Method for identifying port moored ships and their class based on spaceborne sar images

Abstract

This invention discloses a method for identifying moored vessels and their categories in a port based on spaceborne SAR images. The method includes: constructing a vessel dataset composed of port spaceborne SAR images of multiple defined functional zones; training a pre-constructed first network model based on the dataset and labels for each functional zone to obtain a trained first network model; labeling moored and sailing vessels in the dataset, and training a pre-constructed second network model based on the labeled dataset to obtain a trained second network model; slicing the port target image to be identified to obtain multiple slice images; performing functional zone classification and moored vessel detection on each slice image based on the first and second network models respectively, and determining the category of valid moored vessels in each slice image based on the comprehensive detection results. This application can accurately identify valid moored vessels in a port and determine their categories.

Description

Technical Field

[0001] This invention relates to the field of target recognition technology, and in particular to a method for identifying port vessels and their categories based on spaceborne SAR images. Background Technology

[0002] Ports, located at the junction of land and sea, are trade hubs where the density of vessels is much higher than at sea. Therefore, mastering port target information is crucial for port resource scheduling, shipping safety management, and marine economic analysis, contributing to improved port operational efficiency and the scientific basis of management decisions. Spaceborne synthetic aperture radar (SAR), with its all-weather, 24 / 7 monitoring capabilities, has become an important technical means for dynamic monitoring of port functional areas. However, spaceborne SAR images suffer from complex imaging mechanisms and limited resolution. Vessels moored within the port are susceptible to strong land scattering interference in images, and different types of vessels exhibit overlapping and similar geometric structures and textures in SAR images. Furthermore, the dense distribution of vessels in the port environment, susceptibility to obstruction by shore-based equipment, and variations in imaging angle and sea surface conditions further complicate the identification of moored vessel types.

[0003] Traditional methods for ship target detection and identification, such as SAR ship target recognition algorithms based on traditional feature descriptions like HOG, SIFT, and LBP, and classification algorithms that rely on manual feature extraction, struggle to effectively represent the complex and varied characteristics of ships in SAR images and exhibit poor robustness in complex port scenarios. Methods based on deep learning intelligent network models, on the other hand, rely on the electromagnetic scattering characteristics of targets for SAR imaging. Structures with near-dihedral angles generate stronger scattered echo energy than smooth planar structures. Existing intelligent deep learning methods simply search for target features at the pixel level. Therefore, certain cluttered areas in SAR images, or areas with similar structures and strong scattering, are easily misdetected as targets. Furthermore, docked targets in ports are often obscured by shore-based equipment such as cranes and pylons, leading to misclassification.

[0004] Therefore, in complex port environments, existing methods cannot accurately identify ships in a moored state. There is an urgent need for a method based on spaceborne SAR images to identify moored ships and their categories in ports to solve the above problems. Summary of the Invention

[0005] This invention provides a method for identifying moored vessels and their categories in ports based on spaceborne SAR images. This method can accurately identify validly moored vessels in a port and determine their categories. The technical solution is as follows:

[0006] On the one hand, a method for identifying port vessels and their categories based on spaceborne SAR imagery is provided, the method comprising:

[0007] Construct a ship dataset consisting of port spaceborne SAR images with multiple defined functional zones;

[0008] The pre-built first network model is trained based on the dataset and the labels of each functional area to obtain the trained first network model;

[0009] The moored and sailing vessels in the dataset are labeled, and the pre-built second network model is trained based on the labeled dataset to obtain the trained second network model;

[0010] The port target image to be identified is sliced ​​to obtain multiple slice images;

[0011] Based on the first network model and the second network model, functional area classification and moored vessel detection are performed on each slice image, and the category of valid moored vessels in each slice image is determined based on the comprehensive detection results.

[0012] On the other hand, an apparatus for identifying port-berthed vessels and their categories based on spaceborne SAR imagery is provided, the apparatus comprising:

[0013] The building unit is used to construct a ship dataset consisting of port spaceborne SAR images with multiple defined functional areas;

[0014] The first training unit is used to train the pre-built first network model based on the dataset and the labels of each functional area to obtain the trained first network model.

