region where the center point is valid data is sampled, such as... Figure 3 As shown in the right figure; finally, logarithmic transformation and normalization are performed on the deformation rate obtained in step 1.1, which can initially reduce outliers and facilitate calculation.

[0075] In this embodiment, the neural network model training step S3 includes:

[0076] After obtaining the loss function through the second formula, the first optimizer is set using gradient descent and cosine annealing hot restart methods to obtain the second optimizer;

[0077] After constructing the neural network model, the neural network is trained using the third training set and the second deformation rate, and the prediction result is output.

[0078] Based on the prediction results, the training loss is calculated using the loss function.

[0079] Based on the training loss, the parameters of the neural network model are updated by the second optimizer to obtain the trained neural network model.

[0080] In specific implementation, such as Figure 3As shown, the detailed training process for the neural network model is as follows: First, the loss function is determined. The problem of identifying ice channels is treated as a pixel-based classification problem, and cross-entropy is used as the loss function. The formula for calculating the cross-entropy loss in binary classification is as follows:

[0081]

[0082] Where N represents the total number of images in this batch, y i This represents the label of sample i, with 1 for positive class and 0 for negative class, p i This represents the probability that sample i is predicted as positive; secondly, the learning rate optimization method is determined. Using the Adam optimizer, gradient descent with adaptive moment estimation is employed for backpropagation of the model parameters, while cosine annealing with a hot restart method is used for learning rate scheduling; secondly, such as Figure 3 As shown, a neural network model is defined, and a LadderNet model based on two U-shaped encoder-decoder structures is constructed. Each U-shaped structure contains four downsampling operations and four upsampling operations. Features of the same scale are connected using skip-layer connections. To avoid excessive parameters, parameters are shared in the two convolutions of all downsampling modules, and the traditional pooling method for downsampling is replaced with convolutions with a stride of 2. Finally, the model is trained using the third training set and the second deformation rate, as follows: Figure 3 As shown, the input feature map is a single channel, the network outputs a two-channel prediction result, and the result is passed through the softmax function to obtain the probability output. Then, the loss of each iteration is calculated through the loss function set in step 3.1. The model parameters are updated through the set optimizer. At the same time, an early stopping mechanism is set. If the loss does not decrease for 6 consecutive training rounds, the training will automatically terminate, or the total number of training rounds will exceed the set value, the training will terminate, and the training parameters and model will be saved.

[0083] In this embodiment, the visualization result output step S4 includes:

[0084] After the test data is input into the trained neural network model, the probability output corresponding to the test data is obtained.

[0085] After binarizing the probability output according to a preset threshold to obtain a binarized image, the binarized image is then thinned to obtain the skeleton map.

[0086] The first marked result of the interglacial waterway is obtained by performing a connected component solution operation on the skeleton graph.

[0087] Based on the first marking result, the ice channel is divided with the intersection point as the center to obtain the division result;

[0088] Based on the splitting results, the skeleton graph is again subjected to connected region solving to obtain the second labeling results of the split interglacial waterways;

[0089] Based on the second labeling result, the split interglacial channel is segmented, and the cosine similarity between each segment of the same interglacial channel is calculated to obtain the location coordinates of the split interglacial channel.

[0090] After converting the location coordinates to latitude and longitude coordinates using a file containing location coordinates and latitude and longitude coordinates of the same scale, the location coordinates of the interglacial waterway are displayed on the static map using the Basemap method in Python.

[0091] In specific implementation, such as Figure 5 As shown, the model results output and visualization are detailed as follows: First, the model results are post-processed. The trained neural network model is read, the test data is input, and the corresponding probability output is obtained. First, the probability output of the image is binarized according to the set threshold. Then, the obtained binarized image is thinned to obtain a skeleton map. Then, the skeleton map is subjected to connected component solving to obtain all connected components existing in the current image. Each connected component is used as a preliminary label for different waterways. Second, connected component splitting. Since ideal instantaneously generated waterways generally do not have bifurcations, bends, or other shapes, if bifurcations or other situations occur, they are often regarded as multiple different waterways. To deal with such situations, it is necessary to split the waterway labeling results.

[0092] The detailed process for splitting the waterway marking results is as follows: First, based on the characteristic that the connectivity of intersecting waterways in the bifurcation point region is often greater than 3, the bifurcation waterway is split with the intersection point as the center. This splitting is repeated iteratively for intersection points with a connectivity greater than 3 until no intersection points with a connectivity greater than or equal to 3 remain in the graph. Then, a connected component calculation is performed again, and the split waterways are marked. Second, to handle phenomena such as loops and bends in the waterway recognition results, each waterway is first segmented, and then the cosine similarity between segments of the same waterway is calculated. The cosine similarity calculation formula is as follows:

[0093]

[0094] Where A i With B i This is a vector representation of every two waterway segments.

[0095] The calculated cosine similarity and the set threshold can be used to determine the bending angle between each segment. If the bending angle is too large, i.e., the cosine similarity is less than 0.7, the segment is broken at the bending point. For a circular waterway, the number of its endpoints is detected. If there is no endpoint with a connectivity of 1, it is determined to be a circular waterway, and any point is randomly selected to break it. For the visualization of the post-processing results, the position coordinates of the current post-processing results are first converted into latitude and longitude coordinates using a file containing position coordinates and latitude and longitude coordinates of the same scale. Then, the Basemap method in Python is used to visualize the post-processing results on a static map.

