SAR remote sensing image aircraft detection method, device, electronic equipment and medium

Abstract

The embodiments of the present invention relate to the technical fields of computer vision and image processing, and provide a method, device, electronic device and medium for detecting aircraft in SAR remote sensing images. The method comprises: pre-training a model of a multi-detection head of an improved ConvNeXt‑v2 convolutional neural network combined with a Yolov10 target detection algorithm to obtain a SAR remote sensing aircraft detection model; obtaining a target SAR remote sensing image to be detected; inputting the target SAR remote sensing image into the trained SAR remote sensing aircraft detection model, performing aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtaining a target aircraft rotation frame corresponding to the aircraft. Therefore, SAR aircraft detection based on the method of ConvNeXt‑v2 and multi-detection heads effectively improves the detection speed and detection accuracy of aircraft in SAR remote sensing images, which is of great significance for the rapid and accurate extraction of aircraft targets in SAR remote sensing images.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and image processing, and in particular to a SAR remote sensing image aircraft detection method, device, electronic equipment and medium. Background Art

[0002] Synthetic Aperture Radar (SAR) is an active microwave remote sensing imaging system that images surface features by emitting electromagnetic waves and receiving signals reflected by surface targets. Compared with optical remote sensing imaging technology, SAR can provide continuous image acquisition under various complex environmental conditions, including at night and in bad weather. With the continuous development of SAR technology, the diversity and spatial resolution of acquired data have been significantly improved, and sub-meter SAR imaging has been achieved, which greatly meets the needs of refined analysis of ground targets and landforms. SAR technology plays a vital role in many fields including military reconnaissance and surveillance, precision strike guidance, and natural disaster response.

[0003] With the improvement of SAR image resolution and the increase of information contained, image interpretation tasks are gradually developing towards higher-level fine-grained analysis, and the demand for intelligent image processing technology is also growing. With its powerful ability to automatically extract image features, deep learning technology can process complex image data sets without manual feature design, reveal the inherent complex patterns of the data, and show excellent feature expression performance. Compared with traditional image processing methods, the application of deep learning in the field of automatic target recognition (ATR) of SAR images has significantly improved the accuracy of target detection and recognition. Despite this, the task of aircraft target detection and recognition in SAR images still faces many challenges. The structural complexity of aircraft targets causes them to appear as scattered scattering points in SAR images, lacking coherent edge features and complete detail information, which increases the risk of missed detection. In addition, artificial structures in the airport environment, such as bridges and terminals, may generate strong scattering signals, which interfere with the accurate detection and recognition of aircraft targets. Summary of the invention

[0004] The present invention provides a SAR remote sensing image aircraft detection method, device, electronic equipment and medium, which are used to solve the defect of low accuracy in the prior art for SAR remote sensing image aircraft detection, and improve the detection speed and accuracy of aircraft in SAR remote sensing images.

[0005] The present invention provides a SAR remote sensing image aircraft detection method, comprising:

[0006] Acquire SAR remote sensing images of the target to be detected;

[0007] Inputting the target SAR remote sensing image into the trained SAR remote sensing aircraft detection model, performing aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtaining a target aircraft rotation frame corresponding to the aircraft;

[0008] The SAR remote sensing aircraft detection model is trained by the following steps:

[0009] Collect multiple SAR remote sensing images of different types of aircraft and annotate them, and divide the annotated SAR remote sensing images into training set, validation set and test set;

[0010] Based on the training set, the improved ConvNeXt-v2 convolutional neural network combined with the multi-detection head of the Yolov10 target detection algorithm is trained to obtain a trained initial SAR remote sensing image aircraft detection model;

[0011] Verifying the initial SAR remote sensing image aircraft detection model through the verification set;

[0012] When the initial SAR remote sensing image aircraft detection model is verified to be successful, the initial SAR remote sensing image aircraft detection model is tested using the test set, and the initial SAR remote sensing image aircraft detection model that has passed the test is used as the SAR remote sensing aircraft detection model.

