Ship target identification method based on overall and local satellite-borne SAR images

Abstract

The invention relates to the technical field of target identification, and discloses a ship target identification method based on overall and local satellite-borne SAR images. Comprising the following steps: inputting a to-be-recognized satellite-borne SAR original image into a trained ship recognition model to obtain an initial recognition result; the ship identification model is obtained based on the overall SAR image training of a plurality of ships of known types; based on the normalized position of each ship target, generating a region slice corresponding to the ship target; respectively adjusting the size of each region slice to obtain a new image with a preset size corresponding to the region slice; inputting each new image into the trained component identification model to obtain a component identification result of each ship target; the component identification model is obtained based on known component image training of a plurality of ships of known types; and correcting the initial type of each ship target based on the component identification result of each ship target to obtain a final identification result. The ship target can be accurately identified, and the false alarm rate is low.

Description

Technical Field

[0001] This invention relates to the field of target recognition technology, and in particular to a method for ship target recognition based on overall and local spaceborne SAR images. Background Technology

[0002] Spaceborne SAR (Synthetic Aperture Radar) imagery is one of the important sources of information on naval targets. Currently, the image resolution can reach high resolution levels such as 1.0 meter and 0.5 meter, achieving the standard for fine identification of naval targets and possessing the ability to analyze and obtain the target's location and category.

[0003] Spaceborne SAR imaging offers significant advantages such as all-weather, all-time, and high resolution, making it suitable for ship target identification. However, due to the coherent imaging characteristics of spaceborne SAR images, ship targets are susceptible to noise, shadows, and geometric distortions, especially in complex sea conditions and scenarios with significant near-shore scattering interference, resulting in a high false alarm rate and misidentification issues.

[0004] Therefore, there is an urgent need for a ship target identification method based on global and local spaceborne SAR images to solve the above problems. Summary of the Invention

[0005] This invention provides a ship target identification method based on global and local spaceborne SAR images, which can accurately identify ship targets with a low false alarm rate. The technical solution is as follows: On the one hand, a method for ship target identification based on global and local spaceborne SAR images is provided, the method comprising: The raw spaceborne SAR image to be identified is input into a pre-trained ship identification model to obtain initial identification results; the initial identification results include the initial type and normalized position of each ship target; the ship identification model is trained based on the overall SAR images of multiple known types of ships. A corresponding region slice is generated based on the normalized position of each ship target; The size of each region slice is adjusted to obtain a new image with a corresponding preset size; Each new image is input into a pre-trained component recognition model to obtain the component recognition result for each ship target; the component recognition model is trained based on known component images of multiple known types of ships; The initial type of each ship target is corrected based on the component identification results to obtain the final identification result.

[0006] On the other hand, a ship target identification device based on global and local spaceborne SAR images is provided, the device comprising: The pre-identification unit is used to input the original spaceborne SAR image to be identified into a pre-trained ship identification model to obtain an initial identification result; the initial identification result includes the initial type and normalized position of each ship target; the ship identification model is trained based on the overall SAR images of multiple known types of ships; The generation unit is used to generate a corresponding region slice based on the normalized position of each ship target. The adjustment unit is used to adjust the size of each region slice individually to obtain a new image with a corresponding preset size; The component recognition unit is used to input each new image into a pre-trained component recognition model to obtain the component recognition result for each ship target; the component recognition model is trained based on known component images of multiple known types of ships; The correction unit is used to correct the initial type of each ship target based on the component identification results to obtain the final identification result.

[0007] 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 ship target identification method based on global and local spaceborne SAR images described above.

[0008] On the other hand, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when the computer program is executed by a processor, it implements the steps of the ship target recognition method based on global and local spaceborne SAR images described above.

[0009] 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 ship target recognition method based on global and local spaceborne SAR images described above.