[0015] The second training unit is used to label the moored and sailing vessels in the dataset, and to train the pre-built second network model based on the labeled dataset to obtain the trained second network model.

[0016] The slicing unit slices the port target image to be identified, resulting in multiple slice images;

[0017] The detection unit performs functional area classification and moored vessel detection on each slice image based on the first network model and the second network model, and determines the category of valid moored vessels in each slice image based on the comprehensive detection results.

[0018] On the other hand, a computer device is provided, the computer device including a memory and a processor, the memory for storing computer programs, and the processor for executing the computer programs stored in the memory to implement the steps of the method for identifying port moored vessels and their categories based on spaceborne SAR images as described above.

[0019] On the other hand, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when executed by a processor, the computer program implements the steps of the method for identifying port moored vessels and their categories based on spaceborne SAR images described above.

[0020] On the other hand, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method for identifying port-berthed vessels and their categories based on spaceborne SAR images described above.

[0021] This invention provides a method for identifying moored vessels and their categories in a port based on spaceborne SAR images. First, two network models are trained using deep learning methods, enabling them to perform functional area identification and moored vessel detection, respectively. Then, for a real image to be identified, based on the functional area categories and vessel detection results output by the two models, the category of moored vessels in that area is inferred using the functional area-to-moored vessel category correspondence. This method can infer the category of moored vessels in a port, especially obscured vessels, even in low-quality SAR images, thus providing more detailed information about the targets. This application improves the utilization rate of spaceborne SAR images while mining scene information. By using scene analysis to assist in target category inference, deeper image information mining is achieved, significantly improving the accuracy of vessel identification and classification. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of a method for identifying port vessels and their categories based on spaceborne SAR images, according to an embodiment of the present invention.

[0024] Figure 2 This is a structural diagram of a device for identifying port vessels and their categories based on spaceborne SAR images, according to an embodiment of the present invention.

[0025] Figure 3 This is a hardware architecture diagram of a computer device provided in an embodiment of the present invention;

[0026] Figures 4-9 These are schematic diagrams of a ship dataset composed of port spaceborne SAR images provided in an embodiment of the present invention.

[0027] Figure 10This is a schematic diagram of a first network model provided in an embodiment of the present invention;

[0028] Figures 11-13 These are schematic diagrams of samples after ship labeling of datasets for bulk cargo area, container area and oil tank area provided in an embodiment of the present invention;

[0029] Figure 14 This is a schematic diagram illustrating the result of ship identification using the method described in this application, according to an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0031] As mentioned earlier, existing methods cannot accurately detect moored vessels and their types in ports, mainly because the port environment is complex and can easily interfere with target detection.

[0032] However, the inventors discovered that while ships in ports are susceptible to significant interference, the spatiotemporal changes within the port environment are relatively slow, and the berthing positions of ships are closely related to their functional attributes. Therefore, the inventors proposed identifying the types of berthed ships by starting with the ship's location within the port's functional areas and combining this with the correspondence between port functional zones and ship types.

[0033] The specific implementation of the method in this application is described in detail below.

[0034] Please refer to Figure 1 This invention provides a method for identifying port vessels and their categories based on spaceborne SAR images. The method includes:

[0035] Step 100: Construct a ship dataset consisting of port spaceborne SAR images with multiple defined functional areas;

[0036] Step 102: Train the pre-built first network model based on the dataset and the labels of each functional area to obtain the trained first network model;

[0037] Step 104: Label the moored and sailing vessels in the dataset, and train the pre-built second network model based on the labeled dataset to obtain the trained second network model;

[0038] Step 106: Slice the port target image to be identified to obtain multiple slice images;

[0039] Step 108: Based on the first network model and the second network model, perform functional area classification and moored vessel detection on each slice image, and determine the category of valid moored vessels in each slice image based on the comprehensive detection results.

