Medium-orbit spaceborne SAR ship target detection method based on distance compressed domain

Abstract

The invention discloses a medium-orbit spaceborne SAR ship target detection method based on a distance compression domain. The method comprises the following steps: acquiring to-be-detected medium-orbit spaceborne SAR ship echo data; performing range direction matched filtering processing on the echo data to obtain range compressed domain data; inputting the distance compressed domain data into a pre-trained target detection model to obtain a ship target detection result; the target detection model is constructed based on an Anchor-Basic structure, when the model generates a prediction frame, anchor frame priori knowledge generated based on medium orbit detection parameters and ship target parameters in a data set is used as a constraint condition, and prediction frame size and anchor frame priori knowledge offset penalty constraints are fused in a loss function; the data set is composed of SAR distance compressed domain data samples generated by multiple known types of ship targets under known medium orbit satellite detection parameters. According to the method, the detection efficiency and precision of the medium-orbit SAR ship target can be remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of target recognition technology, and in particular to a method for detecting ship targets on a medium-orbit spaceborne SAR based on range compression domain. Background Technology

[0002] In the field of synthetic aperture radar (SAR) for maritime observation, medium-orbit spaceborne SAR, due to its orbital characteristics, inherently presents a contradiction between coverage width and imaging resolution when achieving wide-area ship surveillance in the open sea. Current mainstream detection technologies are all based on the traditional "image first, detection later" process, which involves first performing complex two-dimensional focusing processing on the full-scene echo data to generate a SAR image, and then identifying targets based on the visual features in the image. However, this approach faces challenges due to the unique characteristics of medium-orbit SAR. On the one hand, the large image swath width of medium-orbit SAR (typically thousands of kilometers) and the sparse distribution of ship targets on the vast sea surface result in a computationally intensive and resource-intensive method. On the other hand, the long synthetic aperture time unique to medium-orbit SAR amplifies the influence of ship motion, leading to severe defocusing of targets in the final image, reduced saliency of target visual features, and decreased detection accuracy.

[0003] To improve detection efficiency and accuracy, a method has emerged that bypasses the imaging process and directly performs target detection in the range compression domain. However, this method is only suitable for low-Earth orbit or airborne SAR designs. When facing complex scenarios with drastic spatial variations in the target range migration curve caused by the large field of view and long synthetic aperture time of medium-Earth orbit SAR, the detection accuracy is poor and the false alarm rate is high.

[0004] Therefore, there is an urgent need for a medium-orbit spaceborne SAR ship target detection method based on range compression domain to solve the above problems. Summary of the Invention

[0005] This invention provides a method for detecting ship targets using medium-orbit spaceborne SAR based on range compression domain, which can significantly improve the detection efficiency and accuracy of medium-orbit SAR ship targets. The technical solution is as follows: On the one hand, a method for detecting ship targets on a medium-orbit spaceborne SAR system based on range compression domain is provided, including: Acquire echo data of the medium-orbit spaceborne SAR ship to be detected; The echo data is subjected to range-direction matched filtering to obtain range-compressed domain data; The range-compressed domain data is input into a pre-trained target detection model to obtain ship target detection results. The target detection model is built on an anchor-based structure and trained on the dataset. When generating prediction boxes, the model uses prior knowledge of anchor boxes generated based on the medium-Earth orbit detection parameters and ship target parameters in the dataset as constraints, and the loss function incorporates a penalty constraint for the offset between the prediction box size and the prior knowledge of the anchor boxes. The dataset consists of SAR range-compressed domain data samples generated by various known types of ship targets under known medium-Earth orbit satellite detection parameters.

[0006] On the other hand, a mid-orbit spaceborne SAR ship target detection device based on range compression domain is provided, including: The acquisition unit is used to acquire the echo data of the medium-orbit spaceborne SAR ship to be detected; The preprocessing unit is used to perform range-direction matched filtering on the echo data to obtain range-compressed domain data; The detection unit is used to input the range compressed domain data into a pre-trained target detection model to obtain ship target detection results. The target detection model is built on an anchor-based structure and trained on the dataset. When generating prediction boxes, the model uses prior knowledge of anchor boxes generated based on the medium orbit detection parameters and ship target parameters in the dataset as constraints, and the loss function incorporates the offset penalty constraint between the prediction box size and the prior knowledge of the anchor boxes. The dataset consists of SAR range compressed domain data samples generated by various known types of ship targets under known medium orbit satellite detection parameters.