[0010] This invention provides a ship target recognition method based on global and local spaceborne SAR images. First, the ship target is identified as a whole, yielding an initial recognition result. Then, based on the initial recognition result, key components at smaller scales are further identified. Next, by designing basic discrimination logic, the component recognition results are used to correct the initial recognition result, resulting in three possible final recognition results: classifying the initial recognition result as background, retaining the original result, or obtaining a new result through further correction. This approach considers both ship and component recognition, resulting in higher accuracy. Therefore, this application not only retains the automation advantages of deep learning but also significantly reduces the false alarm rate and misidentification rate through a knowledge-guided mechanism, maintaining stable detection performance even under complex sea conditions and land interference, thus improving the accuracy of ship target recognition. Attached Figure Description

[0011] 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.

[0012] Figure 1 This is a flowchart of a ship target identification method based on global and local spaceborne SAR images provided by an embodiment of the present invention; Figure 2 This is a structural diagram of a ship target identification device based on overall and local spaceborne SAR images provided in an embodiment of the present invention; Figure 3 This is a hardware architecture diagram of a computer device provided in an embodiment of the present invention; Figure 4 This is an example diagram of a sample set composed of overall SAR images of multiple known types of ships provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of component images corresponding to a bulk carrier provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of component image samples corresponding to a container ship provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of component image samples corresponding to an oil tanker provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the determination result when the bridge is not present in the component identification result according to an embodiment of the present invention; Figure 9 This is a schematic diagram of the determination result obtained by using the similarity distance between the identified architecture vector and the standard architecture vector, provided by an embodiment of the present invention; Figure 10 This is a schematic diagram of the determination result provided by an embodiment of the present invention when the bridge is present in the component identification result, and the component type of the distinguishing component with the highest confidence level corresponds to the initial type of the ship target. Detailed Implementation

[0013] 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.

[0014] The specific implementation of the method in this application is described in detail below.

[0015] Please refer to Figure 1 This invention provides a method for ship target identification based on global and local spaceborne SAR images, the method comprising: Step 100: Input the original spaceborne SAR image to be identified into the pre-trained ship identification model to obtain the initial identification result; the initial identification result includes the initial type and normalized position of each ship target; the ship identification model is trained based on the overall SAR image of multiple known types of ships. Step 102: Generate a corresponding region slice based on the normalized position of each ship target; Step 104: Adjust the size of each region slice to obtain a new image with a corresponding preset size; Step 106: Input each new image into the pre-trained component recognition model to obtain the component recognition result for each ship target; the component recognition model is trained based on known component images of multiple known types of ships; Step 108: Correct the initial type of each ship target based on the component identification results to obtain the final identification result.

[0016] In this embodiment, the ship target is first identified as a whole to obtain an initial identification result. Then, based on the initial identification result, key components at smaller scales are further identified. Next, by designing basic discrimination logic, the component identification results are used to correct the initial identification result, resulting in three possible final identification results: classifying the initial identification result as background, retaining the original result, or obtaining a new result through further correction. This approach considers both ship identification and component identification results, leading to higher accuracy. Therefore, this application not only retains the automation advantages of deep learning but also significantly reduces the false alarm rate and misidentification rate through a knowledge-guided mechanism, maintaining stable detection performance even under complex sea conditions and land interference, thus improving the accuracy of ship target identification.

[0017] The following description Figure 1 The execution method of each step is shown.

[0018] First, regarding step 100: The overall SAR images of multiple known ship types are high-resolution SAR ship image samples. Each ship type within each image sample is labeled with a rotated bounding box. Ship types can be civilian or military vessels, such as bulk carriers, container ships, and oil tankers. Figure 4 As shown. Of course, some ships that do not require precise identification can be labeled as other types. After the sample is determined, the various types of ships can be divided into training and validation sets according to a set ratio, such as 4:1.

[0019] Furthermore, the ship identification model can employ a YOLOv8 network. To reduce the possibility of missed targets by the YOLOv8 pre-identification network, this step uses a YOLOv8-p6 architecture, deepening the network structure and optimizing the localization and classification accuracy of ship targets under complex sea conditions and land interference, ensuring that as many ship targets as possible are included in the pre-identification results. Of course, users can also use other network models, and this application is not limited to them. The model training process is a common technique in this field and will not be described in detail here.