[0040] In this embodiment, two network models are first trained using deep learning methods to enable them to identify functional areas and detect moored vessels, respectively. Then, for a real image to be identified, based on the functional area categories and vessel detection results output by the two models, the category of moored vessels within that area is inferred using the functional area-to-moored vessel category correspondence. This allows for category inference of moored vessels in ports, especially obscured vessels, even in low-quality SAR images, thus providing more detailed target information. This application improves the utilization rate of spaceborne SAR images while mining scene information. By using scene analysis to assist target category inference, deeper image information mining is achieved, significantly improving the accuracy of vessel identification and classification.

[0041] The following description Figure 1 The execution method of each step is shown.

[0042] First, regarding step 100:

[0043] We collected spaceborne SAR port images, including full-width images from the GF3 spaceborne SAR and partial port scene slices from the open-source dataset FAIR-CSAR. These slices were uniformly segmented into slices of a set size, such as 1024×1024. We then manually selected slices to include only those within a specified functional area and constructed a spaceborne SAR port vessel dataset with a certain degree of generalization through data augmentation methods such as rotation, flipping, and noise addition.

[0044] In addition, the functional areas include a bulk cargo area, a container area, and an oil tank area, each with its own unique characteristics.

[0045] (1) The bulk cargo area generally consists of a conveyor belt for cargo transportation and a storage area for sand, gravel, timber, etc. Docked bulk carriers need to unload their cargo via the conveyor belt and transfer it to the sand and gravel storage area, or load cargo from the storage area. The conveyor belt is not obvious in actual identification, but it often extends to the sea surface; the storage area has different scattering characteristics depending on the type of cargo. Most of the cargo, represented by sand and gravel, is piled up in a mountain shape, and there are also open spaces in the area where no cargo has been piled up yet.

[0046] (2) Container areas generally consist of container handling equipment such as cranes and container stacking areas. Container ships that dock need to load or unload containers through shore loading and unloading equipment. Shore loading and unloading equipment extends its supports over the sea, which has strong scattering and can obstruct the identification of ships; cargo is uniformly loaded in containers, and containers are densely distributed in the stacking area, but there are also situations where the containers are relatively empty.

[0047] (3) Tank farms generally consist of pipelines for transporting oil resources and storage tanks. Docked tankers need to transfer oil to tanks via pipelines or load oil from tanks. Pipelines are not obvious in actual identification, but they often extend to the sea surface; tanks are essential for oil storage, and their edges have obvious scattering characteristics in SAR images.

[0048] Of course, users can also choose other regions as needed; this application does not impose specific restrictions.

[0049] Regarding step 102:

[0050] like Figures 4-9 As shown, each functional area in the dataset has its own unique characteristics. The figure shows SAR slices and optical slices of the port scene functional areas. By comparing and analyzing the two, we can obtain the scattering characteristics of each type of port functional area of ​​spaceborne SAR, thereby realizing functional area classification.

[0051] Here, each sample image is labeled according to the characteristics of the functional area, and then the first network model is trained to enable it to identify the functional area to which the image belongs.

[0052] In addition, the first network model can be a ResNet101 network, with network parameters as follows: Figure 10 As shown, ResNet101 is a deep residual neural network model. Its basic architecture consists of multiple stacked residual blocks, including convolutional layers, batch normalization layers, ReLU activation functions, and pooling layers. By introducing residual connections, it addresses the gradient vanishing and degradation problems that occur as the number of neural network layers increases. The advantage of ResNet101 lies in its ability to construct extremely deep network structures, thereby learning richer and more abstract features. It exhibits excellent performance in various computer vision tasks such as image recognition and object detection, demonstrating high accuracy and generalization ability. For port scene functional area classification tasks, the large scene area necessitates a ResNet101 model with more layers and parameters to capture more complex image features, resulting in superior performance.