[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 medium-orbit spaceborne SAR ship target detection method based on the range compression domain 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, the steps of the above-described method for detecting ship targets in medium-orbit spaceborne SAR based on range compression domain are implemented.

[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 above-described method for detecting medium-orbit spaceborne SAR ship targets based on range compression domain.

[0010] This invention provides a method for detecting ship targets using mid-orbit spaceborne SAR based on range compression domain. First, by conducting detection directly in the range compression domain, the time-consuming two-dimensional focusing processing in the traditional "image first, detect later" process is eliminated, significantly reducing computational complexity and processing latency. In typical mid-orbit SAR wide-area scenarios, the detection speed is increased by more than five times, meeting the requirements of high-timeliness surveillance. Second, prior knowledge of anchor frames is generated based on mid-orbit detection parameters and ship target parameters in the dataset. For the loss function during training, a penalty is added to the bounding box regression term to offset the anchor frame width and height from the prior dimensions. This allows the model to learn size constraints conforming to mid-orbit imaging geometry during training, thereby enhancing positioning accuracy and model convergence speed. Finally, using the aforementioned prior knowledge as constraints for predicted boxes during model training establishes a mapping relationship between target range compression domain features and anchor frame dimensions. During inference, the anchor frame adaptively adjusts with changes in the range gate, significantly improving the matching degree between the anchor frame and the real target, and reducing missed detections and false alarms caused by size drift. Therefore, this application can improve the detection efficiency and accuracy of mid-orbit SAR ship targets. 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 medium-orbit spaceborne SAR ship target detection method based on range compression domain provided by an embodiment of the present invention; Figure 2 This is a network schematic diagram of a target detection model provided in an embodiment of the present invention; Figure 3 This is a structural diagram of a medium-orbit spaceborne SAR ship target detection device based on range compression domain provided in an embodiment of the present invention; Figure 4 This is a hardware architecture diagram of a computer device provided in an embodiment of the present invention. 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 detecting ship targets on a medium-orbit spaceborne SAR system based on range compression domain. The method includes: Step 100: Obtain the echo data of the medium-orbit spaceborne SAR ship to be detected; Step 102: Perform range-direction matched filtering on the echo data to obtain range-compressed domain data; Step 104: Input the range compressed domain data into a pre-trained target detection model to obtain the ship target detection result; the target detection model is built based on an anchor-based structure. When generating prediction boxes, the model uses prior knowledge of anchor boxes generated based on the mid-orbit detection parameters and ship target parameters in the dataset as constraints, and the loss function incorporates the offset penalty constraint between the prediction box size and the prior knowledge of the anchor boxes; the dataset consists of SAR range compressed domain data samples generated by various known types of ship targets under known mid-orbit satellite detection parameters.

[0016] In this embodiment, firstly, by performing detection directly in the range compression domain, the time-consuming two-dimensional focusing processing in the traditional "imaging first, detection later" process can be eliminated, significantly reducing computational complexity and processing latency. In typical medium-orbit SAR wide-area scenarios, the detection speed is improved by more than 5 times, meeting the requirements of high-timeliness surveillance. Secondly, by generating anchor frame prior knowledge based on the medium-orbit detection parameters and ship target parameters in the dataset, and by adding a penalty for the offset between the anchor frame width and height and the prior size to the bounding box regression term in the loss function during training, the model learns size constraints conforming to the medium-orbit imaging geometry during the training phase, thereby enhancing positioning accuracy and model convergence speed. Finally, using the aforementioned prior knowledge as constraints for the predicted bounding boxes in model training, a mapping relationship between target range compression domain features and anchor frame size can be established. During inference, the anchor frame adaptively adjusts with changes in the range gate, significantly improving the matching degree between the anchor frame and the real target, and reducing missed detections and false alarms caused by size drift. Therefore, this application can improve the detection efficiency and accuracy of medium-orbit SAR ship targets.

