Synthetic aperture radar image target detection method, apparatus and device, and storage medium

Abstract

The invention discloses a synthetic aperture radar image target detection method and device, equipment and a storage medium. The method comprises the following steps: acquiring a to-be-detected SAR image; performing denoising processing on the SAR image by using a denoising network; establishing an integrated target detection model which comprises a denoising network and a subsequent open vocabulary target detection model, and training the integrated model by using an improved bounding box regression loss function constructed based on angular point distance and side edge overlapping information; and inputting the denoised SAR image and text prompt into the trained open vocabulary target detection model for feature extraction and target detection, and outputting a detection result. According to the method, the denoising technology and open vocabulary target detection are combined, the problem that speckle noise is not fully processed in the current SAR target detection field is solved, meanwhile, the limitation of traditional target detection on fixed categories is overcome, the target detection precision is improved, and the applicability and robustness of the model are improved.

Description

Technical Field

[0001] The present disclosure relates to the field of synthetic aperture radar and machine vision, and in particular to a synthetic aperture radar image target detection method, device, equipment and storage medium. Background Art

[0002] Synthetic Aperture Radar (SAR) has important applications in military, geological and other fields due to its high resolution and all-weather and all-day operation capabilities. SAR technology can work under various lighting and climatic conditions, reflect the physical properties and surface roughness of objects, and provide information different from optical and infrared images. These advantages make SAR image target detection an increasingly important research field.

[0003] However, SAR image target detection faces unique challenges. Due to the coherence of the SAR system, SAR images inherently contain speckle noise, which severely limits the effectiveness of detection algorithms. Compared with optical images, the features of targets in SAR images are often not clear enough, which increases the difficulty of detection. In addition, SAR images may contain various complex terrains and backgrounds, making it difficult to distinguish between targets and backgrounds. In addition, targets in SAR images may have different sizes, shapes, and directions, which increases the complexity of detection.

[0004] In response to these challenges, researchers have conducted many explorations. For example, through image preprocessing, including traditional wavelet shrinkage, spatial filtering, frequency domain filtering and other technologies, as well as emerging deep learning denoising methods, these methods aim to reduce noise in SAR images and improve image quality. Researchers have tried various feature extraction methods to better capture the characteristics of targets in SAR images, and have also improved the detection ability of small targets by improving detection algorithms (such as improving target positioning algorithms). In recent years, various deep learning models have also been applied to SAR image target detection in an attempt to improve the accuracy and robustness of detection.

[0005] Despite the continuous development of SAR image target detection technology, existing methods still have many shortcomings. Existing technologies have limited effects in processing speckle noise unique to SAR images, and it is difficult to fully utilize the unique information of SAR images. Although some methods attempt to improve the detection capabilities of targets of different types and scales, generalization still needs to be improved. Moreover, most current methods treat denoising and target detection as independent steps, lacking a unified end-to-end framework, which not only reduces processing efficiency but also may lead to information loss. At the same time, existing target localization methods are not accurate enough when processing SAR images, especially in complex backgrounds or small targets, and the loss function design fails to fully consider the special properties of SAR images. In addition, the training of target detection models often requires a large amount of labeled data, and it is challenging to obtain large-scale labeled SAR image datasets, which to some extent also limits the further improvement of model performance. Summary of the invention

[0006] The present invention proposes a SAR target detection model that combines a denoising network and an open set target detection technology, which solves the problem of insufficient processing of speckle noise in the current SAR target detection field, overcomes the limitation of traditional target detection on fixed categories, improves the target detection accuracy, and expands the applicability and robustness of the model.

[0007] According to an embodiment of the present disclosure, a synthetic aperture radar image target detection method is proposed, comprising:

[0008] Acquire a synthetic aperture radar (SAR) image to be detected;

[0009] Performing denoising on the SAR image using a denoising network to obtain a denoised SAR image;

[0010] Establishing an integrated target detection model, the integrated target detection model includes a denoising network and an open vocabulary target detection model followed by the denoising network, and training the integrated target detection model using an improved bounding box regression loss function constructed based on corner distance and side overlap information;

[0011] The denoised SAR image and text prompt are input into a trained open vocabulary target detection model for feature extraction and target detection, and a detection result is output, wherein the detection result includes a bounding box and a category of the identified target.

[0012] In some embodiments, the denoising network is a pre-trained AGSDNet model.

[0013] In some embodiments, the base model of the open vocabulary object detection model is a GroundingDINO model.

[0014] In some embodiments, the open vocabulary object detection model uses a pre-trained Swin Transformer for image feature extraction and / or a pre-trained BERT for text feature extraction.

[0015] In some embodiments, the following improved bounding box regression loss function L is constructed: SSCA :

[0016] L SSCA =L SSO +α*L CD ,

[0017] Among them, L SSO is the scaled side overlap loss, L CD is the corner distance loss, and α is the weight coefficient.

[0018] In some embodiments, the scaled side overlap loss L is obtained based on the following formula: SSO :

[0019] L SSO =-0.5*SO,

[0020] Among them, SO is the overlap ratio between the predicted box and the real box in width and height.

