Ship navigation method, device and equipment in sea ice area and storage medium

Abstract

The invention relates to the technical field of ship navigation, and provides a ship navigation method, device and equipment for a sea ice area and a storage medium, and the method comprises the steps: obtaining a remote sensing image of the sea ice area where a target ship is to pass, and detecting the feature information of each piece of sea ice in the sea ice area based on the remote sensing image; the feature information comprises contour information and category information, and the category information is used for representing the size of the sea ice; clustering the sea ice according to the feature information to obtain a sea ice cluster; and performing path planning on the target ship based on the sea ice cluster, and navigating the target ship according to the planned target path. According to the method, the sea ice is clustered, the path of the ship is planned according to the clustering result, the calculated amount of path planning is reduced, the efficiency and accuracy of path planning are improved, the ship is navigated according to the planned target path, and the passing safety when the ship passes through the sea ice area can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of ship navigation, and particularly to a ship navigation method, device, equipment and storage medium in sea ice areas. Background Art

[0002] In the field of maritime shipping, there is a high requirement for the real-time nature of ship route planning. In high-latitude regions, natural factors such as monsoon ocean currents cause sea ice to accumulate or disperse, which easily affects the normal navigation of ships. In the extreme environment of large sea ice areas, generally, icebreakers are used to open up channels for ships for route planning, or the method of using unmanned aerial vehicles or manual observation is adopted to observe the situation of the ship's forward channel for navigation decision-making. Among them, for the method of using icebreakers to open up channels, since the situation of sea ice is unknown, the planned travel route is difficult to ensure the safety of ship navigation; while for the method of using unmanned aerial vehicles or manual observation for navigation decision-making, the observation range of the ship's forward channel is limited, and the accuracy of the observed channel situation is difficult to guarantee, which affects the accurate planning of the ship's travel route and is also difficult to ensure the safety of ship navigation. Summary of the Invention

[0003] The present invention provides a ship navigation method, device, equipment and storage medium in sea ice areas to solve the defect that the existing ship path planning methods in sea ice areas are difficult to ensure the safety of ship navigation, and to achieve accurate planning of the travel route of ships in sea ice areas and improve the safety of ship navigation.

[0004] The present invention provides a ship navigation method in sea ice areas, including the following steps: Obtain a remote sensing image of the sea ice area where the target ship is to pass, and detect the characteristic information of each sea ice in the sea ice area based on the remote sensing image; the characteristic information includes contour information and category information, and the category information is used to characterize the size of the sea ice; Cluster the sea ice according to the characteristic information to obtain sea ice clusters; Based on the sea ice clusters, perform path planning for the target ship, and navigate the target ship according to the planned target path.

[0005] According to the ship navigation method in sea ice areas provided by the present invention, the category information at least includes large sea ice, medium sea ice and small sea ice; the step of clustering the sea ice according to the characteristic information to obtain sea ice clusters includes: Select a sea ice set in the sea ice area according to the category information; the sea ice set includes large sea ice and medium sea ice; Calculate the shortest distance between the first sea ice and each second sea ice based on the contour information; the first sea ice is any one of the sea ice set, and the second sea ice is any one of the sea ice set except the first sea ice; Perform density-based clustering processing on the sea ice in the sea ice set according to the shortest distance and the minimum turning radius of the target ship to obtain sea ice clusters; where the shortest distance between adjacent sea ice in the same sea ice cluster is less than the minimum turning radius.

[0006] According to the ship navigation method in the sea ice area provided by the present invention, the path planning for the target ship based on the sea ice clusters includes: Obtain the maximum Feret diameter of the ice cluster area corresponding to each sea ice cluster, and classify the sea ice clusters into medium-sized sea ice clusters and large-sized sea ice clusters according to the maximum Feret diameter of the ice cluster area; Based on the starting position and the ending position of the target ship, perform global optimal obstacle avoidance path planning for the large-sized sea ice clusters and local optimal obstacle avoidance path planning for the medium-sized sea ice clusters to obtain a path array; the path array is used to represent the path distance between key nodes, and the key nodes are selected from the edge pixels of the medium-sized sea ice clusters and the large-sized sea ice clusters during path planning; Based on the path array, use the single-source shortest path algorithm to perform optimal path planning for the target ship.

[0007] According to the ship navigation method in the sea ice area provided by the present invention, after detecting the characteristic information of each sea ice in the sea ice area based on the remote sensing image, it further includes: Obtain the maximum Feret diameter of each sea ice, and convert the maximum Feret diameter of the sea ice into the first evidence information describing the sea ice contour; Convert the category information in the characteristic information into the second evidence information describing the sea ice category; the category information includes the confidence that the sea ice belongs to each preset category; Based on a preset fusion algorithm, fuse the first evidence information and the second evidence information to obtain a fusion feature; Determine the target category to which the sea ice belongs according to the fusion feature, and update the category information based on the target category.

[0008] According to the ship navigation method in the sea ice area provided by the present invention, the detecting the characteristic information of each sea ice in the sea ice area based on the remote sensing image includes: Input the remote sensing image into a pre-trained object detection model to obtain the feature information of each sea ice in the sea ice area output by the object detection model; wherein, the contour information in the feature information is represented by a two-dimensional array, and the two-dimensional array is used to describe the position coordinates of the edge pixel points of the sea ice contour. The object detection model includes a backbone network and a head network. The backbone network is built based on the spatial pyramid mechanism and the convolutional attention mechanism; the backbone network includes a first convolutional layer, a feature extraction layer with residuals, a spatial pyramid layer, and a convolutional attention layer; the head network includes an upsampling layer, a splicing layer, a feature extraction layer without residuals, and a second convolutional layer.

[0009] According to the ship navigation method in the sea ice area provided by the present invention, before detecting the feature information of each sea ice in the sea ice area based on the remote sensing image, it further includes: Obtain a sample data set constructed based on a sea ice data set; the sample data set contains annotation information for sea ice sample images, and the annotation information is used to annotate the sea ice category and the true region box of the sea ice; the sea ice data set is obtained by preprocessing sea ice images, and the preprocessing includes angle adjustment and pixel clipping. Based on the sample data set, perform iterative training on a preset object detection model, and calculate the loss value of the object detection model in each round of iteration. Update the model parameters of the object detection model according to the loss value until the number of iterations reaches a preset maximum number of iterations, or when the loss value converges, obtain a pre-trained object detection model.

[0010] According to the ship navigation method in the sea ice area provided by the present invention, in each round of iteration, calculating the loss value of the object detection model includes: In each round of iteration, obtain the predicted region box of the sea ice in the sea ice sample image output by the object detection model, and calculate the shape intersection over union of the predicted region box and the true region box. Obtain the first horizontal length, the first vertical length of the predicted region box, and the first position coordinates of the center point of the predicted region box, and obtain the second horizontal length, the second vertical length of the true region box, and the second position coordinates of the center point of the true region box. According to the shape intersection over union, the first horizontal length, the first vertical length, the first position coordinates, the second horizontal length, the second vertical length, and the second position coordinates, calculate the regression coefficient of the predicted region box; the regression coefficient is used to represent the regression direction of the predicted region box relative to the true region box. Calculate the loss value of the target detection model based on the intersection over union and the regression coefficient.