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

[0018] First, regarding steps 100 and 102: When detecting real ship targets, SAR ship echo data is first acquired using a medium-orbit satellite-borne radar. Then, the echo data undergoes range-direction matched filtering. The specific process is as follows: First, a range-direction Fourier transform is performed on the echo signal. Then, a complex multiplication operation is performed with the stored range-direction reference function. Finally, an inverse Fourier transform is performed to obtain the range-compressed domain data. This range-compressed domain data is a two-dimensional matrix with dimensions of azimuth time × range gate. The ship target in this domain is represented as a range migration curve spanning multiple range gates.

[0019] In addition, in some implementations, the generated distance compression domain data can be further processed, including histogram equalization to enhance target saliency, downsampling to reduce data volume, and overlapping sliding window cutting along the range and azimuth directions to adapt to the input size of the detection network and ensure target integrity. Users can adjust these settings as needed; no specific limitations are specified here.

[0020] It should also be noted that the orbital altitude of medium-Earth orbit satellites is approximately 2,000-36,000 kilometers. Of course, the specific altitude can be adjusted according to actual circumstances, and this application does not impose a specific limitation.

[0021] Regarding step 104: Because the length and shape of the range migration curve of mid-orbit SAR change drastically with the detection angle and target movement, the universal fixed anchor frame is completely ineffective in this scenario. To solve this specific problem, this step is based on the range history model of mid-orbit SAR to accurately analyze the target range migration characteristics, providing accurate prior information on the anchor frame size for the detection model.

[0022] In this step, the prior knowledge of the anchor frame mainly includes: (1) the minimum basic size of the anchor frame when the ship target is stationary; (2) the dynamic change range of the anchor frame when the ship target is in motion; and (3) the range of change of the basic size of the anchor frame under different detection angles of the medium-orbit satellite.

[0023] The specific content and calculation method of each type of prior knowledge will be explained in detail below.

[0024] (1) When the ship target is stationary, the minimum basic dimensions of the anchor frame include the basic height and the basic width of the anchor frame; the basic height and the basic width are obtained based on the number of pixels occupied by the stationary ship target in the azimuth and range directions in the range compression domain, respectively. The foundation height of the anchor frame is calculated using the following formula: The base width of the anchor frame is calculated using the following formula: In the formula, The base height of the anchor frame is the sum of the number of pixels occupied by the stationary ship target in the azimuth direction within the range compression domain. The base width of the anchor frame is the sum of the number of pixels occupied by the stationary ship target in the distance compression domain. The system pulse repetition frequency; For azimuth synthesis aperture time; The satellite's ground velocity; Wavelength; r The shortest distance between the target and the radar; This refers to the azimuth resolution. This is the radar's equivalent velocity; The range-direction signal sampling rate; This represents the range of variation in signal distance migration. c The speed of sound.

[0025] If the size of the predicted bounding box generated by the object detection model is smaller than the minimum basic size, it is determined that the constraint is not met.

[0026] (2) When the ship target is in motion, the dynamic change range of the anchor frame includes the dynamic change range of the anchor frame height and the dynamic change range of the anchor frame width. The minimum value of the dynamic change range of the anchor frame height is the base height of the anchor frame, and the maximum value is the azimuth pixel position of the target signal in the range compression domain after considering the ship target's speed. .in, Calculated using the following formula: The minimum value of the dynamic variation range of the anchor frame width is the base width of the anchor frame, and the maximum value is the distance-to-pixel distance occupied by the target signal in the distance compression domain after considering the ship target's movement speed. .in, Calculated using the following formula: In the formula, The increase in the synthetic aperture time caused by the target azimuth velocity; This represents the component of the target velocity along the azimuth direction. This is the difference between the maximum and minimum distances from the target to the migration.