[0020] For step 102, the following are included: For each ship target, the following actions are executed: Multiply the normalized position of the ship target by the size of the original image to obtain the coordinate position of the ship target in the original image; Based on the maximum and minimum values ​​of the x and y coordinates of this coordinate position, determine the rotating frame containing the ship target. Generate a minimum bounding rectangle region containing the rotated frame in the original image, and use this rectangle region as a region slice of the ship target.

[0021] In this step, the maximum and minimum values ​​of the x and y coordinates are obtained by comparing the four coordinates of the ship target, as follows: Minimum value of the x-axis The maximum value of the x-axis Minimum value of the ordinate The maximum value of the y-axis This yields the minimum bounding rectangle region of the ship target's rotation bounding box in the input image. As a regional slice of the corresponding ship target I slice .

[0022] In some implementations, the preset size is the size of the input image of the component recognition model; for step 104, it includes: For each region slice, the process expands outwards from the center of the slice until the size of the region slice equals the preset size, resulting in a new image. The pixel value of each pixel in the new image is then determined based on the following formula: In the formula, I g Indicates a new image; Represents a new image ( i,j The pixel value at the given location; I slice Indicates a region slice; Represents a region slice Pixel value at the location; w and h These are the width and height of the new image, respectively; w s and h s The width and height of the respective region slices.

[0023] This step completes the region slices, allowing their size to fit the part recognition model. By setting all pixel values ​​of the supplemented regions to 0, it prevents them from interfering with subsequent part recognition results.

[0024] Regarding step 106: In the ship samples used to train the ship identification model and component identification model, the key components of each known type of ship include the bridge and a distinguishing component. The distinguishing component is different for each type of ship, and the component type of each distinguishing component corresponds one-to-one with the corresponding ship type.

[0025] When constructing image samples for component recognition models, rotating bounding boxes are used to label key components for each type of ship. For example, bulk carriers include the bridge and bulk cargo holds, container ships include the bridge and containers, and oil tankers include the bridge and oil pipelines. Since all ship targets in each class include a bridge and there is no clear distinction between different types of bridges, the bridge is not used to differentiate ship categories. Therefore, during sample labeling, the bridges of all ship types are uniformly labeled as "bridge," without specifying the ship type. Figures 5-7 As shown in the images, the solid lines in all three images represent the bridge. Figure 5 The dashed box in the middle represents the warehouse. Figure 6 The dashed box in the middle represents a container. Figure 7 The dashed box in the middle represents an oil pipeline.

[0026] Image samples for differentiating various parts are also divided into training and validation sets according to a preset ratio, such as 4:1. Furthermore, to improve the accuracy of the YOLOv8 part refinement recognition network, this step employs the YOLOv8-p2 architecture to deepen the network structure, focus on smaller-scale part information, optimize small target localization and classification accuracy, and ensure the accuracy and reliability of part recognition. Similarly, users can also use other network models, and this application is not limited to them. The model training process is a common technique in this field and will not be elaborated upon here.

[0027] In some implementations, the component identification results include the component type, location, and confidence level of each key component; for step 108, the following are included: For each ship target, the following actions are executed: S1, determine whether the ship target's component identification results include the bridge; if yes, proceed to S2; otherwise, determine the ship target as background. S2, determine whether the ship target has other distinguishing components. If yes, execute S3. If no, determine that the ship target is the background. S3, determine whether the component type of the distinguishing component with the highest confidence level corresponds to the initial type of the ship target. If yes, use the initial type as the final identification result; otherwise, proceed to S4. S4. Construct a standard architecture vector for each type of ship and an identification architecture vector between the key components of the target ship. Based on the similarity distance between the identification architecture vector and each standard architecture vector, determine the final identification result.