[0011] The present invention also provides a ship navigation device for sea ice areas, including the following modules: A sea ice detection module, configured to obtain a remote sensing image of a sea ice area where a target ship is to pass, and detect the characteristic information of each sea ice in the sea ice area based on the remote sensing image; the characteristic information includes contour information and category information, and the category information is used to characterize the size of the sea ice; A sea ice clustering module, configured to cluster the sea ice according to the characteristic information to obtain sea ice clusters; A path planning and navigation module, configured to perform path planning for the target ship based on the sea ice clusters, and navigate the target ship according to the planned target path.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and running on the processor, where when the processor executes the computer program, the ship navigation method for sea ice areas as described in any one of the above is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the ship navigation method for sea ice areas as described in any one of the above is implemented.

[0014] The present invention also provides a computer program product, including a computer program, where when the computer program is executed by a processor, the ship navigation method for sea ice areas as described in any one of the above is implemented.

[0015] The ship navigation method, device, equipment, and storage medium for sea ice areas provided by the present invention obtain a remote sensing image of a sea ice area where a target ship is to pass, detect the characteristic information of sea ice based on the remote sensing image of the sea ice area, cluster the sea ice according to the characteristic information to obtain sea ice clusters, perform path planning for the target ship based on the sea ice clusters, and navigate the target ship according to the planned target path. By clustering the sea ice and performing path planning for the ship according to the clustering result, the computational amount of path planning is reduced, the efficiency and accuracy of path planning are improved, and navigating the ship according to the planned target path is beneficial to improving the passage safety of the ship when passing through the sea ice area. Description of the Drawings

[0016] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the implementation examples or the description of the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is a schematic flowchart of the ship navigation method in the sea ice area provided by the present invention.

[0018] Figure 2 It is a schematic structural diagram of the convolutional attention module provided by the present invention.

[0019] Figure 3 It is a schematic structural diagram of the ship navigation device in the sea ice area provided by the present invention.

[0020] Figure 4 It is a schematic structural diagram of the electronic device provided by the present invention. Detailed implementation manners

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention with reference to the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments in the present invention belong to the scope of protection of the present invention.

[0022] An embodiment of the present invention provides a ship navigation method in a sea ice area, which identifies the sizes and contours of each sea ice in the sea ice area based on remote sensing images of the sea ice area, thereby clustering adjacent sea ice of a specific size that will hinder ship navigation according to the distance to obtain corresponding sea ice clusters, plans the navigation path of the ship in the sea ice area according to the sea ice clusters, and navigates the ship in the sea ice area according to the planned path.

[0023] Specifically, referring to Figure 1 , Figure 1 is a schematic flowchart of the ship navigation method in the sea ice area provided by an embodiment of the present invention. As shown in Figure 1 , the ship navigation method in the sea ice area includes the following steps: Step 100: Obtain a remote sensing image of the sea ice area where the target ship is to pass, and detect the feature information of each sea ice in the sea ice area based on the remote sensing image; the feature information includes contour information and category information, and the category information is used to characterize the size of the sea ice; Step 200: Cluster the sea ice according to the feature information to obtain sea ice clusters; Step 300: Perform path planning for the target ship based on the sea ice clusters, and navigate the target ship according to the planned target path.

[0024] First, obtain a remote sensing image of the sea ice area where the target ship is to pass, and detect the characteristic information of each piece of sea ice in the sea ice area based on the remote sensing image. The characteristic information of the sea ice includes the contour information and category information of the sea ice, where the category information is used to characterize the size of the sea ice. Exemplarily, the category information of the sea ice includes large, medium, and small, which are used to characterize the size of the sea ice, and the contour information of the sea ice is used to characterize the shape of the sea ice.

[0025] Among them, the contour information of the sea ice can be characterized by the position coordinates of the contour points on the edge of the sea ice, or the contour information of the sea ice is characterized by the position coordinates of the key contour points on the edge of the sea ice.

[0026] Cluster the sea ice according to the characteristic information of the sea ice to obtain sea ice clusters. In one embodiment, cluster the sea ice based on density according to the characteristic information of the sea ice to obtain sea ice clusters. Optionally, determine the distance between each pair of sea ice according to the contour information in the characteristic information of the sea ice, and cluster the sea ice according to the distance to obtain sea ice clusters with similar distances. Optionally, determine the distance between each pair of sea ice according to the contour information in the characteristic information of the sea ice, cluster the sea ice according to the distance, and on this basis, perform further clustering according to the category information in the characteristic information to obtain sea ice clusters with similar distances and the same category.

[0027] Perform path planning for the target ship based on the sea ice clusters obtained by clustering, and navigate the target ship according to the planned target path. For sea ice with a relatively short distance, the distance between the sea ice is not sufficient for the ship to pass. However, if path planning is performed based on each piece of sea ice, the computational complexity required for path planning is relatively large, which is not conducive to the real-time requirements of the ship for path planning and navigation. Based on this, by clustering the sea ice and performing path planning for the ship based on the clustered sea ice clusters, the computational complexity during path planning can be reduced, and the path planning efficiency can be improved to meet the real-time requirements of the ship for navigation.

[0028] Optionally, the path planning for the target ship is obtained by planning the target ship based on the positions of the sea ice clusters when the starting position and the destination position of the ship relative to the sea ice area are known. Among them, the planned target path starts from the starting position of the target ship relative to the sea ice area, ends at the destination position of the target ship relative to the sea ice area, and the positions where the sea ice clusters are located are used as non-navigable obstacle areas that need to be avoided. The starting position and the destination position of the target ship relative to the sea ice area are determined based on the established navigation route of the ship.

[0029] In this embodiment, by using the remote sensing image of the sea ice area where the target ship is to pass, detecting the characteristic information of the sea ice based on the remote sensing image of the sea ice area, clustering the sea ice according to the characteristic information to obtain sea ice clusters, planning a path for the target ship based on the sea ice clusters, and navigating the target ship according to the planned target path. By clustering the sea ice and planning the ship's path according to the clustering result, the computational amount of path planning is reduced, the efficiency and accuracy of path planning are improved, and navigating the ship according to the planned target path is beneficial to improving the passing safety of the ship when passing through the sea ice area.