[0027] In summary, the dynamic range of the anchor frame height is as follows: The dynamic range of the anchor frame width is as follows: The aforementioned dynamic range is primarily used to analyze the impact of target motion on distance migration. The size of the predicted bounding box generated by the target detection model should fall within this dynamic range.

[0028] (3) The range of variation of the anchor frame foundation size of the medium-orbit satellite under different detection perspectives, including the range of variation of the anchor frame foundation height and the range of variation of the anchor frame foundation width; the maximum values ​​of the range of variation of the anchor frame foundation height and the anchor frame foundation width are respectively the number of pixels of the stationary ship target in the azimuth and range directions in the range compression domain under the set observation perspective.

[0029] The relationship between the number of pixels in the azimuth direction of a stationary point target in the range compression domain and the viewing angle under observation is as follows: The relationship between the number of range pixels of a stationary point target in the range compression domain and the viewing angle under observation is as follows: In the formula, For the perspective under observation At that time, the number of azimuth pixels of a stationary target in the range compression domain; For the perspective under observation At that time, the number of distance pixels of a stationary target in the distance compression domain; The average radius of the Earth This refers to the height of the middle track.

[0030] Based on the above constraints, for any observation viewpoint, the size of the predicted bounding box generated by the target detection model should be within the range of variation of the anchor frame base size.

[0031] In summary, this step, through the modeling and analysis of target range migration characteristics under static, moving, and different perspectives, yields prior knowledge that accurately describes the size range of medium-orbit SAR ship targets in the range compression domain. This includes basic dimensions, dynamic variation range, and a quantitative mapping relationship between anchor frame size and range gate. This provides core physical constraints for the subsequent anchor frame design and inference selection of the target detection model. It ensures that during the inference process, the reasonableness of the anchor frame size is judged on the detection results, targets that do not conform to the range specified by the prior knowledge of the anchor frame are eliminated, thus suppressing false alarms.

[0032] Furthermore, the quality of the dataset directly affects the detection performance of the object detection model; therefore, it is necessary to determine the dataset appropriately. In this step, the dataset is obtained based on at least one method: forward simulation or reverse decoupling. The two methods are explained below: 1. The forward simulation method generates the dataset as follows: based on the known geometric parameters and electromagnetic scattering model of the ship target, combined with the orbital parameters and imaging parameters of the medium-orbit SAR satellite, range-compressed domain data containing the true range migration characteristics is generated through electromagnetic calculations and echo simulation. This data can be mainly used for the verification of the target detection model.

[0033] 2. The inverse decoupling method generates the dataset as follows: Ship target slices are cropped from existing SAR images at a preset resolution; combined with mid-orbit SAR imaging parameters; and through the inverse process of the imaging algorithm, range-compressed domain data corresponding to the target is generated. This method has high generation efficiency and can quickly generate a large number of training samples.

[0034] This application addresses the scarcity of mid-orbit SAR samples by employing a dataset construction method, thus providing data support for model training. Furthermore, the two dataset generation methods mentioned above can be used simultaneously, or only one method can be used, as long as the diversity and accuracy of the dataset are guaranteed.

[0035] Similarly, to further improve the performance of the dataset, the generated distance compression domain data can be further processed, including histogram equalization to enhance target saliency, downsampling to reduce data volume, and overlapping sliding window cutting along the distance direction to adapt to the input size of the detection network and ensure target integrity.

[0036] It should also be noted that the type, geometry, and velocity range of each ship target in the dataset are known; the orbital altitude, synthetic aperture time, and detection parameters of the medium-Earth orbit satellites are also known. Each known ship target in each data sample of the dataset is labeled with a truth box that completely covers the ship target and matches the size of the ship target; When training the object detection model using the dataset, the anchor box prior knowledge is used to constrain the predicted boxes generated by the object detection model, and the loss function is updated based on the relationship between the predicted boxes and the corresponding ground truth boxes.

[0037] In some implementations, the loss function is: In the formula, Loss is the loss function; The area of ​​the intersection between the predicted bounding box and the true bounding box; , These are the areas of the predicted bounding box and the true bounding box, respectively. and The width and height of the prediction box; and These are the average width and height of the dynamically changing anchor frame, respectively; where, , ; These are the weighting coefficients.