[0028] In some implementations, step 104 includes: The orientation of the components is determined based on the maximum and minimum values ​​of the horizontal and vertical coordinates of each key component of the ship target. Based on the arrangement direction, determine the standard architecture vector according to the actual arrangement order of each key component in each type of ship; Based on the arrangement direction and the coordinates of the key components identified in the ship target, they are sorted in ascending order to obtain the identification architecture vector of the ship target. Calculate the similarity distance between the identified architecture vector and each standard architecture vector in turn; If all similarity distances are greater than a set distance threshold, the ship target is determined to be background; otherwise, the ship category corresponding to the smallest similarity distance is taken as the final identification result.

[0029] In this step, the basic arrangement direction of the component architecture is determined by the four-point positions of all key components, specifically as follows: when When arranging, follow the horizontal direction; when When arranged vertically; in, The minimum value of the x-coordinate among all key components; This represents the maximum value of the x-coordinate among all key components; The minimum value of the ordinate among all key components; This represents the maximum value of the ordinate among all critical components.

[0030] Based on the component arrangement direction, for each valid bridge (i.e., including at least one distinguishing component in addition to the bridge), the bridge and other valid components are sorted according to their arrangement direction to form a component identification architecture vector. , cls k To identify the first element in the architecture vector k Category number of each key component.

[0031] Given a standard architecture vector, the positional relationships of key components of a known type of ship are fixed. Therefore, by arranging the key components according to the above-mentioned arrangement direction based on their actual positions, the standard architecture vector of the ship can be obtained.

[0032] For example, considering three types of ships—bulk carriers, container ships, and tankers—bulk carriers consist of a bridge and multiple cargo holds, container ships consist of a bridge and multiple containers, and tankers consist of a bridge and multiple oil pipelines. The bridges of bulk carriers and tankers are located on one side of the hull, while the bridges of container ships are located on both sides of the hull. Furthermore, assuming the bridge's category number is 0, the cargo holds' category number is 1, the containers' category number is 2, and the oil pipelines' category number is 3, then: The standard architecture vector for bulk carriers is: ; The standard architecture vector for container ships is: ; The standard architecture vector for oil tankers is: .

[0033] In some implementations, the similarity distance between the identified architecture vector and each standard architecture vector is calculated using the following formula: When the bridge is on one side of the ship, ; When the bridge is in the middle of the ship, ; In the formula, To identify the architecture vector and the first m The similarity distance between the standard architectural vectors of ship-like vessels; ξ m For the first m Normalized parameters for ship-like targets, m =1~ M , M The number of types of ship targets; ω k , cls k and confidence k These represent the first and second elements in the recognition architecture vector. k The weights, category numbers, and confidence levels of each key component; m k Indicates the first m The first in the standard architecture vector of ship-like vessels k Category numbers for key components; K To identify the number of key components in the architecture vector; E m For the first m The average distance to key components corresponding to ship-like targets; A m and A n The first m and the n Category codes for distinguishing components corresponding to different types of ships. n =1~ M .

[0034] It should be noted that the weight of each critical component in the architecture vector is related to its position relative to the bridge. The closer a critical component is to the bridge, the greater its weight, and vice versa. For example, a component adjacent to the bridge vector with a distance of 1 from the bridge has a weight coefficient of 1 / 2; and so on. tFor key components, the weighting factor is 1 / 2. t Therefore, it is advisable. ω k =1 / 2 t The weight vector composed of each key component. It can be determined using a power function, as shown below: When the number of critical components k approaches In this case, the distance of each component from the bridge T Also tending to It can be guaranteed , t =1,2,…… T , T is The number of digits furthest from the bridge.

[0035] In some implementations, the distance threshold is determined using the following formula: In the formula, θ dis The number of ship categories is M The set distance threshold is determined when the number of ship types is determined. Therefore, the set distance threshold is also determined when the number of ship types is determined.

[0036] To demonstrate the beneficial effects of the present invention, the present invention has been verified using the method of this application, taking three types of ships—bulk cargo ships, container ships, and oil tankers—as well as other types of ships as examples.