[0053] Regarding step 104:

[0054] like Figures 11-13As shown, Labelimg is used to label ships on both the training and test sets. It uses rotated bounding boxes to label moored and sailing ships, without needing to specify the ship's category. Labelimg generates XML files from the drawn bounding boxes, then converts these XML files into YOLO-formatted text labels (txt), completing the dataset required for ship detection.

[0055] Furthermore, the second network model can employ the YOLOv11 network. As the latest version of the YOLO series, the YOLOv11 model architecture deeply integrates multi-scale feature pyramids and dynamic attention mechanisms. It adopts an improved C2f structure to enhance the feature extraction capability of the backbone network, and combines a BiFPN-optimized neck network to achieve efficient cross-layer feature fusion. The head detection module introduces an adaptive anchor box mechanism and a lightweight decoupling head design, significantly improving the model's inference speed and parameter balance. The detection capabilities for small targets and occluded objects are significantly enhanced, and through optimized non-maximum suppression algorithms and multi-scale training strategies, false negatives and false positives can be effectively reduced.

[0056] For step 106, the following are included:

[0057] The target image is preprocessed to obtain an image with a preset format and preset size;

[0058] The preprocessed image is sliced ​​using a sliding window of a set size to obtain multiple image slices; the sliding window of the set size is equal to the input size requirements of the first network model and the second network model.

[0059] For example, for a target image, the high-bit 32-bit, 24-bit, and 16-bit TIFF images are reduced to 8-bit PNG images, and the entire image is sliced ​​by sliding a 1024×1024 window to obtain a slice image group.

[0060] Finally, regarding step 108, the following are included:

[0061] Iterate through each slice image sequentially, and for each slice image encountered, perform the following:

[0062] S1, input the slice image into the first network model, output the functional area category of the slice image, and determine whether the functional area category belongs to one of the set functional areas; if yes, then execute S2; if no, then execute S3.

[0063] S2, input the sliced ​​image into the second network model, output the ship detection result; and determine whether there is at least one valid moored ship in the sliced ​​image based on the ship detection result. If yes, proceed to S4; otherwise, proceed to S3.

[0064] S3, remove the slice image, traverse the next slice image, and return to execute S1;

[0065] S4. Based on the correspondence between the functional area categories and the vessel categories in the slice image, determine the category of each validly moored vessel in the slice image.

[0066] In this step, the second network model's rotating bounding boxes traverse windows of various sizes and multiple directions within the image, and merge all similar boxes using non-maximum suppression (NMS) to finally obtain a simplified ship detection result, including the ship's state (cls). s (Including both anchored and underway states), the ship's four-point coordinates loc = {x1, y1, x2, y2, x3, y3, x4, y4} and confidence level conf s ∈(0,1).

[0067] In some implementations, determining whether at least one valid moored vessel exists in the sliced ​​image based on the vessel detection results includes:

[0068] For each detected vessel, execute:

[0069] If the vessel is in a state of being at sea, no action is taken, and the next vessel is detected;

[0070] If the vessel is in a moored state, calculate the aspect ratio of its rotating frame based on its four-point coordinates; determine whether the calculated aspect ratio is greater than the standard aspect ratio threshold for vessels; if not, remove the vessel and check the next vessel; if yes, determine whether the confidence level of the vessel is greater than the confidence level threshold; if yes, identify the vessel as a valid moored vessel; if not, remove the vessel and check the next vessel.

[0071] The detection results for validly moored vessels in the slice image are obtained after each vessel in the slice image has been detected.

[0072] In some implementations, the length-to-width ratio of the vessel is calculated as follows:

[0073] Calculate the lengths of the two adjacent sides of the rotating frame containing the ship based on the ship's four-point coordinates;

[0074] The ratio of the length of the longest side to the length of the shortest side is taken as the length-to-width ratio of the ship.

[0075] The lengths of two adjacent sides are calculated using the following formulas:

[0076]

[0077] The length-to-width ratio r of a ship is calculated using the following formula:

[0078]

[0079] In the formula, l1 and l2 are two adjacent sides of the ship rotation frame, respectively; x1, y1, x2, y2, x3, y3, x4, y4 are the coordinates of the four points of the ship rotation frame, respectively.