[0038] This loss function adds a penalty for the deviation of the predicted box width and height from the mean of the prior anchor box sizes, and this is done through weighting coefficients. The importance of equilibrium position regression and size regression.

[0039] In some implementations, such as Figure 2 As shown, the target detection model used in this step includes a backbone network, a neck network, and a head network, set sequentially. Among them, (1) The backbone network is used to extract feature maps of different levels from the distance compression domain data to obtain multi-scale features. The backbone network includes P5, P4, P3, P2, and P1 networks connected in sequence. Among them, the P5 network is used to extract deep features, which have strong semantic information and are suitable for detecting large targets. P4 and P3 are used to extract mid-level features, which take into account both semantics and details and are suitable for medium-sized targets. P2 and P1 are used to extract shallow features, which have rich detail information and are suitable for detecting small targets.

[0040] (2) The neck network is used to fuse and enhance the multi-scale features. The neck network includes two feature concatenation modules (i.e., Cat modules, used for concatenation operations), two multi-branch feature fusion modules (i.e., C2f modules), and two upsample modules.

[0041] (3) The detection head is used to output the predicted bounding box and confidence score of the ship target based on the prior knowledge of the anchor frame. The detection head network includes two feature concatenation modules (i.e., the Cat module, used to perform concatenate operation), two multi-branch feature fusion modules (i.e., the C2f module), two convolutional layers and multiple feature extraction layers P5-P3, which are used for target detection for deep features, mid-level features and shallow features, respectively.

[0042] Of course, the above model structure is only a preferred method. Users can add or remove parts of the network as needed. This application does not impose any specific limitations.

[0043] In addition, when training the object detection model, this application prioritizes recall as the optimization objective, uses a low confidence threshold in the inference stage to ensure high recall, and suppresses false alarms through an anchor box adaptive mechanism.

[0044] Once the target detection model is trained, accurate ship target detection results can be obtained simply by inputting the distance compression domain data to be detected into the model.

[0045] like Figure 3 , Figure 4 As shown, this invention provides a medium-orbit spaceborne SAR ship target detection device based on range compression domain. The device embodiment can be implemented through software, hardware, or a combination of both. From a hardware perspective, as... Figure 3The diagram shown is a hardware architecture diagram of a computing device for a medium-orbit spaceborne SAR ship target detection device based on range compression domain, provided in an embodiment of the present invention. (Except for...) Figure 3 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 4 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.

[0046] Please refer to Figure 4 This invention provides a medium-orbit spaceborne SAR ship target detection device based on range compression domain, used to implement any of the above methods. The device includes: Acquisition unit 400 is used to acquire echo data of the medium-orbit spaceborne SAR ship to be detected; Preprocessing unit 402 is used to perform range-direction matched filtering on the echo data to obtain range-compressed domain data; The detection unit 404 is used to input the range compressed domain data into a pre-trained target detection model to obtain the ship target detection result. The target detection model is built based on an anchor-based structure and trained on the dataset. When generating the prediction box, the model uses the prior knowledge of the anchor box generated based on the medium orbit detection parameters and ship target parameters in the dataset as a constraint condition, and the loss function incorporates the offset penalty constraint between the prediction box size and the anchor box prior knowledge. The dataset consists of SAR range compressed domain data samples generated by various known types of ship targets under known medium orbit satellite detection parameters.