[0021] In some embodiments, the corner distance loss L is obtained based on the following formula: CD :

[0022] L CD =(D corner1 / D enclose )+(D corner2 / D enclose ),

[0023] Among them, D corner1 is the distance between the upper left corner of the predicted box and the real box, D corner2 is the distance between the predicted box and the lower right corner of the real box, D enclose is the diagonal distance of the minimum bounding box that contains the predicted box and the true box.

[0024] In some implementations, the weight coefficient α has a value range of 0<α≤1.

[0025] In some embodiments, the integrated target detection model is trained using the AdamW optimizer, and the learning rate and weight decay of the AdamW optimizer are both in the range of 10 -5 to 10 -3 between.

[0026] In some implementations, the text prompt includes at least one category.

[0027] In some embodiments, the detection result also includes a confidence level between each identified target and the corresponding category.

[0028] According to one embodiment of the present disclosure, a synthetic aperture radar image target detection device is proposed, comprising:

[0029] A SAR image acquisition unit, used for acquiring a synthetic aperture radar (SAR) image to be detected;

[0030] A denoising unit, configured to perform denoising processing on the SAR image using a denoising network to obtain a denoised SAR image;

[0031] An integrated model building unit, used to build an integrated target detection model, wherein the integrated target detection model includes a denoising network and an open vocabulary target detection model followed by the denoising network, and trains the integrated target detection model using an improved bounding box regression loss function built based on corner point distance and side overlap information;

[0032] The inference unit is used to input the denoised SAR image and text prompt into the trained open vocabulary target detection model for feature extraction and target detection, and output the detection result, which includes the bounding box and category of the identified target.

[0033] According to one embodiment of the present disclosure, an electronic device is provided, the device comprising a memory and a processor, the memory being used to store computer instructions executable on the processor, the processor being used to implement any of the above methods when executing the computer instructions.

[0034] According to an embodiment of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, any of the above methods is implemented.

[0035] The SAR image target detection method proposed in the present disclosure achieves significant performance improvement by combining a denoising network and an open vocabulary target detection model. First, the introduction of a special denoising network effectively reduces the speckle noise in the SAR image, improves the image quality, and lays a good foundation for subsequent target detection. At the same time, the use of an open vocabulary target detection model greatly expands the range of detectable target categories, allowing the model to flexibly respond to various types of targets, greatly improving the flexibility and applicability of detection. The combination of a denoising network and an open vocabulary target detection model not only overcomes the limitations of traditional methods on fixed categories, but also enhances the model's ability to handle complex scenes. The present disclosure also introduces an improved bounding box regression loss function, which significantly improves the accuracy of target positioning by comprehensively considering the corner distance and side overlap information, especially when processing targets with blurred boundaries in SAR images. It has good performance. In addition, the end-to-end integration solution proposed in the present disclosure not only improves processing efficiency, but also reduces the information loss that may be caused by intermediate steps.

[0036] Other features and advantages of the present disclosure are described in detail below. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the specification and, together with the description, serve to explain the principles of the specification.

[0038] Figure 1 A flowchart of a synthetic aperture radar image target detection method according to an embodiment of the present disclosure is shown.

[0039] Figure 2 The overall architecture of a DGD (Denoise Grounding DINO) model according to an exemplary embodiment of the present disclosure is shown.

[0040] Figure 3 A schematic diagram showing a comparison of an image before denoising and an image after denoising according to an exemplary embodiment of the present disclosure is shown.

[0041] Figure 4 A schematic diagram showing a comparison between a ship detection result obtained by a GD model and a ship detection result obtained by a DGD model according to an exemplary embodiment of the present disclosure is shown.

[0042] Figure 5 A schematic diagram showing a comparison of a vehicle detection result obtained by a GD model and a vehicle detection result obtained by a DGD model according to an exemplary embodiment of the present disclosure is shown.

[0043] Figure 6A schematic diagram comparing the tank detection results obtained by the GD model and the tank detection results obtained by the DGD model according to an exemplary embodiment of the present disclosure is shown.

[0044] Figure 7 It is a schematic diagram of the structure of an electronic device shown in at least one embodiment of the present disclosure. DETAILED DESCRIPTION

[0045] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0046] The disclosed embodiments may be applied to a computer system / server that may operate with numerous other general or special computing system environments or configurations. Examples of well-known computing systems, environments, and / or configurations suitable for use with a computer system / server include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, networked personal computers, minicomputer systems, mainframe computer systems, and distributed cloud computing technology environments including any of the above systems, and the like.

[0047] Computer systems / servers may be described in the general context of computer system executable instructions (such as program modules) executed by a computer system. In general, program modules may include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers may be implemented in a distributed cloud computing environment where tasks are performed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules may be located on local or remote computing system storage media including storage devices.

[0048] Figure 1 A synthetic aperture radar image target detection method according to an embodiment of the present disclosure is shown. As shown in the figure, the method includes steps 1 to 4.

[0049] Step 1: Acquire a synthetic aperture radar (SAR) image to be detected.