[0013] In a possible implementation, the method further includes:

[0014] Extracting different levels of features of the SAR remote sensing images in the training set through the improved ConvNeXt-v2 convolutional neural network;

[0015] The features at different levels are fused by using a BiFPN bidirectional feature pyramid network using a weighted feature fusion and a cross-multi-scale connection method;

[0016] The fused features are input into the input layer corresponding to the multi-detection head of the Yolov10 target detection algorithm for model training to obtain a trained initial SAR remote sensing image aircraft detection model.

[0017] In a possible implementation, the method further includes:

[0018] Input the fused features into the corresponding head structure of the Yolov10 target detection algorithm;

[0019] Presetting the angle prediction branch structure of the Yolov10 target detection algorithm;

[0020] Based on the preset angle prediction branch structure, the aircraft target is identified according to the fused features to obtain the identified aircraft rotation frame;

[0021] When the similarity between the identified aircraft rotation frame and the corresponding aircraft rotation frame marked in the SAR remote sensing image reaches a first preset threshold, a trained initial SAR remote sensing image aircraft detection model is obtained.

[0022] In a possible implementation, the method further includes:

[0023] Inputting the SAR remote sensing images in the verification set into the initial SAR remote sensing image aircraft detection model to verify the initial SAR remote sensing image aircraft detection model;

[0024] When the accuracy of the detection result of the initial SAR remote sensing image aircraft detection model on the verification set reaches a second preset threshold, it is determined that the initial SAR remote sensing image aircraft detection model has passed the verification.

[0025] In a possible implementation, the method further includes:

[0026] Inputting the SAR remote sensing images in the test set into the verified initial SAR remote sensing image aircraft detection model, and calculating the model classification loss function;

[0027] Preset the threshold of online difficult sample mining for classification loss function;

[0028] When the model classification loss function reaches the preset classification loss function online difficult sample mining threshold, it is determined that the initial SAR remote sensing image aircraft detection model test has passed, and the initial SAR remote sensing image aircraft detection model that has passed the test is used as the SAR remote sensing aircraft detection model.

[0029] In a possible implementation, the method further includes:

[0030] The target SAR remote sensing image in which the target aircraft rotation frame is detected is enhanced and output.

[0031] The present invention also provides a SAR remote sensing image aircraft detection device, comprising the following modules:

[0032] An acquisition module is used to acquire a SAR remote sensing image of a target to be detected;

[0033] A detection module is used to input the target SAR remote sensing image into a trained SAR remote sensing aircraft detection model, perform aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtain a target aircraft rotation frame corresponding to the aircraft;

[0034] The model training module is used to collect multiple SAR remote sensing images of different types of aircraft and perform image annotation, and divide the annotated SAR remote sensing images into a training set, a validation set and a test set; based on the training set, the improved ConvNeXt-v2 convolutional neural network combined with the multi-detection head of the Yolov10 target detection algorithm is trained to obtain a trained initial SAR remote sensing image aircraft detection model; the initial SAR remote sensing image aircraft detection model is verified by the validation set; when the initial SAR remote sensing image aircraft detection model is verified, the initial SAR remote sensing image aircraft detection model is tested by the test set, and the initial SAR remote sensing image aircraft detection model that passes the test is used as the SAR remote sensing aircraft detection model.

[0035] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the SAR remote sensing image aircraft detection method as described in any one of the above is implemented.

[0036] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the SAR remote sensing image aircraft detection method as described in any one of the above is implemented.

[0037] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any one of the above-mentioned methods for detecting aircraft in SAR remote sensing images.