[0037] The network model parameters are shown in Table 1: Table 1 Parameters of the Example

[0038] Based on the above parameters, the method of this application can be used to calculate accurate and precise ship target identification results. Figure 4 Example images of a high-resolution SAR ship target dataset are provided, mainly including bulk carriers, container ships, and oil tankers. Other ships are classified as "other" and are not within the scope of fine-grained identification. Figures 5-7 A high-resolution SAR dataset of ship components is provided, corresponding to the ship categories: bulk carriers include a bridge and multiple cargo holds, container ships include a bridge and two containers, and oil tankers include a bridge and an oil pipeline. Figures 8-10 The example of the results using the method of this application shows that, through pre-identification-component fine identification-identification correction, false alarm targets can be identified as background, incorrectly identified targets can be corrected, and correct identification results can be retained. The false alarm rate and misidentification rate are suppressed to below 10%, thereby improving the identification accuracy and verifying the practicality and reliability of the method of this application.

[0039] like Figure 2 , Figure 3 As shown, this embodiment of the invention provides a ship target identification device based on overall and local spaceborne SAR images. The device embodiment can be implemented through software, hardware, or a combination of both. From a hardware perspective, as... Figure 2 The diagram shown is a hardware architecture diagram of a computing device for a ship target identification device based on overall and local spaceborne SAR images, provided by 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.

[0040] Please refer to Figure 3 This invention provides a ship target identification device based on overall and local spaceborne SAR images. The device includes: The pre-identification unit 300 is used to input the original spaceborne SAR image to be identified into the pre-trained ship identification model to obtain the initial identification result; the initial identification result includes the initial type and normalized position of each ship target; the ship identification model is trained based on the overall SAR image of multiple known types of ships; The generation unit 302 is used to generate a corresponding region slice based on the normalized position of each ship target. The adjustment unit 304 is used to adjust the size of each region slice to obtain a new image with a corresponding preset size; The component recognition unit 306 is used to input each new image into the pre-trained component recognition model to obtain the component recognition result for each ship target; the component recognition model is trained based on known component images of multiple known types of ships; The correction unit 308 is used to correct the initial type of each ship target based on the component identification result to obtain the final identification result.

[0041] In some implementations, the generation unit 302 is used to perform the following operations: For each ship target, the following actions are executed: Multiply the normalized position of the ship target by the size of the original image to obtain the coordinate position of the ship target in the original image; Based on the maximum and minimum values ​​of the x and y coordinates of this coordinate position, determine the rotating frame containing the ship target. Generate a minimum bounding rectangle region containing the rotated frame in the original image, and use this rectangle region as a region slice of the ship target.

[0042] In some implementations, the preset size is the size of the input image of the component recognition model; the adjustment unit 304 is used to perform the following operations: For each region slice, the process expands outwards from the center of the slice until the size of the region slice equals the preset size, resulting in a new image. The pixel value of each pixel in the new image is then determined based on the following formula: In the formula, I g Indicates a new image; Represents a new image ( i,j The pixel value at the given location; I slice Indicates a region slice; Represents a region slice Pixel value at the location; w and h These are the width and height of the new image, respectively; w s and h s These represent the width and height of the region slice, respectively.

[0043] In some implementations, in the ship samples used to train the ship identification model and component identification model, the key components of each known type of ship include a bridge and a distinguishing component, and the distinguishing component is different for each type of ship. The component type of each distinguishing component corresponds one-to-one with the corresponding ship type.

[0044] In some implementations, the component identification results include the component type, location, and confidence level of each key component; the correction unit 308 is used to perform the following operations: For each ship target, the following actions are executed: S1, determine whether the ship target's component identification results include the bridge; if yes, proceed to S2; otherwise, determine the ship target as background. S2, determine whether the ship target has other distinguishing components. If yes, execute S3. If no, determine that the ship target is the background. S3, determine whether the component type of the distinguishing component with the highest confidence level corresponds to the initial type of the ship target. If yes, use the initial type as the final identification result; otherwise, proceed to S4. S4. Construct a standard architecture vector for each type of ship and an identification architecture vector between the key components of the target ship. Based on the similarity distance between the identification architecture vector and each standard architecture vector, determine the final identification result.