[0096] Example 2

[0097] This embodiment provides an interglacial waterway identification system based on synthetic aperture radar satellites. Please refer to... Figure 6 As shown, Figure 6 This is a schematic diagram of the interglacial waterway identification system based on synthetic aperture radar satellite of the present invention, as shown below. Figure 6 As shown, the interglacial waterway identification system based on synthetic aperture radar satellites includes:

[0098] Data preprocessing unit 51: After calculating the first deformation rate based on synthetic aperture radar satellite data, it generates interglacial waterway image tags based on the first deformation rate;

[0099] Dataset augmentation unit 52: augments the first training set in the ice channel image label to obtain a second training set, collects a third training set from the second training set through an effective data point sampling method, and processes the first deformation rate to obtain a second deformation rate.

[0100] Neural network model training unit 53: Trains the neural network using the third training set and the second deformation rate to obtain the trained neural network model;

[0101] Visualization output unit 54: Inputs the test data into the trained neural network model to obtain a skeleton map, performs connected component solving on the skeleton map, and displays the location coordinates of the interglacial waterway on a static map based on the labeling and splitting results.

[0102] In summary, the proposed method and system for identifying interglacial waterways based on synthetic aperture radar (SAR) satellites overcomes the limitations of visible light satellites in handling clouds, fog, and shadows by analyzing deformation rates calculated using numerical models and SAR. This enables all-weather, day-and-night identification of interglacial waterways. The system automatically extracts features from training data through neural network learning, reducing reliance on the accuracy of manually designed features and improving the model's recognition and generalization capabilities by learning from new samples. The effective data point sampling method significantly reduces the severe class imbalance problem caused by invalid data due to limitations in satellite observation capabilities during training. Furthermore, the combination of deep learning and traditional graphics methods enables the labeling and visualization of features such as the length, shape, and location of waterways in images, facilitating research and analysis.

[0103] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the scope of the appended claims.

Claims

1. A method for identifying interglacial waterways based on synthetic aperture radar satellites, characterized in that, The method for identifying interglacial waterways includes: Data preprocessing steps: After calculating the first deformation rate based on the synthetic aperture radar satellite data, generate interglacial waterway image tags based on the first deformation rate; Dataset augmentation steps: Augment the first training set in the ice channel image label to obtain the second training set; after collecting the third training set from the second training set using the effective data point sampling method, process the first deformation rate to obtain the second deformation rate. Neural network model training steps: The neural network is trained using the third training set and the second deformation rate to obtain the trained neural network model; Visualization output steps: Input the test data into the trained neural network model to obtain a skeleton map. After performing connected component solving on the skeleton map, display the location coordinates of the interglacial waterway on a static map based on the labeling and splitting results. The visualization result output step includes: After the test data is input into the trained neural network model, the probability output corresponding to the test data is obtained. After binarizing the probability output according to a preset threshold to obtain a binarized image, the binarized image is then thinned to obtain the skeleton map.

2. The method for identifying interglacial waterways according to claim 1, characterized in that, The data preprocessing steps include: The synthetic aperture radar satellite data is obtained through the server, and the first deformation rate is calculated based on the synthetic aperture radar satellite data using the first formula. Based on the first deformation rate, image labels for the interglacial waterway are generated using a classical graphics algorithm.

3. The method for identifying interglacial waterways according to claim 1, characterized in that, The dataset augmentation steps include: The first training set in the ice channel image tags is augmented by random cropping, random left-right flipping, random up-down flipping, and random rotation angle. The first deformation rate is logarithmically transformed and normalized to obtain the second deformation rate.

4. The method for identifying interglacial waterways according to claim 1, characterized in that, The neural network model training steps include: After obtaining the loss function through the second formula, the first optimizer is set using gradient descent and cosine annealing hot restart methods to obtain the second optimizer; After constructing the neural network model, the neural network is trained using the third training set and the second deformation rate, and the prediction result is output.

5. The method for identifying interglacial waterways according to claim 4, characterized in that, The neural network model training steps also include: Based on the prediction results, the training loss is calculated using the loss function. Based on the training loss, the parameters of the neural network model are updated by the second optimizer to obtain the trained neural network model.

6. The method for identifying interglacial waterways according to claim 1, characterized in that, The visualization output step also includes: The first marked result of the interglacial waterway is obtained by performing a connected component solution operation on the skeleton graph. Based on the first marking result, the ice channel is split with the intersection point as the center to obtain the splitting result.

7. The method for identifying interglacial waterways according to claim 1, characterized in that, The visualization output step also includes: Based on the splitting results, the skeleton graph is again subjected to connected region solving to obtain the second labeling results of the split interglacial waterways; Based on the second labeling result, the split interglacial waterway is segmented, and the cosine similarity between each segment of the same interglacial waterway is calculated to obtain the location coordinates of the split interglacial waterway.

8. The method for identifying interglacial waterways according to claim 1, characterized in that, The visualization output step also includes: After converting the location coordinates to latitude and longitude coordinates using a file containing location coordinates and latitude and longitude coordinates of the same scale, the location coordinates of the interglacial waterway are displayed on the static map using the Basemap method in Python.

9. A synthetic aperture radar satellite-based interglacial waterway identification system, characterized in that, The interglacial waterway identification system includes: Data preprocessing unit: After calculating the first deformation rate based on synthetic aperture radar satellite data, it generates interglacial waterway image tags based on the first deformation rate; Dataset augmentation unit: augments the first training set in the ice channel image label to obtain a second training set, and then collects a third training set from the second training set using an effective data point sampling method, and processes the first deformation rate to obtain a second deformation rate; Neural network model training unit: The neural network is trained using the third training set and the second deformation rate to obtain the trained neural network model; Visualization output unit: Input the test data into the trained neural network model to obtain a skeleton map, perform connected component solving on the skeleton map, and display the location coordinates of the interglacial waterway on a static map based on the labeling and splitting results; The visualization result output unit includes: After the test data is input into the trained neural network model, the probability output corresponding to the test data is obtained. After binarizing the probability output according to a preset threshold to obtain a binarized image, the binarized image is then thinned to obtain the skeleton map.

Images