[0030] In one embodiment, the category information in the characteristic information of the sea ice at least includes large sea ice, medium sea ice, and small sea ice. The clustering of the sea ice is implemented based on the sea ice density. Based on this, step 200 further includes: Step 201, selecting a sea ice set in the sea ice area according to the category information; the sea ice set includes large sea ice and medium sea ice; Step 202, calculating the shortest distance between the first sea ice and each second sea ice based on the contour information; the first sea ice is any one in the sea ice set, and the second sea ice is any one in the sea ice set except the first sea ice; Step 203, performing density-based clustering processing on the sea ice in the sea ice set according to the shortest distance and the minimum turning radius of the target ship to obtain sea ice clusters; wherein, the shortest distance between adjacent sea ice in the same sea ice cluster is less than the minimum turning radius.

[0031] According to the category information in the characteristic information of the sea ice, a sea ice set in the sea ice area is selected. The sea ice set includes large sea ice and medium sea ice. Based on the contour information in the characteristic information of the sea ice, the shortest distance between the first sea ice and each second sea ice is calculated. The first sea ice is any one in the sea ice set, and the second sea ice is any one in the sea ice set except the first sea ice.

[0032] Further, according to the shortest distance between any two sea ice in the sea ice set, combined with the minimum turning radius of the target ship, density-based clustering processing is performed on the sea ice in the sea ice set to obtain sea ice clusters, where the shortest distance between adjacent sea ice in the same sea ice cluster is less than the minimum turning radius of the ship.

[0033] For sea ice in sea ice regions, small sea ice does not affect ship passage. Therefore, only large and medium-sized sea ice are regarded as obstacles affecting ship passage. Further, the shortest distance between large and medium-sized sea ice is compared with the minimum turning radius of the ship to determine whether the ship can pass. If the shortest distance is less than the minimum turning radius of the ship, then the ship cannot pass between the sea ice, and the sea ice with the shortest distance less than the minimum turning radius of the ship is regarded as a whole and clustered into the same sea ice cluster.

[0034] In one embodiment, a density clustering algorithm is used to group the sea ice. The similarity between sea ice is measured based on the shortest distance between sea ice, and the shortest distance between sea ice is calculated from the contour information in the feature information of the sea ice. Specifically, for the shortest distance between the first sea ice and the second sea ice, first, based on any edge pixel point in the contour information of the first sea ice, calculate the distance between it and each edge pixel point in the contour information of the second sea ice. In this way, calculate the pairwise distances between each edge pixel point in the contour information of the first sea ice and each edge pixel point in the contour information of the second sea ice, and select the shortest distance among them as the shortest distance between the first sea ice and the second sea ice.

[0035] Further, the shortest distance between sea ice is compared with the minimum turning radius of the ship to further measure the similarity between sea ice and determine whether to cluster the sea ice into the same group. Specifically, if the shortest distance between sea ice is less than or equal to the minimum turning radius of the ship, it means that the shortest distance between sea ice is not sufficient for the ship to pass normally, and the sea ice is clustered into the same group to obtain the corresponding sea ice cluster; if the shortest distance between sea ice is greater than the minimum turning radius of the ship, it means that the shortest distance between sea ice is sufficient for the ship to pass normally, and the sea ice does not belong to the same clustering group.

[0036] The obtained sea ice clusters are classified according to the maximum Feret diameter, so that the sea ice clusters are divided into large sea ice clusters and medium sea ice clusters, and different algorithms are used for path planning for different sizes of sea ice clusters. Based on this, in step 300, path planning for the target ship based on the sea ice cluster further includes: Step 301, obtain the maximum Feret diameter of the ice cluster area corresponding to each of the sea ice clusters, and classify the sea ice clusters into medium sea ice clusters and large sea ice clusters according to the maximum Feret diameter of the ice cluster area; Step 302, based on the starting position and the ending position of the target ship, perform a globally optimal obstacle avoidance path planning for the large sea ice cluster and a locally optimal obstacle avoidance path planning for the medium sea ice cluster to obtain a path array; the path array is used to represent the path distance between key nodes, and the key nodes are selected from the edge pixel points of the medium sea ice cluster and the large sea ice cluster during path planning; Step 303: Based on the path array, use the single-source shortest path algorithm to perform optimal path planning for the target ship.

[0037] The classification of sea ice clusters is determined based on the maximum Feret diameter of the sea ice clusters. When classifying the sea ice clusters, first obtain the maximum Feret diameter of each sea ice cluster, and based on the maximum Feret diameter of each sea ice cluster, classify the sea ice clusters into medium-sized sea ice clusters and large-sized sea ice clusters.

[0038] Optionally, based on the starting position and ending position of the target ship, perform globally optimal obstacle avoidance path planning for large-sized sea ice clusters and locally optimal obstacle avoidance path planning for medium-sized sea ice clusters to obtain a path array, which is used to identify the path distances between key nodes, and the key nodes are selected from the edge pixels of medium-sized sea ice clusters and large-sized sea ice clusters in path planning. Based on the planned path array, use the single-source shortest path algorithm to perform optimal path planning for the target ship.

[0039] In one embodiment, the area corresponding to the sea ice cluster is mostly irregular. If the maximum Feret diameter of the irregular area occupied by all the sea ice in the sea ice cluster is greater than the first number of pixel points, it is a large-sized sea ice cluster. If the maximum Feret diameter of the irregular area occupied by all the sea ice in the sea ice cluster is greater than the second number of pixel points and less than the first number of pixel points, it is a medium-sized sea ice cluster. Among them, the first number of pixel points is greater than the second number of pixel points.

[0040] For the clustered sea ice clusters, perform globally optimal obstacle avoidance path planning for large-sized sea ice clusters. Specifically, for large-sized sea ice clusters, first connect the starting position A and the ending position Z of the sea ice area, then find the tangent points B and C passing through the starting position A and the large-sized sea ice cluster, and calculate the distance AB between the starting position A and the tangent point B, and the distance BC between the tangent points B and C. Finally, take the tangent points B and C as the new starting positions, repeat the above process, and obtain the characteristic graph of path planning composed of nodes and edges.

[0041] For the clustered sea ice clusters, perform locally optimal obstacle avoidance path planning for medium-sized sea ice clusters. Specifically, when the ship encounters a medium-sized sea ice cluster during navigation, it can slowly sail along the periphery of the medium-sized sea ice cluster and use the contour points of the medium-sized sea ice cluster to calculate the paths on both sides for local optimal path planning.

[0042] In this way, the key nodes of each sea ice cluster and the distances between any two adjacent key nodes can be obtained, and then the path array for path planning can be obtained. Use the single-source shortest path algorithm to make a decision on the path array, perform optimal path planning for the target ship, and obtain the target path. Optionally, the single-source shortest path algorithm includes but is not limited to Dijkstra's algorithm, and the optimal path can be the distance-optimal path, that is, the shortest path.

