Method and device for detecting aircraft target in SAR remote sensing image

Abstract

The embodiment of the invention relates to the technical field of SAR remote sensing aircraft detection, and provides a method and device for detecting an aircraft target in an SAR remote sensing image, and the method comprises the steps: obtaining a to-be-detected target SAR remote sensing image; inputting the target SAR remote sensing image into a pre-trained SAR remote sensing aircraft detection model, and extracting target multi-scale local features of a target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model; performing feature aggregation on the target multi-scale local features to obtain target interaction information between the local information and the global information of the target aircraft; and carrying out aircraft target detection on the SAR remote sensing image of the target based on the target interaction information. Therefore, global view angle information is fully captured, the influence caused by aircrafts with different sizes is reduced, and the detection precision and efficiency of targets such as different types of aircrafts in the SAR remote sensing image are improved.

Description

Technical Field

[0001] The invention relates to the technical field of SAR remote sensing aircraft detection, and in particular to a method and a device for detecting aircraft targets in SAR remote sensing images. Background Art

[0002] Synthetic Aperture Radar (SAR) images are not limited by weather conditions and day and night time. They can obtain data in a variety of complex environments and extract useful information about ground targets from complex SAR image data to achieve accurate target detection, playing a vital role in both military and non-military fields. Early SAR aircraft target detection technology mainly emphasized manual feature extraction and processing classifiers, which was difficult to apply on a large scale. Therefore, how to accurately detect aircraft targets in SAR images has always been a research hotspot.

[0003] Early aircraft target detection methods focused on simple machine learning methods. For example, features were extracted using filtering, PCA principal component analysis technology, directional gradient histogram and other methods, and features were classified using Adaboost, KNN, SVM and other methods. After feature extraction and classification operations on the candidate area, it is determined whether the target exists. Similarly, the CFAR method uses a threshold to distinguish the target from noise, clutter and interference to search for the target. Since then, a series of improved CFAR methods have been widely proposed by researchers. For example, Ai J et al. used an adaptive truncated clutter statistical model based on the log-normal distribution model to improve the performance of the TPCFAR detector. Hou B et al. developed a new multi-layer CFAR target detector to alleviate the errors caused by the loss of detail information. Although these traditional machine learning methods can effectively detect targets, they still have problems such as poor robustness, low detection accuracy and weak generalization ability. At the same time, these target detection methods mostly use manual participation in feature design and screening, and the detection accuracy is still limited.

[0004] In recent years, with the rapid development of deep learning technology, it has been widely used in SAR remote sensing target detection and recognition tasks. However, in SAR images with complex backgrounds, it is not easy to extract features of small aircraft targets, and the scattered points are more discrete, making detection more difficult. Therefore, many researchers have designed a variety of solutions to address these limitations and further improve the detection accuracy of SAR targets. For example, Wang Sy et al. designed a new LeNet-5 detection framework and used data enhancement methods to achieve rough and rapid positioning of candidate aircraft targets in large-scale SAR scenarios. Li M et al. proposed a lightweight detection model (LDM), which mainly includes a reuse module (RB) and an information correction module (ICB) based on the Yolov3 framework. The RB module helps the neural network extract rich aircraft features by aggregating multi-layer information, and helps suppress interference and redundant information in complex environments by extracting grayscale features and enhancing spatial information. Zhao S et al. proposed a Res-Clo network to denoise SAR images as a preprocessing step to improve detection accuracy. Subsequently, an improved network DML-YoloV8 was designed based on the YoloV8 network. In the feature extraction layer, a designed MFB module was integrated to effectively broaden the network's receptive field. At the same time, deformable convolution was introduced in the feature fusion layer to enhance the network's multi-scale detection capability. At present, the SAR aircraft target detection algorithms at home and abroad have improved the detection accuracy to a certain extent, but the model complexity is large and the generalization ability is weak.