[0047] In some implementations, the prior knowledge of the anchor frame includes: When the ship target is stationary, the minimum basic dimensions of the anchor frame include the basic height and the basic width of the anchor frame; the basic height and the basic width are obtained based on the number of pixels occupied by the stationary ship target in the azimuth and range directions in the range compression domain, respectively. When the ship target is in motion, the dynamic change range of the anchor frame includes the dynamic change range of the anchor frame height and the dynamic change range of the anchor frame width. The minimum value of the dynamic change range of the anchor frame height is the basic height of the anchor frame, and the maximum value is the number of azimuth pixels occupied by the target signal in the range compression domain after considering the maximum azimuth speed of the ship target. The minimum value of the dynamic change range of the anchor frame width is the basic width of the anchor frame, and the maximum value is the number of range pixels occupied by the target signal in the range compression domain after considering the maximum range speed of the ship target. The range of variation of the anchor frame foundation size under different detection perspectives of the medium-orbit satellite includes the range of variation of the anchor frame foundation height and the range of variation of the anchor frame foundation width; the maximum values ​​of the range of variation of the anchor frame foundation height and the anchor frame foundation width are respectively the number of pixels of the stationary ship target in the azimuth and range directions in the range compression domain under the set observation perspective.

[0048] In some implementations, the type, geometry, and velocity range of each ship target in the dataset are known; the orbital altitude, synthetic aperture time, and detection parameters of the medium-Earth orbit satellites are also known. Each known ship target in each data sample of the dataset is labeled with a truth box that completely covers the ship target and matches the size of the ship target; When training the object detection model using the dataset, the anchor box prior knowledge is used to constrain the predicted boxes generated by the object detection model, and the loss function is updated based on the relationship between the predicted boxes and the corresponding ground truth boxes.

[0049] In some implementations, the loss function is: In the formula, Loss is the loss function; The area of ​​the intersection between the predicted bounding box and the true bounding box; , These are the areas of the predicted bounding box and the true bounding box, respectively. and The width and height of the prediction box; and These are the average width and height of the dynamically changing anchor frame, respectively. These are the weighting coefficients.

[0050] In some implementations, the target detection model includes a backbone network, a neck network, and a detection head arranged sequentially. The backbone network is used to extract multi-scale features from distance compressed domain data; The neck network is used to fuse and enhance the multi-scale features; The detection head is used to output the predicted bounding box and confidence level of the ship target based on the prior knowledge of the anchor frame.

[0051] In some implementations, the foundation height of the anchor frame is calculated using the following formula: The base width of the anchor frame is calculated using the following formula: In the formula, , These are the foundation height and foundation width of the anchor frame, respectively. The system pulse repetition frequency; For azimuth synthesis aperture time; The satellite's ground velocity; Wavelength; r The shortest distance between the target and the radar; This refers to the azimuth resolution. This is the radar's equivalent velocity; The range-direction signal sampling rate; This represents the range of variation in signal distance migration. c The speed of sound.

[0052] It should be noted that the medium-orbit spaceborne SAR ship target detection device based on range compression domain 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 medium-orbit spaceborne SAR ship target detection device based on range compression domain provided in the above embodiments and the medium-orbit spaceborne SAR ship target detection method embodiment based on range compression domain belong to the same concept. The specific implementation process is detailed in the method embodiment, and will not be repeated here.

[0053] Embodiments of this application also provide a computer device, please refer to... Figure 4 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 medium-orbit spaceborne SAR ship target detection method based on the range compression domain provided in the above-described method embodiments.

[0054] 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 medium-orbit spaceborne SAR ship target detection method based on the range compression domain provided in the above-described method embodiments.

[0055] 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 any of the distance compression domain-based medium-orbit spaceborne SAR ship target detection methods described in the above embodiments.

[0056] 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.

[0057] 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.

[0058] 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.

[0059] 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 detecting ship targets on a medium-orbit spaceborne SAR system based on range compression domain, characterized in that, The method includes: Acquire echo data of the medium-orbit spaceborne SAR ship to be detected; The echo data is subjected to range-direction matched filtering to obtain range-compressed domain data; The range-compressed domain data is input into a pre-trained target detection model to obtain ship target detection results. The target detection model is built based on an anchor-based structure. When generating prediction boxes, the model uses prior knowledge of anchor boxes generated based on the medium-Earth orbit detection parameters and ship target parameters in the dataset as constraints, and the loss function incorporates a penalty constraint on the offset between the prediction box size and the prior knowledge of the anchor boxes. The dataset consists of SAR range-compressed domain data samples generated by various known types of ship targets under known medium-Earth orbit satellite detection parameters.