[0038] The SAR remote sensing image aircraft detection method, device, electronic equipment and medium provided by the present invention are as follows: acquiring a target SAR remote sensing image to be detected; inputting the target SAR remote sensing image into a trained SAR remote sensing aircraft detection model, performing aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtaining a target aircraft rotation frame corresponding to the aircraft; wherein the SAR remote sensing aircraft detection model is trained through the following steps: collecting a plurality of SAR remote sensing images of different types of aircraft and annotating the images, and dividing the annotated SAR remote sensing images into a training set, a validation set and a test set; based on the training set, model training is performed on a multi-detection head of an improved ConvNeXt-v2 convolutional neural network combined with a Yolov10 target detection algorithm to obtain a trained initial SAR remote sensing image aircraft detection model; the initial SAR remote sensing image aircraft detection model is verified through the validation set; when the initial SAR remote sensing image aircraft detection model is verified, the initial SAR remote sensing image aircraft detection model is tested through the test set, and the initial SAR remote sensing image aircraft detection model that passes the test is used as the SAR remote sensing aircraft detection model. Compared with the existing technology, the structural complexity of aircraft targets leads to missed detection by deep learning detection methods, or the man-made structures in the airport environment generate strong scattering signals, which interferes with the accurate detection and identification of aircraft targets. This solution, based on ConvNeXt-v2 and multi-detection head methods, effectively improves the detection speed and accuracy of aircraft in SAR remote sensing images, which is of great significance for the rapid and accurate extraction of aircraft targets in SAR remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 The present invention provides a flow chart of the SAR remote sensing image aircraft detection method.

[0041] Figure 2 The present invention provides a flow chart of the SAR remote sensing aircraft detection model training method.

[0042] Figure 3 It is a structural schematic diagram of the SAR remote sensing aircraft detection model provided by the present invention.

[0043] Figure 4The present invention is a schematic diagram of the structure of the SAR remote sensing image aircraft detection device provided by the present invention.

[0044] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] To facilitate understanding of the embodiments of the present invention, specific embodiments will be further explained below in conjunction with the accompanying drawings. The embodiments do not constitute a limitation on the embodiments of the present invention.

[0047] Figure 1 FIG. 1 is a flow chart of the SAR remote sensing image aircraft detection method provided by the present invention, such as Figure 1 As shown, the method includes the following:

[0048] S11, obtaining a SAR remote sensing image of a target to be detected.

[0049] In the embodiment of the present invention, the image to be detected is first acquired. This step involves collecting Synthetic Aperture Radar (SAR) remote sensing images required for aircraft detection. These images can come from different SAR sensors, including but not limited to airborne, satellite-borne or ground-based systems. Image acquisition may involve a series of technical activities related to SAR sensor operation, such as flight path planning, imaging mode selection, and data collection timing.

[0050] Furthermore, the acquired SAR remote sensing images of the target to be detected are subjected to image preprocessing: After the image is acquired, some preprocessing is usually required to ensure that the image quality meets the requirements of subsequent processing. This may include denoising, radiation calibration, geometric correction, etc. to eliminate the influence of sensor characteristics and imaging environment on the image.

[0051] S12: input the target SAR remote sensing image into a trained SAR remote sensing aircraft detection model, perform aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtain a target aircraft rotation frame corresponding to the aircraft.

[0052] The pre-processed target SAR image is fed into a pre-trained aircraft detection model. This model is based on deep learning methods such as ConvNeXt-v2 and multi-detection head based models, which can identify and locate aircraft targets in the image.

[0053] The model uses its trained feature extraction and target recognition capabilities to analyze the input SAR image, identify the aircraft targets in the image, and generate a rotation box for each detected aircraft target. The rotation box is a bounding box that can more accurately describe the posture and position of the aircraft in the image. It includes not only position information but also rotation angle information.

[0054] Furthermore, the model outputs the recognition results, including information such as the position, size, and rotation angle of each detected aircraft target, which is usually expressed in the coordinate form of a rotating box.

[0055] S13, performing image enhancement processing on the target SAR remote sensing image in which the target aircraft rotation frame is detected and outputting the image.

[0056] In the embodiment of the present invention, the SAR image in which the aircraft target has been detected can also be further enhanced. The purpose of image enhancement is to improve the visual effect of the image, enhance specific features, or provide higher quality image data for subsequent analysis and processing. The enhancement process can include contrast enhancement, color correction, edge enhancement, etc.

[0057] According to the characteristics of SAR images, specific filters can also be applied for enhancement processing, such as mean filter, Gaussian filter, median filter and Laplace filter, to reduce image noise or enhance image edge information.