[0005] The main problems in aircraft target detection in SAR remote sensing images are as follows: The content of SAR remote sensing images is relatively complex and has great semantic ambiguity. In addition, different aircraft targets in SAR remote sensing images have similar appearances such as shape and color. This situation can easily lead to misclassification of simple aircraft target detection methods, that is, SAR aircraft targets cannot be directly located and detected by their physical appearance. At the same time, a large amount of noise and false information poses a huge challenge to the accurate detection of aircraft targets.

[0006] The scales of different targets in SAR images span a large range, and a single-scale feature extractor cannot obtain effective global and contextual information from SAR images. Similarly, when using a simple feature extractor to obtain deep features of aircraft targets, redundant information is reused multiple times, which reduces the representation of local detail features and ultimately affects the detection performance of aircraft targets in SAR images. Summary of the invention

[0007] The present invention provides a method and device for detecting aircraft targets in SAR remote sensing images, so as to solve the defects of low accuracy and great difficulty in detecting aircraft targets in SAR remote sensing images in the prior art, and to improve the detection accuracy and efficiency of targets such as different types of aircraft in SAR remote sensing images.

[0008] The present invention provides a method for detecting aircraft targets in SAR remote sensing images, comprising the following steps.

[0009] Acquire SAR remote sensing images of the target to be detected; Inputting the target SAR remote sensing image into a pre-trained SAR remote sensing aircraft detection model, and extracting target multi-scale local features of the target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model; Performing feature aggregation on the multi-scale local features of the target to obtain target interaction information between local and global information of the target aircraft; Aircraft target detection is performed on the target SAR remote sensing image based on the target interaction information, and the detection result includes target frame regression and aircraft classification.

[0010] In a possible implementation, the method further includes: The SAR remote sensing aircraft detection model is trained based on the following method: Collect SAR remote sensing images from different types of aircraft and perform image processing; The SAR remote sensing images after image processing are divided into a model training set, a model verification set and a model test set; The improved YoloV8 network is trained by using the model training set and the model validation set; The trained YoloV8 network is tested by the model test set. When the test result meets the preset conditions, it is determined that the training of the improved YoloV8 network is completed, and the SAR remote sensing aircraft detection model is obtained.

[0011] In a possible implementation, the method further includes: Acquire multi-scale local features of aircraft in the SAR remote sensing image in the model training set through the improved YoloV8 network; Based on the SAR remote sensing images in the model training set, an improved SPPF_AA component is used in the improved YoloV8 network to model global semantics and target edge details to obtain an initial model; Aggregating the multi-scale local features layer by layer in the initial model to obtain interactive information between local and global information; Based on the interactive information, a rotating detection head is used to perform aircraft frame regression and aircraft target classification; When the accuracy of the regression of the aircraft frame and the classification result of the aircraft target by the initial model reaches a first preset threshold, the initial model is verified by the model verification set.

[0012] In a possible implementation, the method further includes: Squeezing the SAR remote sensing images in the model training set through a convolutional layer; The squeezed feature map is passed through multiple maximum pooling operations and residual structures; Using a global average pooling layer and a global maximum pooling layer to obtain global and contextual semantic information, and fusing the global and contextual semantic information with edge detail information; Based on the information fusion results, global semantics and target edge details are modeled to obtain the initial model.

[0013] In a possible implementation, the method further includes: When the initial model is verified, the SAR remote sensing images in the model test set are input into the verified initial model to test the model; When the test result is that the detection accuracy of the initial model for the model test set reaches a second preset threshold, it is determined that the improved YoloV8 network training is completed, and the SAR remote sensing aircraft detection model is obtained.

[0014] In a possible implementation, the method further includes: The SAR remote sensing image is subjected to image processing to obtain a processed SAR remote sensing image, wherein the image processing includes random translation, rotation, color conversion, normalization and flipping processing operations.

[0015] The present invention also provides an aircraft target detection device for SAR remote sensing images, comprising the following modules: An acquisition module is used to acquire a SAR remote sensing image of a target to be detected; A feature extraction module is used to input the target SAR remote sensing image into a pre-trained SAR remote sensing aircraft detection model, and extract the target multi-scale local features of the target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model; A feature aggregation module is used to perform feature aggregation on the multi-scale local features of the target to obtain target interaction information between the local and global information of the target aircraft; The detection module is used to perform aircraft target detection on the target SAR remote sensing image based on the target interaction information, and the detection result includes target frame regression and aircraft classification.