[0080] In addition, the aspect ratio threshold θ of standard ships dis The value is determined based on the actual situation, such as 5 to 8, etc. For example, when r ≤ θ dis If the current moored vessel's length-to-width ratio is incorrect or its shape is not that of a vessel, then the information for that moored vessel is deleted, and the next vessel to be moored is selected.

[0081] In addition, the reliability threshold θ conf It is also determined according to user requirements, such as 0.7, etc. When the ship conf s >θ conf If so, it indicates that the moored vessel is trustworthy, and the vessel's information is retained; if conf s ≤θ conf If the information is invalid, the moored vessel is unreliable, and its information should be deleted.

[0082] In some implementations, the category of each validly moored vessel in the slice image is determined based on the correspondence between the functional area categories and vessel categories, including:

[0083] The category of valid moored vessels is determined based on the functional area categories of the sliced ​​image.

[0084] For example, if the functional area category of the slice image is bulk cargo area, then the category of each valid moored vessel in the slice image is determined to be a bulk cargo ship.

[0085] If the functional area category of the slice image is container area, then the category of each valid moored vessel in the slice image is determined to be a container ship.

[0086] If the functional area category of the slice image is oil tank area, then the category of each valid moored vessel in the slice image is determined to be an oil tanker.

[0087] This method uses scene analysis to assist in inferring target categories and mines image information more deeply, which can significantly improve the accuracy of ship identification and classification.

[0088] To demonstrate the beneficial effects of the method described in this application, the inventors verified the method. The model parameters used are shown in Table 1.

[0089] Table 1. Model parameters for the embodiment:

[0090]

[0091] Recognition results as follows Figure 14 As shown, the yellow area is the bulk cargo area, the blue area is the container area, and the green area is the oil tank area. Correspondingly, the yellow box contains bulk carriers, the blue box contains container ships, and the green box contains oil tankers, verifying the practicality and reliability of the method in this application.

[0092] like Figure 2 , Figure 3 As shown, this invention provides a device for identifying port vessels and their categories based on spaceborne SAR images. The device can be implemented in software, hardware, or a combination of both. From a hardware perspective, such as... Figure 2 The diagram shown is a hardware architecture diagram of a computing device for identifying port vessels and their categories based on spaceborne SAR imagery, according to an embodiment of the present invention. (Except for...) Figure 2 In addition to the processor, memory, network interface, and non-volatile memory shown, the computing device in the embodiment may also include other hardware, such as a forwarding chip responsible for processing packets. Taking software implementation as an example, such as... Figure 3 As shown, a device in a logical sense is formed by the CPU of the computing device in which it is located reading the corresponding computer program from the non-volatile memory into the memory for execution.

[0093] Please refer to Figure 3 This invention provides an apparatus for identifying port vessels and their categories based on spaceborne SAR images. The apparatus includes:

[0094] Construction unit 300 is used to construct a ship dataset consisting of port spaceborne SAR images with multiple defined functional areas;

[0095] The first training unit 302 is used to train the pre-built first network model based on the dataset and the labels of each functional area to obtain the trained first network model.

[0096] The second training unit 304 is used to label the moored and sailing vessels in the dataset, and to train the pre-built second network model based on the labeled dataset to obtain the trained second network model.

[0097] Slicing unit 306 slices the port target image to be identified to obtain multiple slice images;

[0098] The detection unit 308 performs functional area classification and moored vessel detection on each slice image based on the first network model and the second network model, and determines the category of valid moored vessels in each slice image based on the comprehensive detection results.

[0099] In some implementations, the designated functional areas include bulk cargo areas, container areas, and oil tank areas.

[0100] In some implementations, the construction unit 300 is used to perform the following operations:

[0101] The target image is preprocessed to obtain an image with a preset format and preset size;

[0102] The preprocessed image is sliced ​​using a sliding window of a set size to obtain multiple image slices; the sliding window of the set size is equal to the input size requirements of the first network model and the second network model.