[0045] In some implementations, step S4 is used to perform the following operations: The orientation of the components is determined based on the maximum and minimum values ​​of the horizontal and vertical coordinates of each key component of the ship target. Based on the arrangement direction, determine the standard architecture vector according to the actual arrangement order of each key component in each type of ship; Based on the arrangement direction and the coordinates of the key components identified in the ship target, they are sorted in ascending order to obtain the identification architecture vector of the ship target. Calculate the similarity distance between the identified architecture vector and each standard architecture vector in turn; If all similarity distances are greater than a set distance threshold, the ship target is determined to be background; otherwise, the ship category corresponding to the smallest similarity distance is taken as the final identification result.

[0046] In some implementations, the similarity distance between the identified architecture vector and each standard architecture vector is calculated using the following formula: When the bridge is on one side of the ship, ; When the bridge is in the middle of the ship, ; In the formula, To identify the architecture vector and the first m The similarity distance between the standard architectural vectors of ship-like vessels; ξ m For the first m Normalized parameters for ship-like targets, m =1~ M , M The number of types of ship targets; ω k , cls k and confidence k These represent the first and second elements in the recognition architecture vector. k The weights, category numbers, and confidence levels of each key component; m k Indicates the first m The first in the standard architecture vector of ship-like vessels k Category numbers for key components; KTo identify the number of key components in the architecture vector; E m For the first m The average distance to key components corresponding to ship-like targets; A m and A n The first m and the n Category codes for distinguishing components corresponding to different types of ships. n =1~ M .

[0047] It should be noted that the ship target recognition device based on overall and local 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 device can be divided into different functional modules to complete all or part of the functions described above. In addition, the ship target recognition device based on overall and local spaceborne SAR images provided in the above embodiments and the ship target recognition method based on overall and local spaceborne SAR images belong to the same concept. The specific implementation process is detailed in the method embodiments and will not be repeated here.

[0048] 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 ship target recognition method based on global and local spaceborne SAR images provided in the above-described method embodiments.

[0049] The 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 ship target identification method based on global and local spaceborne SAR images provided in the above-described method embodiments.

[0050] 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 ship target identification method based on global and local spaceborne SAR images as described in any of the above embodiments.

[0051] 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.

[0052] 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.

[0053] 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.

[0054] 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 ship target identification based on global and local spaceborne SAR images, characterized in that, The method includes: The raw spaceborne SAR image to be identified is input into a pre-trained ship identification model to obtain initial identification results; the initial identification results include the initial type and normalized position of each ship target; the ship identification model is trained based on the overall SAR images of multiple known types of ships. A corresponding region slice is generated based on the normalized position of each ship target; The size of each region slice is adjusted to obtain a new image with a corresponding preset size; Each new image is input into a pre-trained component recognition model to obtain the component recognition result for each ship target; the component recognition model is trained based on known component images of multiple known types of ships; The initial type of each ship target is corrected based on the component identification results to obtain the final identification result.

2. The method according to claim 1, characterized in that, The step of generating a corresponding region slice based on the normalized position of each ship target includes: For each ship target, the following actions are executed: Multiply the normalized position of the ship target by the size of the original image to obtain the coordinate position of the ship target in the original image; Based on the maximum and minimum values ​​of the x and y coordinates of this coordinate position, determine the rotating frame containing the ship target. A minimum bounding rectangle region containing the rotated frame is generated in the original image, and this rectangle region is used as a region slice of the ship target.