[0043] In one embodiment, after detecting the feature information of sea ice based on the remote sensing image of the sea ice area, the detected feature information is calibrated and updated according to the maximum Feret diameter of the sea ice. Based on this, after step 100, the following steps may further be included: Step 110: Obtain the maximum Feret diameter of each piece of sea ice, and convert the maximum Feret diameter of the sea ice into first evidence information describing the sea ice contour; Step 120: Convert the category information in the feature information into second evidence information describing the sea ice category; the category information includes the confidence levels of the sea ice belonging to each preset category; Step 130: Based on a preset fusion algorithm, fuse the first evidence information and the second evidence information to obtain a fusion feature; Step 140: Determine the target category to which the sea ice belongs according to the fusion feature, and update the category information based on the target category.

[0044] After detecting the feature information of each piece of sea ice, obtain the maximum Feret diameter of each piece of sea ice, and convert the maximum Feret diameter of the sea ice into first evidence information describing the sea ice contour. Optionally, the contour information in the feature information of the sea ice is represented by a two-dimensional array, and the two-dimensional array includes the position coordinates of each edge pixel point of the sea ice. According to the position coordinates of the edge pixel points of the sea ice, calculate the maximum Feret diameter of the sea ice.

[0045] The maximum width of the sea ice is a key factor affecting ship navigation. The maximum Feret diameter is used to represent the maximum width of the sea ice, and based on this, clustering and path planning of the sea ice are performed to ensure the safety of ship navigation. Optionally, for irregular sea ice, its maximum Feret diameter may be the length of the circumscribed rectangle of the sea ice or the diagonal length of the circumscribed rectangle of the sea ice.

[0046] According to the maximum Feret diameter of the sea ice convert the contour information of the sea ice into first evidence information describing the sea ice contour. The conversion of the maximum Feret diameter can be calculated according to the following formula: ; (1) ; (2) ; (3) For the maximum Feret diameter of any piece of sea ice calculate, according to the above formulas 1 to 3, the probabilities that the sea ice belongs to large sea ice , medium sea ice and small sea ice Probabilities of different categories, etc., are used as the first evidence information of the sea ice contour. , and are the evidence conversion functions of large sea ice, medium sea ice, and small sea ice respectively, is the gamma function.

[0047] Furthermore, the category information in the feature information of sea ice is converted into the second evidence information describing the sea ice category, and the category information of sea ice includes the confidence levels of sea ice belonging to each preset category. Optionally, the conversion of the category information can be carried out according to the following formula 4 for the category information of sea ice: , , ; (4) represents the confidence level that the sea ice is small sea ice, represents the confidence level that the sea ice is medium sea ice, represents the confidence level that the sea ice is large sea ice. , and are the second evidence information after conversion of large sea ice, medium sea ice, and small sea ice respectively.

[0048] Based on a preset fusion algorithm, the first evidence information and the second evidence information are fused to obtain a fusion feature, and the target category to which the sea ice belongs is determined according to this fusion feature, and the category information of the sea ice is updated based on this target category. Optionally, according to the category with the highest confidence level in the category information of the sea ice as the initial category of the sea ice, the target category of the sea ice is determined according to the fusion feature. If this target category is different from the initial category of the sea ice, the category information of the sea ice is updated, and the target category of the sea ice is used as the final category.

[0049] In one embodiment, the fusion algorithm for fusing the first evidence information and the second evidence information is specifically as follows: Define ; Let ; ; ; .

[0050] Among them, and are both intermediate feature values, and the category corresponding to the maximum value in the fusion feature ([[]] ) is the target category of the sea ice, that is, the updated category of the sea ice.

[0051] Detecting the characteristic information of sea ice based on the remote sensing images of sea ice areas is implemented based on a pre-trained object detection model. Based on this, in step 100, detecting the characteristic information of each piece of sea ice in the sea ice area based on the remote sensing images of the sea ice area may further include: Step 101, input the remote sensing image into the pre-trained object detection model to obtain the characteristic information of each piece of sea ice in the sea ice area output by the object detection model; wherein, the contour information in the characteristic information is represented by a two-dimensional array, and the two-dimensional array is used to describe the position coordinates of the edge pixel points of the sea ice contour; The object detection model includes a backbone network and a head network. The backbone network is built based on the spatial pyramid mechanism and the convolutional attention mechanism; the backbone network includes a first convolutional layer, a feature extraction layer with residuals, a spatial pyramid layer, and a convolutional attention layer; the head network includes an upsampling layer, a splicing layer, a feature extraction layer without residuals, and a second convolutional layer.

[0052] Input the remote sensing image into the pre-trained object detection model to obtain the characteristic information of each piece of sea ice in the sea ice area output by the object detection model. Among them, the contour information in the characteristic information of the sea ice is represented by a two-dimensional array, and the two-dimensional array is used to describe the position coordinates of the edge pixel points of the sea ice contour.

[0053] Furthermore, the object detection model includes a backbone network and a head network. The backbone network is built based on the spatial pyramid mechanism and the convolutional attention mechanism. The backbone network includes a first convolutional layer, a feature extraction layer with residuals, a spatial pyramid layer, and a convolutional attention layer. The head network includes an upsampling layer, a splicing layer, a feature extraction layer without residuals, and a second convolutional layer.

[0054] Before inputting the remote sensing image into the pre-trained object detection model, it is necessary to train the object detection model. The training process of the object detection model specifically includes: Step 001, obtain a sample data set constructed based on the sea ice data set; the sample data set contains annotation information for the sea ice sample images, and the annotation information is used to annotate the sea ice category and the true region box of the sea ice; the sea ice data set is obtained by preprocessing the sea ice images, and the preprocessing includes angle adjustment and pixel cropping; Step 002, perform iterative training on a preset object detection model based on the sample data set, and calculate the loss value of the object detection model in each round of iteration; Step 003, update the model parameters of the object detection model according to the loss value until the number of iterations reaches the preset maximum number of iterations, or when the loss value converges, obtain the pre-trained object detection model.

[0055] Obtain a sample dataset constructed based on a sea ice dataset, which contains annotation information for sea ice sample images. The annotation information for sea ice sample images is used to annotate the sea ice category and the true region bounding box of the sea ice. The sea ice dataset is obtained by preprocessing sea ice images, and the preprocessing of sea ice images includes, but is not limited to, angle adjustment and pixel cropping. Preprocess the sea ice images to obtain corresponding sea ice sample images, and construct a sample dataset based on the sea ice sample images and their annotation information.

[0056] Furthermore, based on the constructed sample dataset, iteratively train a preset object detection model. In each round of iteration, calculate the loss value of the object detection model, and update the model parameters of the object detection model according to the calculated loss value until the number of iterations reaches the preset maximum number of iterations or the loss value converges, and obtain a pre-trained object detection model.

[0057] Among them, the annotation information for sea ice sample images is obtained by annotating sea ice sample images. Specifically, obtain a sea ice dataset of satellite remote sensing images of the sea ice area, preprocess and annotate the sea ice images therein to obtain a set of visible-light-based sea ice images, that is, the sample dataset.