[0103] In some implementations, the detection unit 308 is used to perform the following operations:

[0104] Iterate through each slice image sequentially, and for each slice image encountered, perform the following:

[0105] S1, input the slice image into the first network model, output the functional area category of the slice image, and determine whether the functional area category belongs to one of the set functional areas; if yes, then execute S2; if no, then execute S3.

[0106] S2, input the sliced ​​image into the second network model, output the ship detection result; and determine whether there is at least one valid moored ship in the sliced ​​image based on the ship detection result. If yes, proceed to S4; otherwise, proceed to S3.

[0107] S3, remove the slice image, traverse the next slice image, and return to execute S1;

[0108] S4. Based on the correspondence between the functional area categories and the vessel categories in the slice image, determine the category of each validly moored vessel in the slice image.

[0109] In some implementations, the ship detection results include the ship's status, the ship's four-point coordinates, and confidence level; the ship's status includes anchored status and sailing status.

[0110] When the detection unit 308 determines whether there is at least one valid moored vessel in the slice image based on the vessel detection results, it performs the following operations:

[0111] For each detected vessel, execute:

[0112] If the vessel is in a state of being at sea, no action is taken, and the next vessel is detected;

[0113] If the vessel is in a moored state, calculate the aspect ratio of its rotating frame based on its four-point coordinates; determine whether the calculated aspect ratio is greater than the standard aspect ratio threshold for vessels; if not, remove the vessel and check the next vessel; if yes, determine whether the confidence level of the vessel is greater than the confidence level threshold; if yes, identify the vessel as a valid moored vessel; if not, remove the vessel and check the next vessel.

[0114] The detection results for validly moored vessels in the slice image are obtained after each vessel in the slice image has been detected.

[0115] In some implementations, the length-to-width ratio of the vessel is calculated as follows:

[0116] Calculate the lengths of the two adjacent sides of the rotating frame containing the ship based on the ship's four-point coordinates;

[0117] The ratio of the length of the longest side to the length of the shortest side is taken as the length-to-width ratio of the ship.

[0118] It should be noted that the apparatus for identifying port vessels and their categories based on spaceborne SAR images provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the apparatus can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the apparatus for identifying port vessels and their categories based on spaceborne SAR images provided in the above embodiments and the method embodiments for identifying port vessels and their categories based on spaceborne SAR images belong to the same concept. The specific implementation process is detailed in the method embodiments and will not be repeated here.

[0119] Embodiments of this application also provide a computer device, please refer to... Figure 3 The computer device includes a processor and a memory, the memory storing at least one instruction, at least one program, code set or instruction set, the at least one instruction, at least one program, code set or instruction set being loaded and executed by the processor to implement the method for identifying port moored vessels and their categories based on spaceborne SAR images provided in the above-described method embodiments.

[0120] Embodiments of this application also provide a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement the method for identifying port moored vessels and their categories based on spaceborne SAR images provided in the above-described method embodiments.

[0121] Embodiments of this application also provide a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium and executes the computer program, causing the computer device to perform the method for identifying port moored vessels and their categories based on spaceborne SAR images as described in any of the above embodiments.

[0122] For ease of description, the above systems or devices are described separately as various modules or units based on their functions. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware components.