[0016] 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 method for detecting aircraft targets in SAR remote sensing images as described in any one of the above is implemented.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for detecting aircraft targets in SAR remote sensing images as described in any one of the above is implemented.

[0018] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for detecting aircraft targets in SAR remote sensing images as described in any one of the above is implemented.

[0019] The present invention provides a method and device for detecting aircraft targets in SAR remote sensing images, which are as follows: obtaining a target SAR remote sensing image to be detected; inputting the target SAR remote sensing image into a pre-trained SAR remote sensing aircraft detection model, extracting the target multi-scale local features of the target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model; performing feature aggregation on the target multi-scale local features to obtain target interaction information between the local and global information of the target aircraft; performing aircraft target detection on the target SAR remote sensing image based on the target interaction information, and the detection results include target frame regression and aircraft classification. Compared with the defects of low accuracy and high detection difficulty of SAR remote sensing image aircraft target detection in the prior art, this solution fully captures global perspective information and reduces the impact of aircraft of different sizes, thereby improving the detection accuracy and efficiency of different types of aircraft and other targets in SAR remote sensing images. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] 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.

[0021] Figure 1 The present invention is a flowchart of a method for detecting aircraft targets in SAR remote sensing images.

[0022] Figure 2 This is one of the flow charts of the training method of the SAR remote sensing aircraft detection model provided by the present invention.

[0023] Figure 3 This is the second flow chart of the training method of the SAR remote sensing aircraft detection model provided by the present invention.

[0024] Figure 4 The present invention is a schematic diagram of the structure of an aircraft target detection device in a SAR remote sensing image.

[0025] Figure 5 The figure is a schematic diagram of the network structure of the SAR remote sensing aircraft detection model provided by the present invention.

[0026] Figure 6 It is a schematic diagram of the detection results of the SAR remote sensing aircraft detection model provided by the present invention.

[0027] Figure 7 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0028] 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.

[0029] 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.

[0030] Figure 1 FIG. 1 is a flow chart of a method for detecting aircraft targets in SAR remote sensing images provided by the present invention. Figure 1 As shown, the method specifically includes: S11, obtaining a SAR remote sensing image of a target to be detected.

[0031] Synthetic Aperture Radar (SAR) remote sensing technology is a remote sensing technology that uses microwaves for imaging. It has all-weather and all-day observation capabilities and is particularly suitable for the detection of targets such as aircraft.

[0032] In the embodiment of the present invention, first, it is necessary to obtain the SAR remote sensing image to be detected from the SAR remote sensing data source. These data sources may include satellites, unmanned aerial vehicles or ground SAR systems. The acquired image has high resolution and can clearly display targets such as aircraft. The acquired SAR remote sensing image may be in a specific image format, resolution, and size to ensure smooth subsequent processing.

[0033] S12, inputting the target SAR remote sensing image into a pre-trained SAR remote sensing aircraft detection model, and extracting target multi-scale local features of the target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model.

[0034] The SAR remote sensing aircraft detection model is a deep learning-based model that can extract aircraft features from SAR remote sensing images for aircraft target detection through training. The specific training process is described in Figure 2 and Figure 3 The corresponding embodiments are described in detail and will not be described in detail here.

[0035] Then, the acquired target SAR remote sensing image is input into the pre-trained SAR remote sensing aircraft detection model. This model is usually trained on a large number of SAR remote sensing image data sets and has high detection accuracy. After the image is input, the model will automatically extract the target aircraft features in the image, which may include the shape, size, texture, etc. of the aircraft. The extracted features will be used for the subsequent extraction of multi-scale local features of the target.

[0036] S13, performing feature aggregation on the target multi-scale local features to obtain target interaction information between the local and global information of the target aircraft.