[0058] Furthermore, the enhanced image is output and can be used for further analysis, archiving or display. The enhanced image not only improves the visual effect, but also may improve the accuracy and reliability of target recognition in subsequent processing steps.

[0059] The SAR remote sensing image aircraft detection method provided by the present invention comprises the following steps: acquiring a target SAR remote sensing image to be detected; inputting the target SAR remote sensing image into a trained SAR remote sensing aircraft detection model, performing aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtaining a target aircraft rotation frame corresponding to the aircraft; wherein the SAR remote sensing aircraft detection model is trained through the following steps: collecting a plurality of SAR remote sensing images of different types of aircraft and annotating the images, and dividing the annotated SAR remote sensing images into a training set, a validation set and a test set; performing model training on a multi-detection head of an improved ConvNeXt-v2 convolutional neural network combined with a Yolov10 target detection algorithm based on the training set, and obtaining a trained initial SAR remote sensing image aircraft detection model; verifying the initial SAR remote sensing image aircraft detection model through the validation set; when the initial SAR remote sensing image aircraft detection model is verified, testing the initial SAR remote sensing image aircraft detection model through the test set, and using the initial SAR remote sensing image aircraft detection model that passes the test as the SAR remote sensing aircraft detection model. Compared with the existing technology, the structural complexity of aircraft targets leads to missed detection by deep learning detection methods, or the man-made structures in the airport environment generate strong scattering signals, which interferes with the accurate detection and identification of aircraft targets. This method, based on ConvNeXt-v2 and multi-detection head methods, effectively improves the detection speed and accuracy of aircraft in SAR remote sensing images, which is of great significance for the rapid and accurate extraction of aircraft targets in SAR remote sensing images.

[0060] Figure 2 FIG. 1 is a flow chart of the SAR remote sensing aircraft detection model training method provided by the present invention, such as Figure 2 As shown, the method includes the following:

[0061] S21. Collect multiple SAR remote sensing images of different types of aircraft and annotate them, and divide the annotated SAR remote sensing images into a training set, a validation set, and a test set.

[0062] Collect SAR remote sensing images from various types of aircraft. These images may come from different SAR sensors and platforms, such as satellites, aircraft, or drones. When collecting images, it is necessary to ensure that different types of aircraft are covered to increase the diversity and generalization ability of model training.

[0063] The collected SAR images are annotated, including information such as the position, size, and attitude of the aircraft. Annotation usually involves drawing a bounding box or a rotation box to locate the aircraft target, and may include a class label for marking the aircraft. This step is a key step in supervised learning because the model needs this annotation information to learn how to identify and locate aircraft.

[0064] The annotated SAR remote sensing images are divided into three parts according to the preset ratio: training set, validation set and test set. Usually, the ratio of these data sets may be 70% for training, 15% for validation, and 15% for testing, but the specific ratio can be adjusted according to the actual situation.

[0065] S22. Based on the training set, a model is trained on the improved ConvNeXt-v2 convolutional neural network combined with the multi-detection head of the Yolov10 target detection algorithm to obtain a trained initial SAR remote sensing image aircraft detection model.

[0066] The improved ConvNeXt-v2 convolutional neural network is used as the backbone network, combined with the multi-detection head architecture of the Yolov10 target detection algorithm. ConvNeXt-v2 is an efficient convolutional neural network, and Yolov10 is a real-time target detection algorithm. The combination of the two aims to improve the detection speed and accuracy of the model.

[0067] The network is trained using the training set. During the training process, the network learns feature representations through forward propagation and adjusts weights through backpropagation to minimize the loss function. This step may involve the adjustment of hyperparameters, optimization of learning rates, and the application of regularization techniques to prevent overfitting. After the training is completed, an initial SAR remote sensing image aircraft detection model is obtained, which can identify and locate aircraft targets in SAR images. The structure of the model is as follows Figure 3 shown.

[0068] S23. Verifying the initial SAR remote sensing image aircraft detection model through the verification set.