[0058] In one embodiment, extract the red, green, and blue bands from the sea ice dataset, and perform pixel-level fusion on these three bands to obtain a high-resolution visible-light sea ice image. Then, through preprocessing such as segmentation operations, divide the high-resolution visible-light sea ice image into image blocks of a certain size to obtain sea ice sample images. Then annotate the sea ice image blocks in the sea ice sample images. Exemplarily, generate the circumscribed rectangle of each sea ice image block, and label the sea ice with the longest side of the circumscribed rectangle of the sea ice target within the range of 8 pixels to 32 pixels as small sea ice, the sea ice within 32 pixels to 128 pixels as medium sea ice, and the sea ice within 128 pixels to 256 pixels as large sea ice.

[0059] Furthermore, based on the sample dataset constructed from the annotated sea ice sample images, train the object detection model. In one embodiment, the object detection model is built using the YOLOv8 network model, and the training process of the object detection model mainly includes: Step 1: Configure the environment required for model training; Step 2: Build an improved YOLOv8 network model as the object detection model, such as Figure 2As shown in the figure, in the object detection model, the Convolutional Block Attention Module (CBAM) is added to the 10th layer of the backbone network of YOLOv8. On this basis, the architecture of the head network of YOLOv8 is improved; the Shape-IoU (Shape Intersection over Union) is used as the loss function for bounding box regression, and then, the evidence fusion module is incorporated into the non-maximum suppression module of the object detection model. Among them, the convolutional attention module is constructed based on the convolutional attention mechanism.

[0060] Step 3: Model training and verification: The improved YOLOv8 network model is trained and verified using the sample data set constructed from sea ice images.

[0061] The last module in the backbone network of the original YOLOv8 is the improved Spatial Pyramid Pooling - Fast (SPPF) module, and in this application, CBAM is set after the SPPF module.

[0062] The method for CBAM to extract features is as follows: The first step: The first sea ice feature map output by the 9th layer (i.e., the spatial pooling layer) of the object detection model is used as the input of the convolutional attention module, and the convolutional attention module includes a channel attention module and a spatial attention module.

[0063] The second step: The channel attention module performs a one-dimensional convolution operation on the first sea ice feature map F to obtain a one-dimensional convolution result , and then multiplies the one-dimensional convolution result by the first sea ice feature map to obtain a second sea ice feature map .

[0064] The third step: In the spatial attention module, the second sea ice feature map is used as the input, and a two-dimensional convolution operation is performed on the second sea ice feature map to obtain a two-dimensional convolution result , and then multiplies the two-dimensional convolution result by the second sea ice feature map to obtain a third sea ice feature map .

[0065] There are three branches in the head network of YOLOv8, which are used to detect small targets, medium targets, and large targets respectively. The branches for detecting medium targets and large targets are connected after the convolutional attention module, and the branch for detecting small targets remains unchanged and is still connected after the SPPF module.

[0066] Based on this, the process of the object detection model detecting the feature information of sea ice specifically includes the following steps: Input the first sea ice feature map into the channel attention module to obtain the second sea ice feature map output by the channel attention module; the first sea ice feature map is the output of the spatial pooling layer before the convolutional attention module in the backbone network. Inputting the first sea ice feature map into the channel attention module to obtain the second sea ice feature map output by the channel attention module further includes: Perform one-dimensional convolution on the first sea ice feature map; based on the one-dimensional convolution result and the first sea ice feature map, determine the second sea ice feature map.

[0067] Further, input the second sea ice feature map into the spatial attention module to obtain the third sea ice feature map output by the spatial attention module; Inputting the second sea ice feature map into the spatial attention module to obtain the third sea ice feature map output by the spatial attention module further includes: Perform two-dimensional convolution on the second sea ice feature map; based on the two-dimensional convolution result and the second sea ice feature map, determine the third sea ice feature map.

[0068] Specifically, the explanation of the above is shown in Equation 5. Pass the input first sea ice feature map through two parallel average pooling AvgPool and max pooling MaxPool. At this time, the shape of the feature map is 1×1×C; input the feature map into a multi-layer perceptron module MLP, which is also called an artificial neural network and includes an input layer, a hidden layer, and an output layer. Add the elements of these two outputs one by one, and then pass through a sigmoid (activation function) to obtain the output result of the channel attention module, that is, the second sea ice feature map in this embodiment.

[0069] ; (5) In this embodiment, the channel attention module in the convolutional attention module performs one-dimensional convolution operation on the first sea ice feature map.

[0070] The explanation of the above is shown in Equation 6. Pass the input second sea ice feature map through two parallel average pooling AvgPool and max pooling MaxPool. At this time, the shape of the feature map is H×W×1; then use the concatenation operation to concatenate the two feature maps. Then use a 7×7 convolution to make the feature map become a 1-channel feature map; perform a sigmoid operation on this 1-channel feature map to obtain the output result of the spatial attention module, that is, the third sea ice feature map in this embodiment.

[0071] ; (6) In this embodiment, a two-dimensional convolution operation is performed on the second sea ice feature map through the spatial attention module in the convolutional attention module.

[0072] During the training process of the object detection model, the Shape-IoU (Shape Intersection over Union) is used to construct the loss function of the object detection model. The Shape-IoU is determined based on the predicted bounding box of the object detection model and the ground truth bounding box of the annotated sea ice sample image.

[0073] Specifically, the training method of the improved YOLOv8 model is as follows: The annotated sea ice sample images are divided into a training set and a validation set according to a certain ratio (such as 7:3), the training coefficient is set (such as 300), a certain number of sea ice sample images (such as 16) are input for each training, and after the training is completed, the weight parameters of the model after training are saved.

[0074] As described above, during the training process of the object detection model, the intersection over union of the predicted bounding box and the ground truth bounding box of the sea ice sample image is used as the loss function to calculate the loss value of the object detection model. Based on this, the calculation process of the loss value during the training process of the object detection model mainly includes the following steps: Step 10: Determine the first size data of the predicted bounding box; the first size data includes the first center point and the first size; Step 20: Determine the second size data of the ground truth bounding box; the second size data includes the second center point and the second size; Step 30: Determine the third size data of the bounding rectangle enclosing the predicted bounding box and the ground truth bounding box; the third size data includes the horizontal and vertical lengths and the scale parameter; the bounding rectangle is the largest bounding matrix of the predicted bounding box and the ground truth bounding box; Step 40: Calculate the loss value of the object detection model based on the first size data, the second size data, and the third size data.

[0075] Specifically, the loss function of the object detection model constructed based on Shape-IoU, and the loss value is calculated based on this loss function, as follows: Define the first center point of the predicted bounding box Anchor as (x, y), the horizontal length as and the vertical length as , that is, the first size data in this embodiment.

[0076] Define the second center point of the ground truth bounding box Ground_Truth as ( ), the horizontal length as and the vertical length as , that is, the second size data in this embodiment.