[0037] Feature aggregation is a method of fusing multiple features to extract higher-level features of the target aircraft.

[0038] First, the extracted multi-scale local features of the target need to be screened and integrated to remove redundant and noisy features.

[0039] Then, these features are fused using feature aggregation methods to obtain the interactive information between the local and global information of the target aircraft. This interactive information may include the relative position relationship between the various parts of the aircraft, texture changes, etc.

[0040] Feature aggregation methods may include pooling layers in convolutional neural networks (CNNs), attention mechanisms, etc.

[0041] S14. Performing aircraft target detection on the target SAR remote sensing image based on the target interaction information, wherein the detection result includes target frame regression and aircraft classification.

[0042] Target box regression and classification are two core tasks in target detection, which are used to determine the location and category of the target respectively.

[0043] Based on the extracted target interaction information, the target box regression method is used to determine the position and size of the target aircraft. This is usually achieved by predicting the four boundary coordinates of the target box. At the same time, the classification method is used to determine the category of the target aircraft. The probability values ​​of different categories can be calculated, and the category with the highest probability value can be selected as the final classification result. In practical applications, target box regression and classification are usually performed simultaneously, and the detection accuracy can be improved through joint optimization. Figure 6 A schematic diagram of the test results.

[0044] The above steps constitute a complete SAR remote sensing aircraft detection process. By acquiring the target SAR remote sensing image to be detected, inputting it into the pre-trained model to extract features, performing feature aggregation, and performing target box regression and classification based on interactive information, accurate detection of aircraft in SAR remote sensing images can be achieved. This process has broad application prospects in military reconnaissance, civil aviation and other fields.

[0045] The present invention provides a method for detecting aircraft targets in SAR remote sensing images, which comprises the following steps: obtaining a target SAR remote sensing image to be detected; inputting the target SAR remote sensing image into a pre-trained SAR remote sensing aircraft detection model, extracting target multi-scale local features of the target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model; performing feature aggregation on the target multi-scale local features to obtain target interaction information between local and global information of the target aircraft; and performing aircraft target detection on the target SAR remote sensing image based on the target interaction information, and the detection results include target frame regression and aircraft classification. Compared with the defects of low accuracy and high detection difficulty of SAR remote sensing image aircraft target detection in the prior art, the present method fully captures global perspective information and reduces the impact of aircraft of different sizes, thereby improving the detection accuracy and efficiency of different types of aircraft and other targets in SAR remote sensing images.

[0046] Figure 2 It is one of the flow charts of the training method of the SAR remote sensing aircraft detection model provided by the present invention, such as Figure 2 As shown, the method specifically includes: S21. Collect SAR remote sensing images of different types of aircraft and perform image processing.

[0047] Collect SAR remote sensing image datasets of different types of aircraft and aircraft in different postures (take-off, landing, cruising, etc.). For example, there are 1,000 images in total, including 7 categories: Boeing787, A220, ARJ21, A330, A320 / 321, Boeing737-800 and others, and the data categories are unbalanced.

[0048] The collected original SAR images may contain interference information such as noise and clutter, and preprocessing is required to improve the image quality. In order to improve the detection accuracy of SAR aircraft targets and increase the number of samples, multiple preprocessing operations such as random translation, rotation, color transformation, normalization and flipping are used. Secondly, in order to further improve the detection performance, the larger size images are cropped to 1024 × 1024 size according to the cropping step r = 500. This processing method can ensure the influence of different image sizes.

[0049] S22. Divide the processed SAR remote sensing images into a model training set, a model verification set and a model test set.

[0050] In order to ensure the smooth progress of the experiment in the embodiment of the present invention and to demonstrate the superiority of the SAR aircraft target detection framework on small samples, all samples are divided into three parts: training, testing and validation sets. 400 and 100 SAR remote sensing images after image processing are randomly selected as training and validation samples, and the remaining 500 are test samples.

[0051] S23, training the improved YoloV8 network using the model training set and the model verification set.