[0069] Use the validation set to validate the trained model. The purpose of this step is to evaluate the performance of the model on unseen data and adjust the model parameters to optimize performance. The performance of the model is evaluated by calculating indicators such as precision, recall, and F1 score. If the model does not perform well on the validation set, it may be necessary to return to the training step for further adjustments and optimization.

[0070] S24. When the initial SAR remote sensing image aircraft detection model is verified to be successful, the initial SAR remote sensing image aircraft detection model is tested using the test set, and the initial SAR remote sensing image aircraft detection model that has passed the test is used as the SAR remote sensing aircraft detection model.

[0071] If the model performs as expected on the validation set, the test set is then used for a final test of the model. The test set provides an independent dataset for evaluating the performance of the model in real applications. The final performance of the model is evaluated on the test set to ensure that the model has sufficient generalization ability. The test results will determine whether the model is ready to be deployed in real applications. After the test passes, the initial SAR remote sensing image aircraft detection model is used as the final SAR remote sensing aircraft detection model for actual aircraft detection tasks.

[0072] The model training steps of the embodiment of the present invention are described in detail below:

[0073] Step 1: Preprocess the SAR remote sensing images and annotate them into data sets. Divide the rotating box detection targets of different aircraft categories according to the preset ratio and randomly divide them into training set, validation set and test set. The ratio of training set, validation set and test set can be 8:1:1.

[0074] Step 2: Extract feature branches P2, P3, and P4 layers from stage2, stage3, and stage4 of ConvNeXt-v2 respectively, and fuse P2, P3, and P4 feature layers through BiFPN feature pyramid network using weighted feature fusion and cross-scale connection and convert them into P3, P4, and P5 layers for detection head input. Considering the balance between accuracy and efficiency, ConvNeXt-v2-base model is selected as the backbone network, and BiFPN is selected as the Neck network structure.

[0075] Furthermore, the ConvNeXt-v2-base model consists of four stages. The number of ConvNeXt-v2 Blocks in each stage are 3, 3, 27, and 3 respectively, and the dimensions of each Block are 96, 192, 384, and 768 respectively.

[0076] Furthermore, the BiFPN feature pyramid network uses weighted feature fusion and cross-scale connections to fuse the P2, P3, and P4 feature layers and convert them into the P3, P4, and P5 layers of the detection head input. The specific structure is as follows: The BiFPN feature pyramid is divided into two stages: upsampling weighted fusion and downsampling weighted fusion. The P2, P3, and P4 feature layers and each level of the two stages are weightedly connected to each other through skip connections, and C2fCIB is added to enhance the fusion features of upsampling and concat. It is used to capture the dependency between different features and improve the accuracy of feature extraction.

[0077] Furthermore, upsampling weighted fusion is performed to obtain the aggregated feature map by weighting the Pi feature layer and the Pi+1 feature layer. The formula is as follows:

[0078]

[0079] in, is the up-sampled aggregate feature, is the weight, is the weight, A non-zero decimal number to prevent the denominator from being 0.

[0080] Downsampling weighted fusion is Pi feature layer, Pi+1 feature layer and Pi-1 feature layer to obtain the aggregate feature map through weighting. The formula is as follows:

[0081]

[0082] in, is the down-sampling aggregation feature, is the weight, is the weight, is the weight.

[0083] Step 3: Combined with the NMS-free detection head of Yolov10, the P3, P4, and P5 layers are concatenated and input into the one-to-manyHead and one-to-one Head to identify SAR aircraft targets. The one-to-many Head only participates in the training process, not in evaluation and reasoning.

[0084] Yolov10's NMS-free detection head is a target detection head that does not rely on NMS. It uses one-to-many Head and one-to-one Head as the Head network structure to improve the accuracy and generalization of the model during training. Only one-to-one Head is used in the evaluation and reasoning detection stages to improve the speed of model detection.

[0085] Furthermore, the NMS-free detection head of the original Yolov10 is a positive box detection head and cannot train and predict rotated box targets. Therefore, an angle prediction branch is added to the detection head. The angle prediction branch structure is two 3×3 convolutions and one 1×1 convolution.