3. The method according to claim 1, characterized in that, The preset size is the size of the input image of the component recognition model; The step of adjusting the size of each region slice to obtain a new image with a corresponding preset size includes: For each region slice, the image is expanded outwards from the center of the region slice until the size of the region slice equals the preset size, resulting in a new image. The pixel value of each pixel in the new image is then determined based on the following formula: In the formula, I g Indicates a new image; Represents a new image ( i,j The pixel value at the given location; I slice Indicates a region slice; Represents a region slice Pixel value at the location; w and h These are the width and height of the new image, respectively; w s and h s These represent the width and height of the region slice, respectively.

4. The method according to claim 1, characterized in that, In the ship samples used to train the ship identification model and the component identification model, the key components of each known type of ship include a bridge and a distinguishing component, and the distinguishing component is different for each type of ship. The component type of each distinguishing component corresponds one-to-one with the corresponding ship type.

5. The method according to claim 4, characterized in that, The component identification results include the component type, location, and confidence level of each key component; The initial type is corrected based on the component identification results of each ship target to obtain the final identification result, including: For each ship target, the following actions are executed: S1, determine whether the ship target's component identification results include the bridge; if yes, proceed to S2; otherwise, determine the ship target as background. S2, determine whether the ship target has other distinguishing components. If yes, execute S3. If no, determine that the ship target is the background. S3, determine whether the component type of the distinguishing component with the highest confidence level corresponds to the initial type of the ship target. If yes, use the initial type as the final identification result; otherwise, proceed to S4. S4. Construct a standard architecture vector for each type of ship and an identification architecture vector between the key components of the target ship. Based on the similarity distance between the identification architecture vector and each standard architecture vector, determine the final identification result.

6. The method of claim 5, characterized in that, S4 includes: The orientation of the components is determined based on the maximum and minimum values ​​of the horizontal and vertical coordinates of each key component of the ship target. Based on the arrangement direction, the standard architecture vector is determined according to the actual arrangement order of each key component in each type of ship. Based on the arrangement direction and the coordinates of the identification positions of each key component in the ship target, they are sorted in ascending order to obtain the identification architecture vector of the ship target. Calculate the similarity distance between the identified architecture vector and each standard architecture vector in turn; If all similarity distances are greater than a set distance threshold, the ship target is determined to be background; otherwise, the ship category corresponding to the smallest similarity distance is taken as the final identification result.

7. The method of claim 6, characterized in that, The similarity distance between the identified architecture vector and each standard architecture vector is calculated using the following formula: When the bridge is on one side of the ship, ; When the bridge is in the middle of the ship, ; In the formula, To identify the architecture vector and the first m The similarity distance between the standard architectural vectors of ship-like vessels; ξ m For the first m Normalized parameters for ship-like targets, m =1~ M , M The number of types of ship targets; ω k , cls k and confidence k These represent the first and second elements in the recognition architecture vector. k The weights, category numbers, and confidence levels of each key component; m k Indicates the first m The first in the standard architecture vector of ship-like vessels k Category numbers for key components; K To identify the number of key components in the architecture vector; E m For the first m The average distance to key components corresponding to ship-like targets; A m and A n The first m and the n Category codes for distinguishing components corresponding to different types of ships. n =1~ M .

8. A ship target identification device based on overall and local spaceborne SAR images, characterized in that, The device includes: The pre-identification unit is used to input the original spaceborne SAR image to be identified into a pre-trained ship identification model to obtain an initial identification result; the initial identification result includes the initial type and normalized position of each ship target; the ship identification model is trained based on the overall SAR images of multiple known types of ships; The generation unit is used to generate a corresponding region slice based on the normalized position of each ship target. The adjustment unit is used to adjust the size of each region slice individually to obtain a new image with a corresponding preset size; The component recognition unit is used to input each new image into a pre-trained component recognition model to obtain the component recognition result for each ship target; the component recognition model is trained based on known component images of multiple known types of ships; The correction unit is used to correct the initial type of each ship target based on the component identification results to obtain the final identification result.

9. A computer device, characterized in that, The computer device includes a memory and a processor. The memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory to implement the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method described in any one of claims 1-7.

Images