[0052] The SAR remote sensing aircraft detection model framework of the embodiment of the present invention mainly obtains the multi-scale local details of the aircraft in the image with an improved YoloV8 network, and uses the improved SPPF_AA component in the network to realize the modeling of global semantics and target edge details. At the same time, these multi-scale features are aggregated layer by layer to realize the interaction between local and global information. Then, the rotating detection head (OBB) is used to regress the target box and classify the target. The overall network structure of the framework is as follows: Figure 5 As shown in the figure, Conv(·) represents the convolution operation; SCDown represents the spatial and channel decoupled downsampling operation; Usample represents the upsampling operation; Concat represents the feature concatenation operation; C2F and C2FCIB represent the feature extraction layer; SPPF_AA represents the feature fusion module of the improved pyramid structure; PSA represents the local self-attention module; Concat represents the feature concatenation.

[0053] The improved YoloV8 network is trained using the training set data. During the training process, the model learns the mapping relationship between image features and labels through continuous iterations, and gradually optimizes the model parameters. At the same time, the model is verified using the validation set data, and the training strategy is adjusted according to the verification results (such as stopping training early, adjusting the learning rate, etc.).

[0054] S24, testing the trained YoloV8 network through the model test set, when the test result meets the preset conditions, determining that the improved YoloV8 network training is completed, and obtaining the SAR remote sensing aircraft detection model.

[0055] The trained YoloV8 network is tested using the test set data. During the test, the model predicts each image in the test set and outputs information such as the location and category of the aircraft target.

[0056] Furthermore, based on the test results, the performance indicators of the model (such as accuracy, recall, F1 score, etc.) are calculated. These indicators are used to evaluate the model's ability to detect aircraft in SAR remote sensing images.

[0057] When the performance indicators of the test results meet the preset conditions (such as accuracy higher than a certain threshold), the model training is considered complete. At this point, the improved YoloV8 network obtained is the SAR remote sensing aircraft detection model. This model can be used in subsequent aircraft detection tasks to achieve automatic recognition and positioning of aircraft in SAR remote sensing images.

[0058] The training method of the SAR remote sensing aircraft detection model provided by the embodiment of the present invention performs a series of enhancement processing on the aircraft target in the SAR image, so that the aircraft and the background have stronger differences, and further reduce the ambiguity between similar aircraft targets. On the basis of the SPPF module, the global average pooling layer and the global maximum pooling layer are used to add some global background information and edge detail information of the target to help the network make better judgments. At the same time, the global perspective information is fully captured and the impact of aircraft of different sizes is reduced.

[0059] Figure 3 FIG. 2 is a flow chart of the training method of the SAR remote sensing aircraft detection model provided by the present invention. Figure 3 As shown, the method specifically includes: S31. Collect SAR remote sensing images of different types of aircraft and perform image processing.

[0060] Collect SAR remote sensing image datasets of different types of aircraft and aircraft in different postures (take-off, landing, cruising, etc.). For example, there are 1,000 images in total, including 7 categories: Boeing787, A220, ARJ21, A330, A320 / 321, Boeing737-800 and others, and the data categories are unbalanced.

[0061] The collected original SAR images may contain interference information such as noise and clutter, and preprocessing is required to improve the image quality. In order to improve the detection accuracy of SAR aircraft targets and increase the number of samples, multiple preprocessing operations such as random translation, rotation, color transformation, normalization and flipping are used. Secondly, in order to further improve the detection performance, the larger size images are cropped to 1024 × 1024 size according to the cropping step r = 500. This processing method can ensure the influence of different image sizes.

[0062] S32, dividing the SAR remote sensing images after image processing into a model training set, a model verification set and a model test set.

[0063] The processed SAR remote sensing image dataset is divided into three parts according to a certain ratio (such as 70% training set, 15% validation set, and 15% test set). The training set is used to train the model, the validation set is used to adjust the model parameters (such as learning rate, batch size, etc.) during the training process, and the test set is used to evaluate the final performance of the model.