2. The method according to claim 1, characterized in that, The prior knowledge of the anchor frame includes: When the ship target is stationary, the minimum basic dimensions of the anchor frame include the basic height and the basic width of the anchor frame; the basic height and the basic width are obtained based on the number of pixels occupied by the stationary ship target in the azimuth and range directions in the range compression domain, respectively. When the ship target is in motion, the dynamic change range of the anchor frame includes the dynamic change range of the anchor frame height and the dynamic change range of the anchor frame width. The minimum value of the dynamic change range of the anchor frame height is the basic height of the anchor frame, and the maximum value is the number of azimuth pixels occupied by the target signal in the range compression domain after considering the maximum azimuth speed of the ship target. The minimum value of the dynamic change range of the anchor frame width is the basic width of the anchor frame, and the maximum value is the number of range pixels occupied by the target signal in the range compression domain after considering the maximum range speed of the ship target. The range of variation of the anchor frame foundation size under different detection perspectives of the medium-orbit satellite includes the range of variation of the anchor frame foundation height and the range of variation of the anchor frame foundation width; the maximum values ​​of the range of variation of the anchor frame foundation height and the anchor frame foundation width are respectively the number of pixels of the stationary ship target in the azimuth and range directions in the range compression domain under the set observation perspective.

3. The method according to claim 1, characterized in that, In the dataset, the type, geometry, and velocity range of each ship target are known; the orbital altitude, synthetic aperture time, and detection parameters of the medium-Earth orbit satellites are also known. Each known ship target in each data sample of the dataset is labeled with a truth box that completely covers the ship target and matches the size of the ship target; When training the object detection model using the dataset, the anchor box prior knowledge is used to constrain the predicted boxes generated by the object detection model, and the loss function is updated based on the relationship between the predicted boxes and the corresponding ground truth boxes.

4. The method according to claim 1, characterized in that, The loss function is: In the formula, Loss is the loss function; The area of ​​the intersection between the predicted bounding box and the true bounding box; , These are the areas of the predicted bounding box and the true bounding box, respectively. and The width and height of the prediction box; and These are the average width and height of the dynamically changing anchor frame, respectively. These are the weighting coefficients.

5. The method according to claim 1, characterized in that, The target detection model includes a backbone network, a neck network, and a detection head arranged sequentially. The backbone network is used to extract multi-scale features from distance compressed domain data; The neck network is used to fuse and enhance the multi-scale features; The detection head is used to output the predicted bounding box and confidence level of the ship target based on the prior knowledge of the anchor frame.

6. The method according to claim 2, characterized in that, The foundation height of the anchor frame is calculated using the following formula: The base width of the anchor frame is calculated using the following formula: In the formula, , These are the foundation height and foundation width of the anchor frame, respectively. The system pulse repetition frequency; For azimuth synthesis aperture time; The satellite's ground velocity; Wavelength; r The shortest distance between the target and the radar; This refers to the azimuth resolution. This is the radar's equivalent velocity; The range-direction signal sampling rate; This represents the range of variation in signal distance migration. c The speed of sound.

7. A medium-orbit spaceborne SAR ship target detection device based on range compression domain, characterized in that, include: The acquisition unit is used to acquire the echo data of the medium-orbit spaceborne SAR ship to be detected; The preprocessing unit is used to perform range-direction matched filtering on the echo data to obtain range-compressed domain data; The detection unit is used to input the distance compression domain data into a pre-trained target detection model to obtain the ship target detection result; The target detection model is built on an anchor-based structure and trained on a dataset. When generating prediction boxes, the model uses prior knowledge of anchor boxes generated based on the mid-orbit detection parameters and ship target parameters in the dataset as constraints. The loss function incorporates a penalty constraint on the offset between the prediction box size and the prior knowledge of the anchor boxes. The dataset consists of SAR range compressed domain data samples generated by various known types of ship targets under known mid-orbit satellite detection parameters.

8. 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 1-6.

9. 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 1-6.

10. A computer program product, characterized in that, Includes a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1-6.

Images