[0077] Define the horizontal length of the largest circumscribed rectangle of the two bounding boxes of the predicted region box and the ground-truth region box as W, the vertical length as H, and the scale parameter scale can be adjusted according to the actual situation, with the value range: scale ≥ 1, to obtain the third size data in this embodiment.

[0078] The loss function constructed based on the Shape Intersection over Union (Shape-IoU) is beneficial to accelerating the convergence speed of the object detection model, thereby improving the detection accuracy of the object detection model.

[0079] Furthermore, the process of calculating the loss value of the object detection model according to the above first size data, second size data, and third size data is shown in the following formulas 7 to 12.

[0080] ; (7) ; (8) ; (9) ; (10) ; (11) ; (12) Among them, represents the predicted region box, represents the ground-truth region box, is the loss function used to calculate the loss value.

[0081] The Shape Intersection over Union (Shape-IoU) is a loss function that comprehensively considers the relative position difference between the center points of the predicted region box and the ground-truth region box, and the relative shape difference between the predicted region box and the ground-truth region box In addition, the Shape Intersection over Union (Shape-IoU) also calculates a regression coefficient based on the ground-truth region box, which is used to refine whether the predicted region box regresses from the short side direction or the long side direction of the ground-truth region box during the regression process. The loss function constructed by the Shape Intersection over Union (Shape-IoU) improves the detection accuracy of the object detection model.

[0082] The first size data includes the first horizontal length and the first vertical length of the predicted region box, the second size data includes the second horizontal length and the second vertical length of the ground-truth region box, the first center point of the predicted region box is characterized by the first position coordinates, and the second center point of the ground-truth region box is characterized by the second position coordinates. Based on this, in step 002, during each round of iteration, calculating the loss value of the object detection model specifically includes: Step 021, in each round of iteration, obtain the predicted region bounding box of the sea ice region in the sea ice sample image output by the target detection model, and calculate the shape intersection over union (IoU) between the predicted region bounding box and the ground-truth region bounding box; Step 022, obtain the first horizontal length, the first vertical length of the predicted region bounding box, and the first position coordinates of the center point of the predicted region bounding box, and obtain the second horizontal length, the second vertical length of the ground-truth region bounding box, and the second position coordinates of the center point of the ground-truth region bounding box; Step 023, calculate the regression coefficient of the predicted region bounding box according to the shape IoU, the first horizontal length, the first vertical length, the first position coordinates, the second horizontal length, the second vertical length, and the second position coordinates; the regression coefficient is used to represent the regression direction of the predicted region bounding box relative to the ground-truth region bounding box; Step 024, calculate the loss value of the target detection model based on the shape IoU and the regression coefficient.

[0083] In each round of iteration during model training, obtain the predicted region bounding box of the sea ice in the sea ice sample image output by the target detection model, and calculate the shape IoU between the predicted region bounding box and the ground-truth region bounding box annotated in the sea ice sample image.

[0084] Further, obtain the first size data of the predicted region bounding box and the first position coordinates of its center point, and obtain the second size data of the ground-truth region bounding box and the second position coordinates of its center point. Wherein, the first size data includes the first horizontal length and the first vertical length of the predicted region bounding box, and the second size data includes the second horizontal length and the second vertical length of the ground-truth region bounding box.

[0085] Calculate the regression coefficient of the predicted region bounding box according to the calculated shape IoU, the first size data, the second size data, the first position coordinates, and the second position coordinates. The regression coefficient is used to represent the regression direction of the predicted region bounding box relative to the ground-truth region bounding box. Calculate the loss value of the target detection model based on the shape IoU and the regression coefficient of the target detection model.

[0086] In one embodiment, during the training process of the target detection model, non-maximum suppression is used to update the model parameters, which specifically includes the following steps: Step 50, input the longest side of the predicted region bounding box into the non-maximum suppression module, and determine the first uncertain evidence and the second uncertain evidence based on the longest side of the predicted region bounding box; the first uncertain evidence is the evidence used to describe the uncertainty of the bounding box of the sea ice; the second uncertain evidence is the evidence used to describe the class uncertainty of the sea ice; Step 60: Fuse and reason the first uncertain evidence and the second uncertain evidence to obtain a new predicted category of the predicted region box; Step 70: Update the model parameters of the object detection model according to the new predicted category of the predicted region box.

[0087] Specifically, this application incorporates evidence fusion into the non-maximum suppression of YOLOv8, adjusts the predicted sea ice size category information, improves the non-maximum suppression module of the original YOLOv8, and adds an evidence fusion module to the non-maximum suppression module. The non-maximum suppression module of the original YOLOv8 is used to eliminate redundant detection boxes in the prediction results and thus obtain the final prediction results.

[0088] Furthermore, on this basis, this application extracts the prediction results obtained by the original YOLOv8 in the non-maximum suppression module, including the position of the predicted region box and the predicted category, and processes the position and category of the predicted region box. For the position information of the predicted region box, determine the longest side of the predicted region box l , and convert it into two types of evidence, a and b, describing uncertainty.

[0089] Based on the improved DSmT (Dezert-Smarandache Theory, an algorithm for processing information fusion) fusion reasoning algorithm, fuse the above two types of evidence to obtain a new predicted category of sea ice size. By incorporating an evidence fusion module into the non-maximum suppression module, adjust the predicted category of sea ice size, and improve the detection accuracy of sea ice size in the model training stage and the model application stage, thereby further improving the path planning and navigation accuracy of ships in sea ice areas.

[0090] By adding a convolutional attention module to the object detection model and correspondingly improving the structure of the head network of the object detection model, guide the object detection model to evenly extract target features of different size types; in terms of model loss, introduce a boundary regression loss function of shape intersection over union to reduce the error of the object detection model in detecting the shape and position of the target, thereby improving the accuracy of ship navigation in sea ice areas.

[0091] Using a pre-trained object detection model, after obtaining a remote sensing image of the sea ice area where the ship is to pass, the contour information and size categories of the sea ice and other feature information are detected from the remote sensing image. Among them, one or more remote sensing images of the sea ice area are obtained. When there are multiple ones, the remote sensing images of multiple sea ice areas form a sea ice image set. The remote sensing images in the sea ice image set are visible light-based sea ice images obtained from satellite remote sensing data of the sea ice in a specified area. The sea ice images contain regional positioning information. For example, the four vertices of a rectangular sea ice image correspond to four longitude and latitude information. Through the distance conversion algorithm between the image and the real world, the real distance between a certain sea ice in the sea ice image and the image vertices can be obtained, and further, based on the determination of the longitude and latitude of the image vertices, the coordinates of the sea ice in the real world in the image can be calculated.