[0064] Before or after dividing the data set, the aircraft targets in the image need to be annotated to generate label files containing information such as target location and category. These label files will be used as supervision information for model training.

[0065] Specifically, in order to ensure the smooth progress of the experiment in the embodiment of the present invention and to demonstrate the superiority of the SAR aircraft target detection framework on small samples, all samples are divided into three parts: training, testing and validation sets. 400 and 100 SAR remote sensing images after image processing are randomly selected as training and validation samples, and the remaining 500 are test samples.

[0066] S33. Acquire multi-scale local features of aircraft in the SAR remote sensing image in the model training set through the improved YoloV8 network.

[0067] S34. Based on the SAR remote sensing images in the model training set, an improved SPPF_AA component is used in the improved YoloV8 network to model global semantics and target edge details to obtain an initial model.

[0068] S35. Aggregate the multi-scale local features layer by layer in the initial model to obtain interactive information between local and global information.

[0069] S36: Based on the interactive information, use the rotating detection head to perform aircraft frame regression and aircraft target classification.

[0070] The following is a unified description of S33-S36: Assume that the feature map of the input SPPF_AA is F(X) ∈ RH×W×C, where H, W, and C represent the height, width, and channel dimensions of the feature map, respectively. In order to obtain effective local features, these input features are transmitted to a 1 × 1 convolutional layer for squeezing to obtain F1 (X). The specific formula is as follows.

[0071] F1(X) = Conv1×1(F(X))∈ RH×W×C Among them, Conv1×1 (·) represents a 1 × 1 convolution.

[0072] Furthermore, the squeezed feature map F1 (X) is transmitted layer by layer. At the same time, considering that the information flow will be reused between layers during the transmission process, a residual structure is introduced, and the maximum pooling operation is used multiple times to reduce the use of irrelevant information. The three maximum pooling operations are shown in the following equations.

[0073] F2 (X) = MaxPool(F1 (X)) F3 (X) = MaxPool(F1 (X), F2 (X)) F4 (X) = MaxPool(F1 (X), F2 (X), F3 (X)) Among them, F2 (X), F3 (X), F4 (X) represent the output features of different maximum pooling layers.

[0074] MaxPool(·) represents the maximum pooling operation.

[0075] In order to obtain global and contextual semantic information, the global average pooling layer and the global maximum pooling layer are used to add some global background information and edge detail information of the target to help the network make better judgments. At the same time, it fully captures the global perspective information and reduces the impact of aircraft of different sizes. The fused output feature F(X) is shown in the following equation.

[0076] F(X) = Concat[F2 (X), F3 (X), F4 (X), F1 (X), Adaptive_AvgPool(F1(X)), Adaptive_MaxPool(F1 (X))] Where Adaptive_AvgPool(·) and Adaptive_MaxPool(·) represent the adaptive average pooling operation and the adaptive maximum pooling operation respectively. Then, the fusion feature is stimulated, that is, the input convolution layer is stimulated.

[0077] S37: When the accuracy of the initial model's regression of the aircraft frame and the classification result of the aircraft target reaches a first preset threshold, the initial model is verified using the model verification set.

[0078] The improved YoloV8 network is trained using the training set data. During the training process, the model learns the mapping relationship between image features and labels through continuous iterations, and gradually optimizes the model parameters. When the accuracy of the initial model's regression of the aircraft frame and the classification results of the aircraft target reaches the first preset threshold (for example, 98%), the model is verified using the validation set data, and the training strategy is adjusted according to the verification results (such as early stopping of training, adjusting the learning rate, etc.).

[0079] S38. When the initial model is verified, the SAR remote sensing images in the model test set are input into the verified initial model to test the model.

[0080] The trained YoloV8 network is tested using the test set data. During the test, the model predicts each image in the test set and outputs information such as the location and category of the aircraft target.

[0081] Furthermore, based on the test results, the performance indicators of the model (such as accuracy, recall, F1 score, etc.) are calculated. These indicators are used to evaluate the model's ability to detect aircraft in SAR remote sensing images.