[0086] Step 4: Use the training set and validation set to train the network model, and use the test data for testing and verification to obtain the SAR remote sensing aircraft detection model.

[0087] The OHEM online difficult sample mining strategy is used for training difficult samples. The classification loss function calculation formula is as follows:

[0088]

[0089] in is the probability distribution, is the probability distribution of the true label, is the probability distribution of the predicted label, For a certain category, for Category The prediction probability of each data set is N, the batch size is N, and threshold is the threshold for mining difficult samples, which is generally set to 0.7.

[0090] Compared with the prior art, the SAR aircraft detection method based on ConvNeXt-v2 and multiple detection heads in the embodiment of the present invention comprehensively considers the balance between accuracy and efficiency, selects the ConvNeXt-v2-base model as the backbone network, has more efficient feature extraction capability, a simpler model structure and stronger performance improvement, and selects the NMS-free detection heads of BiFPN and Yolov10 as the Neck network structure and Head structure respectively, which enhances the model feature fusion capability, significantly improves the speed of aircraft target detection, and meets the application scenarios with high real-time requirements; greatly improves the detection accuracy, and reduces the probability of false detection and missed detection.

[0091] The SAR remote sensing image aircraft detection device provided by the present invention is described below. The SAR remote sensing image aircraft detection device described below and the SAR remote sensing image aircraft detection method described above can be referenced to each other.

[0092] Figure 4 This is a schematic diagram of the structure of the SAR remote sensing image aircraft detection device provided by the present invention. Specifically comprising:

[0093] The acquisition module 401 is used to acquire a SAR remote sensing image of a target to be detected. For detailed description, please refer to the relevant description corresponding to the above method embodiment, which will not be repeated here.

[0094] The detection module 402 is used to input the target SAR remote sensing image into the trained SAR remote sensing aircraft detection model, perform aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtain a target aircraft rotation frame corresponding to the aircraft. For detailed description, please refer to the relevant description corresponding to the above method embodiment, which will not be repeated here.

[0095] The model training module 403 is used to collect multiple SAR remote sensing images of different types of aircraft and perform image annotation, and divide the annotated SAR remote sensing images into a training set, a validation set and a test set; based on the training set, the improved ConvNeXt-v2 convolutional neural network combined with the multi-detection head of the Yolov10 target detection algorithm is trained to obtain a trained initial SAR remote sensing image aircraft detection model; the initial SAR remote sensing image aircraft detection model is verified by the validation set; when the initial SAR remote sensing image aircraft detection model is verified, the initial SAR remote sensing image aircraft detection model is tested by the test set, and the initial SAR remote sensing image aircraft detection model that passes the test is used as the SAR remote sensing aircraft detection model. For detailed descriptions, please refer to the relevant descriptions corresponding to the above method embodiments, which will not be repeated here.

[0096] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 810, a communications interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communications interface 820 and the memory 830 communicate with each other via the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the SAR remote sensing image aircraft detection method, the method comprising: obtaining a target SAR remote sensing image to be detected; inputting the target SAR remote sensing image into a trained SAR remote sensing aircraft detection model, performing aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtaining a target aircraft rotation frame corresponding to the aircraft; wherein the SAR remote sensing aircraft detection model is trained by the following steps: collecting multiple SAR remote sensing images of different types of aircraft and annotating the images, dividing the annotated SAR remote sensing images into The invention relates to a training set, a validation set and a test set; based on the training set, the improved ConvNeXt-v2 convolutional neural network combined with the multi-detection head of the Yolov10 target detection algorithm is trained to obtain the trained initial SAR remote sensing image aircraft detection model; the initial SAR remote sensing image aircraft detection model is verified by the validation set; when the initial SAR remote sensing image aircraft detection model is verified, the initial SAR remote sensing image aircraft detection model is tested by the test set, and the initial SAR remote sensing image aircraft detection model that passes the test is used as the SAR remote sensing aircraft detection model.