[0092] The sea ice size can be divided into small, medium, and large. The medium and large sizes are regarded as the sea ice sizes that affect ship navigation. Therefore, by calculating the positions of the medium-sized sea ice and the large-sized sea ice in the sea ice image, the navigation path and navigation information of the ship are planned. The navigation information includes the position information of the medium-sized sea ice and the position information of the large-sized sea ice. By obtaining the sea ice positions, a safe driving path is planned for the ship.

[0093] In this embodiment, by obtaining the remote sensing image of the sea ice and performing preprocessing such as annotation on it, a sample data set is constructed, the object detection model is trained, and the pre-trained object detection model is used to detect the remote sensing image of the sea ice area where the ship is to pass, obtaining the detection results of the sea ice, including the sea ice contour and the category of the sea ice size, and based on this, path planning and navigation are performed on the ship, improving the navigation efficiency of the ship in the sea ice area and ensuring the navigation safety of the ship.

[0094] Furthermore, when performing path planning for the ship, a density clustering algorithm is adopted to cluster the medium and large-sized sea ice that affects ship navigation according to the distance that the ship can pass, so as to reduce the calculation amount during path planning, improve the path planning efficiency, and meet the real-time requirements of ship navigation. And for the path planning of the ship, a combination of global optimal and local optimal is adopted to ensure that the ship can cross the sea ice area at the shortest distance.

[0095] The ship navigation device in the sea ice area provided by the present invention will be described below. The ship navigation device in the sea ice area described below can be mutually referred to corresponding to the ship navigation method in the sea ice area described above.

[0096] Referring to Figure 3 , the ship navigation device in the sea ice area provided by the embodiment of the present invention includes: The sea ice detection module 10 is used to obtain a remote sensing image of the sea ice area where the target ship is to pass, and detect the characteristic information of each piece of sea ice in the sea ice area based on the remote sensing image; the characteristic information includes contour information and category information, and the category information is used to characterize the size of the sea ice; The sea ice clustering module 20 is used to cluster the sea ice according to the characteristic information to obtain sea ice clusters; The path planning and navigation module 30 is used to perform path planning on the target ship based on the sea ice clusters, and navigate the target ship according to the planned target path.

[0097] In one embodiment, the category information at least includes large sea ice, medium sea ice, and small sea ice; the sea ice clustering module 20 is further used for: Select a set of sea ice in the sea ice area according to the category information; the set of sea ice includes large sea ice and medium sea ice; Calculate the shortest distance between a first piece of sea ice and each second piece of sea ice based on the contour information; the first piece of sea ice is any one in the set of sea ice, and the second piece of sea ice is any one in the set of sea ice other than the first piece of sea ice; Perform density-based clustering processing on the sea ice in the set of sea ice according to the shortest distance and the minimum turning radius of the target ship to obtain sea ice clusters; wherein, the shortest distance between adjacent sea ice in the same sea ice cluster is less than the minimum turning radius.

[0098] In one embodiment, the path planning and navigation module 30 is further used for: Obtain the maximum Feret diameter of the ice cluster area corresponding to each sea ice cluster, and classify the sea ice clusters into medium sea ice clusters and large sea ice clusters according to the maximum Feret diameter of the ice cluster area; Perform global optimal obstacle avoidance path planning on the large sea ice clusters and local optimal obstacle avoidance path planning on the medium sea ice clusters based on the starting position and the ending position of the target ship to obtain a path array; the path array is used to represent the path distance between key nodes, and the key nodes are selected from the edge pixels of the medium sea ice clusters and the large sea ice clusters in the path planning; Perform optimal path planning on the target ship by using the single-source shortest path algorithm based on the path array.

[0099] In one embodiment, the sea ice detection module 10 is further used for: Obtain the maximum Feret diameter of each piece of sea ice, and convert the maximum Feret diameter of the sea ice into first evidence information describing the sea ice contour; Convert the category information in the feature information into second evidence information describing sea ice categories; the category information includes the confidence levels of the sea ice belonging to each preset category. Based on a preset fusion algorithm, fuse the first evidence information and the second evidence information to obtain a fused feature. Determine the target category to which the sea ice belongs according to the fused feature, and update the category information based on the target category.

[0100] In one embodiment, the sea ice detection module 10 is further configured to: Input the remote sensing image into a pre-trained target detection model to obtain the feature information of each piece of sea ice in the sea ice area output by the target detection model; wherein, the contour information in the feature information is represented by a two-dimensional array, and the two-dimensional array is used to describe the position coordinates of the edge pixel points of the sea ice contour. The target detection model includes a backbone network and a head network. The backbone network is built based on a spatial pyramid mechanism and a convolutional attention mechanism; the backbone network includes a first convolutional layer, a feature extraction layer with residuals, a spatial pyramid layer, and a convolutional attention layer; the head network includes an upsampling layer, a splicing layer, a feature extraction layer without residuals, and a second convolutional layer.

[0101] In one embodiment, the ship navigation device in the sea ice area further includes a model training module, which is configured to: Obtain a sample data set constructed based on a sea ice data set; the sample data set includes annotation information for sea ice sample images, and the annotation information is used to annotate sea ice categories and real region boxes of sea ice; the sea ice data set is obtained by preprocessing sea ice images, and the preprocessing includes angle adjustment and pixel cropping. Iteratively train a preset target detection model based on the sample data set, and calculate the loss value of the target detection model in each round of iteration. Update the model parameters of the target detection model according to the loss value until the number of iterations reaches a preset maximum number of iterations, or when the loss value converges, obtain a pre-trained target detection model.

[0102] In one embodiment, the model training module is further configured to: In each round of iteration, obtain the predicted region box of the sea ice in the sea ice sample image output by the target detection model, and calculate the shape intersection over union of the predicted region box and the real region box. Obtain the first horizontal length, the first vertical length of the predicted region box, and the first position coordinates of the center point of the predicted region box, and obtain the second horizontal length, the second vertical length of the ground-truth region box, and the second position coordinates of the center point of the ground-truth region box; Calculate the regression coefficient of the predicted region box according to the intersection over union (IoU), the first horizontal length, the first vertical length, the first position coordinates, the second horizontal length, the second vertical length, and the second position coordinates; the regression coefficient is used to represent the regression direction of the predicted region box relative to the ground-truth region box; Calculate the loss value of the object detection model based on the intersection over union (IoU) and the regression coefficient.

[0103] Figure 4 An example of the physical structure diagram of an electronic device is shown as Figure 4 shown. The electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440. Among them, the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 may call the logical instructions in the memory 430 to execute the ship navigation method in the sea ice area. The method includes: Obtain a remote sensing image of the sea ice area where the target ship is to pass, and detect the feature information of each piece of sea ice in the sea ice area based on the remote sensing image; the feature information includes contour information and category information, and the category information is used to characterize the size of the sea ice; Cluster the sea ice according to the feature information to obtain sea ice clusters; Perform path planning for the target ship based on the sea ice clusters, and navigate the target ship according to the planned target path.