[0082] S39. When the test result shows that the detection accuracy of the initial model for the model test set reaches a second preset threshold, it is determined that the training of the improved YoloV8 network is completed, and the SAR remote sensing aircraft detection model is obtained.

[0083] When the performance indicators of the test results meet the preset conditions (such as the accuracy rate is higher than the second preset threshold of 99%), the model training is considered complete. At this time, the improved YoloV8 network obtained is the SAR remote sensing aircraft detection model. This model can be used in subsequent aircraft detection tasks to realize the automatic recognition and positioning of aircraft in SAR remote sensing images.

[0084] The experimental results of aircraft detection in SAR remote sensing images are shown in Table 1. Figure 6 shown.

[0085] Table 1 Experimental results of aircraft detection in SAR remote sensing images

[0086] The training method of the SAR remote sensing aircraft detection model provided by the embodiment of the present invention performs a series of enhancement processing on the aircraft target in the SAR image, so that the aircraft and the background have stronger differences, and further reduce the ambiguity between similar aircraft targets. On the basis of the SPPF module, the global average pooling layer and the global maximum pooling layer are used to add some global background information and edge detail information of the target to help the network make better judgments. At the same time, the global perspective information is fully captured and the impact of aircraft of different sizes is reduced.

[0087] The following is a description of the aircraft target detection device for SAR remote sensing images provided by the present invention. The aircraft target detection device for SAR remote sensing images described below and the aircraft target detection method for SAR remote sensing images described above can be referenced to each other.

[0088] Figure 4The present invention provides a schematic diagram of the structure of an aircraft target detection device in a SAR remote sensing image, which specifically includes: 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.

[0089] The feature extraction module 402 is used to input the target SAR remote sensing image into the pre-trained SAR remote sensing aircraft detection model, and extract the target multi-scale local features of the target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model. For detailed description, please refer to the relevant description corresponding to the above method embodiment, which will not be repeated here.

[0090] The feature aggregation module 403 is used to perform feature aggregation on the target multi-scale local features to obtain target interaction information between the local and global information of the target aircraft. For detailed description, please refer to the relevant description corresponding to the above method embodiment, which will not be repeated here.

[0091] The detection module 404 is used to detect the aircraft target in the target SAR remote sensing image based on the target interaction information, and the detection result includes target frame regression and aircraft classification. For detailed description, please refer to the relevant description corresponding to the above method embodiment, which will not be repeated here.

[0092] 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 communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute a method for detecting an aircraft target in a SAR remote sensing image, the method comprising: obtaining a target SAR remote sensing image to be detected; inputting the target SAR remote sensing image into a pre-trained SAR remote sensing aircraft detection model, extracting target multi-scale local features of the target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model; performing feature aggregation on the target multi-scale local features to obtain target interaction information between local and global information of the target aircraft; performing aircraft target detection on the target SAR remote sensing image based on the target interaction information, and the detection result includes target frame regression and aircraft classification.

[0093] 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.

[0094] 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 method for detecting aircraft targets in SAR remote sensing images provided by the above methods, and the method includes: obtaining a target SAR remote sensing image to be detected; inputting the target SAR remote sensing image into a pre-trained SAR remote sensing aircraft detection model, and extracting target multi-scale local features of the target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model; performing feature aggregation on the target multi-scale local features to obtain target interaction information between local and global information of the target aircraft; and performing aircraft target detection on the target SAR remote sensing image based on the target interaction information, and the detection results include target box regression and aircraft classification.

[0095] 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 methods for detecting aircraft targets in SAR remote sensing images provided by the above-mentioned methods, the method comprising: obtaining a target SAR remote sensing image to be detected; inputting the target SAR remote sensing image into a pre-trained SAR remote sensing aircraft detection model, and extracting target multi-scale local features of the target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model; performing feature aggregation on the target multi-scale local features to obtain target interaction information between local and global information of the target aircraft; and performing aircraft target detection on the target SAR remote sensing image based on the target interaction information, and the detection results include target box regression and aircraft classification.

[0096] 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.