[0097] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0098] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the SAR remote sensing image aircraft detection method provided by the above methods, the method comprising: obtaining a target SAR remote sensing image to be detected; inputting the target SAR remote sensing image into a trained SAR remote sensing aircraft detection model, performing aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtaining a target aircraft rotation frame corresponding to the aircraft; wherein the SAR remote sensing aircraft detection model is trained by the following steps: collecting multiple images of different types; The invention discloses a method for obtaining a SAR remote sensing image of a type of aircraft and annotating the image, dividing the annotated SAR remote sensing image into a training set, a validation set and a test set; based on the training set, model training is performed on an improved ConvNeXt-v2 convolutional neural network combined with a multi-detection head of a Yolov10 target detection algorithm to obtain a trained initial SAR remote sensing image aircraft detection model; the initial SAR remote sensing image aircraft detection model is verified by the validation set; when the initial SAR remote sensing image aircraft detection model is verified, the initial SAR remote sensing image aircraft detection model is tested by the test set, and the initial SAR remote sensing image aircraft detection model that passes the test is used as the SAR remote sensing aircraft detection model.

[0099] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the SAR remote sensing image aircraft detection method provided by the above-mentioned methods, the method comprising: acquiring a target SAR remote sensing image to be detected; inputting the target SAR remote sensing image into a trained SAR remote sensing aircraft detection model, performing aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtaining a target aircraft rotation frame corresponding to the aircraft; wherein the SAR remote sensing aircraft detection model is trained by the following steps: acquiring multiple SAR remote sensing images of different types of aircraft and performing image detection; The invention discloses a method for annotating SAR remote sensing images, dividing the annotated SAR remote sensing images into a training set, a validation set and a test set; based on the training set, a multi-detection head of an improved ConvNeXt-v2 convolutional neural network combined with a Yolov10 target detection algorithm is trained to obtain a trained initial SAR remote sensing image aircraft detection model; the initial SAR remote sensing image aircraft detection model is verified by the validation set; when the initial SAR remote sensing image aircraft detection model is verified, the initial SAR remote sensing image aircraft detection model is tested by the test set, and the initial SAR remote sensing image aircraft detection model that passes the test is used as the SAR remote sensing aircraft detection model.

[0100] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0101] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting aircraft in SAR remote sensing images, characterized in that: include: Acquire SAR remote sensing images of the target to be detected; Inputting the target SAR remote sensing image into the trained SAR remote sensing aircraft detection model, performing aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtaining a target aircraft rotation frame corresponding to the aircraft; The SAR remote sensing aircraft detection model is trained by the following steps: Collect multiple SAR remote sensing images of different types of aircraft and annotate them, and divide the annotated SAR remote sensing images into training set, validation set and test set; Based on the training set, the improved ConvNeXt-v2 convolutional neural network combined with the multi-detection head of the Yolov10 target detection algorithm is trained to obtain a trained initial SAR remote sensing image aircraft detection model; The improved ConvNeXt-v2 convolutional neural network combined with the multi-detection head of the Yolov10 target detection algorithm is trained based on the training set to obtain the trained initial SAR remote sensing image aircraft detection model, including: Extracting different levels of features of the SAR remote sensing images in the training set through the improved ConvNeXt-v2 convolutional neural network; The features at different levels are fused by using a BiFPN bidirectional feature pyramid network using a weighted feature fusion and a cross-multi-scale connection method; The fused features are input into the input layer corresponding to the multi-detection head of the Yolov10 target detection algorithm for model training to obtain a trained initial SAR remote sensing image aircraft detection model; Verifying the initial SAR remote sensing image aircraft detection model through the verification set; When the initial SAR remote sensing image aircraft detection model is verified to be successful, the initial SAR remote sensing image aircraft detection model is tested using the test set, and the initial SAR remote sensing image aircraft detection model that has passed the test is used as the SAR remote sensing aircraft detection model.

2. The method according to claim 1, characterized in that The fused features are input into the input layer corresponding to the multi-detection head of the Yolov10 target detection algorithm for model training to obtain the trained initial SAR remote sensing image aircraft detection model, including: Input the fused features into the corresponding head structure of the Yolov10 target detection algorithm; Presetting the angle prediction branch structure of the Yolov10 target detection algorithm; Based on the preset angle prediction branch structure, the aircraft target is identified according to the fused features to obtain the identified aircraft rotation frame; When the similarity between the identified aircraft rotation frame and the corresponding aircraft rotation frame marked in the SAR remote sensing image reaches a first preset threshold, a trained initial SAR remote sensing image aircraft detection model is obtained.