[0104] In addition, when the logical instructions in the above-mentioned memory 430 can be implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0105] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the ship navigation method in the sea ice area provided by the above-mentioned various methods. The method includes: Obtain a remote sensing image of the sea ice area where the target ship is to pass, and detect the feature information of each piece of sea ice in the sea ice area based on the remote sensing image; the feature information includes contour information and category information, and the category information is used to characterize the size of the sea ice; Cluster the sea ice according to the feature information to obtain sea ice clusters; Perform path planning for the target ship based on the sea ice clusters, and navigate the target ship according to the planned target path.

[0106] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the ship navigation method in the sea ice area provided by the above-mentioned various methods. The method includes: Obtain a remote sensing image of the sea ice area where the target ship is to pass, and detect the feature information of each piece of sea ice in the sea ice area based on the remote sensing image; the feature information includes contour information and category information, and the category information is used to characterize the size of the sea ice; Cluster the sea ice according to the feature information to obtain sea ice clusters; Perform path planning for the target ship based on the sea ice clusters, and navigate the target ship according to the planned target path.

[0107] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative effort.

[0108] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A ship navigation method in sea ice area, characterized in that, Including: Obtaining a remote sensing image of the sea ice area where the target ship is to pass, and detecting the characteristic information of each piece of sea ice in the sea ice area based on the remote sensing image; The characteristic information includes contour information and category information, and the category information is used to characterize the size of the sea ice; Clustering the sea ice according to the characteristic information to obtain sea ice clusters; Performing path planning on the target ship based on the sea ice clusters, and navigating the target ship according to the planned target path.

2. The ship navigation method in the sea ice area according to claim 1, characterized in that, The category information at least includes large sea ice, medium sea ice and small sea ice; the clustering the sea ice according to the characteristic information to obtain sea ice clusters includes: Selecting a set of sea ice in the sea ice area according to the category information; the set of sea ice includes large sea ice and medium sea ice; Calculating the shortest distance between a first piece of sea ice and each second piece of sea ice based on the contour information; the first piece of sea ice is any one in the set of sea ice, and the second piece of sea ice is any one in the set of sea ice other than the first piece of sea ice; Performing density-based clustering processing on the sea ice in the set of sea ice according to the shortest distance and the minimum turning radius of the target ship to obtain sea ice clusters; wherein, the shortest distance between adjacent sea ice in the same sea ice cluster is less than the minimum turning radius.

3. The method for ship navigation in sea ice area according to claim 1, wherein, The performing path planning on the target ship based on the sea ice clusters includes: Obtaining the maximum Feret diameter of the ice cluster area corresponding to each sea ice cluster, and classifying the sea ice clusters into medium sea ice clusters and large sea ice clusters according to the maximum Feret diameter of the ice cluster area; Performing a globally optimal obstacle avoidance path planning on the large sea ice clusters and a locally optimal obstacle avoidance path planning on the medium sea ice clusters based on the starting position and the ending position of the target ship to obtain a path array; the path array is used to represent the path distance between key nodes, and the key nodes are selected from the edge pixel points of the medium sea ice clusters and the large sea ice clusters in the path planning; Performing an optimal path planning on the target ship by using a single-source shortest path algorithm based on the path array.

4. The ship navigation method in the sea ice area according to claim 1, wherein, After detecting the characteristic information of each piece of sea ice in the sea ice area based on the remote sensing image, it further includes: Obtaining the maximum Feret diameter of each piece of sea ice, and converting the maximum Feret diameter of the sea ice into first evidence information for describing the sea ice contour; Converting the category information in the characteristic information into second evidence information for describing the sea ice category; the category information contains the confidence that the sea ice belongs to each preset category; Fusing the first evidence information and the second evidence information based on a preset fusion algorithm to obtain a fusion feature; Determining the target category to which the sea ice belongs according to the fusion feature, and updating the category information based on the target category.

5. The ship navigation method in the sea ice area according to claim 1, characterized in that, The detecting the characteristic information of each piece of sea ice in the sea ice area based on the remote sensing image includes: Input the remote sensing image into a pre-trained object detection model to obtain the feature information of each sea ice in the sea ice area output by the object detection model; wherein, the contour information in the feature information is represented by a two-dimensional array, and the two-dimensional array is used to describe the position coordinates of the edge pixel points of the sea ice contour. The object detection model includes a backbone network and a head network. The backbone network is built based on the spatial pyramid mechanism and the convolutional attention mechanism; the backbone network includes a first convolutional layer, a feature extraction layer with residuals, a spatial pyramid layer, and a convolutional attention layer; the head network includes an upsampling layer, a splicing layer, a feature extraction layer without residuals, and a second convolutional layer.

6. The method for ship navigation in sea ice areas according to claim 5, characterized in that Before detecting the feature information of each sea ice in the sea ice area based on the remote sensing image, it further includes: Obtain a sample data set constructed based on a sea ice data set; the sample data set contains annotation information for sea ice sample images, and the annotation information is used to annotate the sea ice category and the true region box of the sea ice; the sea ice data set is obtained by preprocessing sea ice images, and the preprocessing includes angle adjustment and pixel cropping. Based on the sample data set, perform iterative training on a preset object detection model, and calculate the loss value of the object detection model in each round of iteration. Update the model parameters of the object detection model according to the loss value until the number of iterations reaches a preset maximum number of iterations, or when the loss value converges, obtain a pre-trained object detection model.

7. The method for ship navigation in sea ice areas according to claim 6, wherein, In each round of iteration, calculating the loss value of the object detection model includes: In each round of iteration, obtain the predicted region box of the sea ice in the sea ice sample image output by the object detection model, and calculate the shape intersection over union of the predicted region box and the true region box. Obtain the first horizontal length, the first vertical length of the predicted region box, and the first position coordinates of the center point of the predicted region box, and obtain the second horizontal length, the second vertical length of the true region box, and the second position coordinates of the center point of the true region box. According to the shape intersection over union, the first horizontal length, the first vertical length, the first position coordinates, the second horizontal length, the second vertical length, and the second position coordinates, calculate the regression coefficient of the predicted region box; the regression coefficient is used to represent the regression direction of the predicted region box relative to the true region box. Based on the shape intersection over union and the regression coefficient, calculate the loss value of the object detection model.

8. A ship navigation device for sea ice areas, characterized in that, It includes: A sea ice detection module, configured to obtain a remote sensing image of a sea ice area where a target ship is to pass, and detect the feature information of each sea ice in the sea ice area based on the remote sensing image. The feature information includes contour information and category information, and the category information is used to characterize the size of the sea ice. A sea ice clustering module, configured to cluster the sea ice according to the feature information to obtain sea ice clusters. A path planning and navigation module, configured to perform path planning on the target ship based on the sea ice clusters, and navigate the target ship according to the planned target path.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the ship navigation method in the sea ice area as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the ship navigation method in the sea ice area as described in any one of claims 1 to 7.

Images