[0097] 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.

[0098] 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 targets 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 a pre-trained SAR remote sensing aircraft detection model, and extracting target multi-scale local features of the target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model; Performing feature aggregation on the multi-scale local features of the target to obtain target interaction information between local and global information of the target aircraft; Aircraft target detection is performed on the target SAR remote sensing image based on the target interaction information, and the detection result includes target frame regression and aircraft classification.

2. The method according to claim 1, characterized in that The SAR remote sensing aircraft detection model is trained based on the following method: Collect SAR remote sensing images from different types of aircraft and perform image processing; The SAR remote sensing images after image processing are divided into a model training set, a model verification set and a model test set; The improved YoloV8 network is trained by using the model training set and the model validation set; The trained YoloV8 network is tested by the model test set. When the test result meets the preset conditions, it is determined that the training of the improved YoloV8 network is completed, and the SAR remote sensing aircraft detection model is obtained.

3. The method according to claim 2, characterized in that The improved YoloV8 network is trained by the model training set and the model verification set, including: Acquire multi-scale local features of aircraft in the SAR remote sensing image in the model training set through the improved YoloV8 network; Based on the SAR remote sensing images in the model training set, an improved SPPF_AA component is used in the improved YoloV8 network to model global semantics and target edge details to obtain an initial model; Aggregating the multi-scale local features layer by layer in the initial model to obtain interactive information between local and global information; Based on the interactive information, a rotating detection head is used to perform aircraft frame regression and aircraft target classification; When the accuracy of the regression of the aircraft frame and the classification result of the aircraft target by the initial model reaches a first preset threshold, the initial model is verified by the model verification set.

4. The method according to claim 3, characterized in that Based on the SAR remote sensing image in the model training set, the improved SPPF_AA component is used in the improved YoloV8 network to model global semantics and target edge details to obtain an initial model, including: Squeezing the SAR remote sensing images in the model training set through a convolutional layer; The squeezed feature map is passed through multiple maximum pooling operations and residual structures; Using a global average pooling layer and a global maximum pooling layer to obtain global and contextual semantic information, and fusing the global and contextual semantic information with edge detail information; Based on the information fusion results, global semantics and target edge details are modeled to obtain the initial model.

5. The method according to claim 4, characterized in that The trained YoloV8 network is tested by the model test set, and when the test result meets the preset conditions, it is determined that the improved YoloV8 network training is completed, and the SAR remote sensing aircraft detection model is obtained, including: When the initial model is verified, the SAR remote sensing images in the model test set are input into the verified initial model to test the model; When the test result is that the detection accuracy of the initial model for the model test set reaches a second preset threshold, it is determined that the improved YoloV8 network training is completed, and the SAR remote sensing aircraft detection model is obtained.

6. The method according to claim 2, characterized in that The collecting of SAR remote sensing images of different types of aircraft and performing image processing include: The SAR remote sensing image is subjected to image processing to obtain a processed SAR remote sensing image, wherein the image processing includes random translation, rotation, color conversion, normalization and flipping processing operations.

7. A device for detecting aircraft targets in SAR remote sensing images, characterized in that: include: An acquisition module is used to acquire a SAR remote sensing image of a target to be detected; A feature extraction module is used to input the target SAR remote sensing image into a pre-trained SAR remote sensing aircraft detection model, and extract the target multi-scale local features of the target aircraft in the target SAR remote sensing image through the SAR remote sensing aircraft detection model; A feature aggregation module is used to perform feature aggregation on the multi-scale local features of the target to obtain target interaction information between the local and global information of the target aircraft; The detection module is used to perform aircraft target detection on the target SAR remote sensing image based on the target interaction information, and the detection result includes target frame regression and aircraft classification.

8. 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 method for detecting aircraft targets in SAR remote sensing images according to any one of claims 1 to 6 is implemented.

9. 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 method for detecting aircraft targets in SAR remote sensing images according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for detecting aircraft targets in SAR remote sensing images according to any one of claims 1 to 6 is implemented.

Images