[0123] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0124] Finally, it should be noted that in this document, relational terms such as first, second, third, and fourth are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0125] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for identifying port moored ships and their classes based on spaceborne SAR images, characterized by, The method comprises: constructing a ship dataset composed of a plurality of port starboard SAR images of set functional areas; training a pre-constructed first network model based on the dataset and labels of the functional areas to obtain a trained first network model; annotating the berthed ships and the ships at sea in the dataset, and training a pre-constructed second network model based on the annotated dataset to obtain a trained second network model; slicing a port target image to be recognized to obtain a plurality of slice images; performing functional area classification and berthed ship detection on each slice image based on the first network model and the second network model respectively, and determining the category of each effective berthed ship in each slice image based on the comprehensive detection results; the first network model is a ResNet101 network, the second network model is a YOLOv11 network, the YOLOv11 model architecture deeply fuses a multi-scale feature pyramid and a dynamic attention mechanism, uses an improved C2f structure to enhance the feature extraction capability of the backbone network, realizes efficient fusion of cross-layer features in combination with a BiFPN optimized neck network, and introduces an adaptive anchor frame mechanism and a lightweight decoupled head design in the head detection module; the method of performing functional area classification and berthed ship detection on each slice image based on the first network model and the second network model respectively, and determining the category of each effective berthed ship in each slice image based on the comprehensive detection results, comprises: iterating through each slice image in turn, and for each slice image that is iterated through, performing the following steps: S1, inputting the slice image into the first network model, outputting the functional area category of the slice image, and determining whether the functional area category belongs to one of the set functional areas; if yes, performing S2; if no, performing S3; S2, inputting the slice image into the second network model to output a ship detection result; and determining whether there is at least one effective berthed ship in the slice image based on the ship detection result, if yes, performing S4; if no, performing S3; S3, removing the slice image and iterating through the next slice image, and returning to perform S1; S4, determining the category of each effective berthed ship in the slice image based on the corresponding relationship between the functional area category and the ship category of the slice image; the ship detection result comprises the state of the ship, the four-point coordinates of the ship, and the confidence; the state of the ship comprises the berthing state and the sailing state; the method of determining whether there is at least one effective berthed ship in the slice image based on the ship detection result, comprises: for each ship that is detected, performing the following steps: if the state of the ship belongs to the sailing state, do not process it, and detect the next ship; if the state of the ship belongs to the berthing state, calculate the aspect ratio of the rotation frame based on the four-point coordinates; and determine whether the calculated aspect ratio is greater than the aspect ratio threshold of the standard ship; if no, remove the ship, and detect the next ship; if yes, determine whether the confidence of the ship is greater than the confidence threshold, if yes, determine the ship as an effective berthed ship, if no, remove the ship, and detect the next ship; Until each ship in the slice image is detected, the detection result of the effective berthing ship in the slice image is obtained.

2. The method of claim 1, wherein, The set function area includes bulk cargo area, container area and oil tank area.

3. The method of claim 1, wherein, The target port image to be recognized is sliced to obtain a plurality of slice images, including: The target image is preprocessed to obtain an image of a preset format and a preset size; The preprocessed image is cut by a sliding window of a set size to obtain a plurality of image slices; the sliding window of the set size is equal to the input size requirement of the first network model and the second network model.

4. The method of claim 1, wherein, The length-width ratio of the ship is calculated by the following method: The length of the adjacent two sides of the rotating frame where the ship is located is calculated based on the four coordinates of the ship. The ratio of the length of the largest side to the length of the smallest side is taken as the length-width ratio of the ship.

5. An apparatus for identifying ported ships and their classes based on space-borne SAR images, characterized by, The device for implementing the method of any one of claims 1-4 comprises: A construction unit for constructing a ship data set composed of a plurality of set function areas of a port space SAR image; A first training unit for training a first network model constructed in advance based on the data set and the labels of the function areas to obtain a trained first network model; A second training unit for labeling the berthing ships and the ships at sea in the data set and training a second network model constructed in advance based on the labeled data set to obtain a trained second network model; A slicing unit for slicing the target port image to be recognized to obtain a plurality of slice images; A detection unit for classifying the function areas and detecting the berthing ships of each slice image based on the first network model and the second network model, respectively, and determining the category of the effective berthing ship in each slice image based on the comprehensive detection result.

6. A computer device, comprising: The computer device comprises a memory and a processor, the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to realize the steps of the method of any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and the computer program is executed by the processor to realize the steps of the method of any one of claims 1-4.

8. A computer program product, characterised in that, The computer program is executed by the processor to realize the steps of the method of any one of claims 1-4.

Images