3. The method according to claim 2, characterized in that The verifying the initial SAR remote sensing image aircraft detection model by using the verification set includes: Inputting the SAR remote sensing images in the verification set into the initial SAR remote sensing image aircraft detection model to verify the initial SAR remote sensing image aircraft detection model; When the accuracy of the detection result of the initial SAR remote sensing image aircraft detection model on the verification set reaches a second preset threshold, it is determined that the initial SAR remote sensing image aircraft detection model has passed the verification.

4. The method according to claim 3, characterized in that When the initial SAR remote sensing image aircraft detection model is verified to be passed, the initial SAR remote sensing image aircraft detection model is tested by using the test set, and the initial SAR remote sensing image aircraft detection model that passes the test is used as the SAR remote sensing aircraft detection model, including: Inputting the SAR remote sensing images in the test set into the verified initial SAR remote sensing image aircraft detection model, and calculating the model classification loss function; Preset the threshold of online difficult sample mining for classification loss function; When the model classification loss function reaches the preset classification loss function online difficult sample mining threshold, it is determined that the initial SAR remote sensing image aircraft detection model test has passed, and the initial SAR remote sensing image aircraft detection model that has passed the test is used as the SAR remote sensing aircraft detection model.

5. The method according to claim 1, characterized in that The target SAR remote sensing image is input into the trained SAR remote sensing aircraft detection model, and aircraft detection is performed on the target SAR remote sensing image by the SAR remote sensing aircraft detection model to obtain a target aircraft rotation frame corresponding to the aircraft, including: Perform image enhancement processing on the target SAR remote sensing image where the target aircraft rotation frame is detected and output; The SAR remote sensing aircraft detection model is optimized based on the target aircraft rotation frame.

6. A SAR remote sensing image aircraft detection device, characterized in that: include: An acquisition module is used to acquire a SAR remote sensing image of a target to be detected; A detection module is used to input the target SAR remote sensing image into a trained SAR remote sensing aircraft detection model, perform aircraft detection on the target SAR remote sensing image through the SAR remote sensing aircraft detection model, and obtain a target aircraft rotation frame corresponding to the aircraft; A model training module is used to collect multiple SAR remote sensing images of different types of aircraft and perform image annotation, and divide the annotated SAR remote sensing images into a training set, a validation set and a test set; based on the training set, an improved ConvNeXt-v2 convolutional neural network combined with a multi-detection head of a Yolov10 target detection algorithm is model trained to obtain a trained initial SAR remote sensing image aircraft detection model; the model training based on the training set of the improved ConvNeXt-v2 convolutional neural network combined with a multi-detection head of a Yolov10 target detection algorithm to obtain a trained initial SAR remote sensing image aircraft detection model includes: extracting different levels of features of the SAR remote sensing images in the training set through the improved ConvNeXt-v2 convolutional neural network; The features at different levels are fused by using BiFPN bidirectional feature pyramid network with weighted feature fusion and cross-multi-scale connection method; the fused features are input into the input layer corresponding to the multi-detection head of the Yolov10 target detection algorithm for model training to obtain the trained initial SAR remote sensing image aircraft detection model; the initial SAR remote sensing image aircraft detection model is verified by the verification set; when the initial SAR remote sensing image aircraft detection model is verified, the initial SAR remote sensing image aircraft detection model is tested by the test set, and the initial SAR remote sensing image aircraft detection model that passes the test is used as the SAR remote sensing aircraft detection model.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the SAR remote sensing image aircraft detection method according to any one of claims 1 to 5 is implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the SAR remote sensing image aircraft detection method according to any one of claims 1 to 5 is implemented.

9. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the SAR remote sensing image aircraft detection method according to any one of claims 1 to 5 is implemented.

Images