[0050] A synthetic aperture radar system (SAR system) can be used to scan a target area to collect SAR images. SAR systems can be mounted on satellites, aircraft, or other platforms. SAR systems transmit microwave signals and receive signals reflected from the ground, forming high-resolution two-dimensional images through complex signal processing techniques.

[0051] The raw SAR images can be preprocessed by geometric correction, radiation correction, etc., to eliminate systematic errors and atmospheric effects. The processed SAR images can be stored in digital formats, such as TIFF, JPEG2000, etc.

[0052] In some real-time monitoring systems, SAR images may be directly transmitted from the radar system to the processing unit, while in a research environment, pre-collected SAR image datasets are often used. In some embodiments, suitable SAR images may be selected from the image library according to the specific detection task. The selection criteria may include image resolution, coverage area, imaging time, etc.

[0053] Step 2: Use a denoising network to denoise the SAR image to obtain a denoised SAR image.

[0054] Currently, many SAR target detection models only use a single feature extraction detection network, and rarely consider denoising. This embodiment introduces denoising into target detection to optimize the extracted features and improve detection performance.

[0055] In some preferred embodiments, the denoising network is a pre-trained AGSDNet model.

[0056] After in-depth research and comparison, the inventors believe that the AGSDNet model combines dilated convolution and gradient information modules as a denoising design. Adding the AGSDNet denoising network before the image encoder has a good effect on removing speckle noise in SAR images.

[0057] Specifically, the AGSDNet model is a convolutional neural network consisting of a gradient calculation module (GCB), a feature denoising module (FDB), and two attention modules. The feature denoising module contains a convolutional layer and a residual connection block, which contains average pooling and convolution layers to simulate median filtering denoising.

[0058] The AGSDNet model can be pre-trained using SAR images so that the network can learn how to effectively remove noise from SAR images. When the integrated target detection model is subsequently trained, the AGSDNet model initialized by the pre-trained weights is frozen, that is, the weights of the AGSDNet model are no longer updated during the training process.

[0059] Step 3, establishing an integrated target detection model, wherein the integrated target detection model includes a denoising network and an open vocabulary target detection model followed by the denoising network, and using an improved bounding box regression loss function based on corner point distance and side overlap information to train the integrated target detection model.

[0060] The inventors believe that compared with the traditional closed-set object detection model, the open vocabulary object detection model introduces language information, trains the visual-language alignment space through image-text pairs, has zero-sample learning ability and stronger fine-tuning ability, and can handle a wider range of object categories.

[0061] The object detection model integrates a denoising network and an open vocabulary object detection model followed by a denoising network, which contains two main input branches: SAR images and text prompts. The object detection model includes the following key parts:

[0062] Denoising network: located before the image encoder, used to remove speckle noise in SAR images;

[0063] Two encoders: one for feature extraction of text and one for denoised image;

[0064] Feature enhancement module: used to fuse cross-modal features;

[0065] Cross-modal query selection module: comprehensively determines cross-modal queries.

[0066] Decoder: Outputs the recognized bounding box and category results.

[0067] By combining the denoising network with the open vocabulary object detection model, an end-to-end integrated object detection model is formed. This integration enables the model to handle both SAR image denoising and open vocabulary object detection tasks, significantly improving the overall detection performance. The model can effectively handle the noise problem in SAR images while taking advantage of the advantages of open vocabulary object detection to achieve more flexible and accurate object detection.

[0068] In some embodiments, the GroundingDINO model can be selected as the base model of the open vocabulary object detection model to further improve the object detection performance. The feature enhancement module of the GroundingDINO model can fuse the image and text features through self-attention and text-image cross-attention mechanisms, and the query selection module can use the fused features to comprehensively determine the cross-modal query. The detection decoder uses self-attention, cross-modal attention, and feed-forward layers to output the final bounding box and category results based on the cross-modal query.

[0069] In some embodiments, a pre-trained Swin Transformer can be used as an image encoder of an open vocabulary object detection model for image feature extraction. In some embodiments, a pre-trained BERT can be used as a text encoder of an open vocabulary object detection model for text feature extraction.

[0070] In an example, the GroundingDINO model can use a pre-trained Swin Transformer as an image encoder, a pre-trained BERT as a text encoder, and add an AGSDNet model as a denoising network before the image encoder to obtain an integrated target detection model Denoise GroundingDINO, referred to as DGD.

[0071] In order to improve the accuracy of target detection in SAR images, the inventors proposed an improved bounding box regression loss function, called scaled side-corner aligned (SSCA) loss. The SSCA loss function is constructed based on corner point distance and side overlap information to replace the traditional GIoU loss function. It can better adapt to the characteristics of SAR images and improve the accuracy of target positioning.

[0072] In some embodiments, the following improved bounding box regression loss function L is constructed: SSCA :

[0073] L SSCA =L SSO +α*L CD

[0074] Among them, L SSO is the scaled side overlap loss, L CD is the corner distance loss, and α is the weight coefficient. This loss function combines the side overlap information and corner distance information of the bounding box, and can comprehensively evaluate the difference between the predicted box and the real box.

[0075] By combining the scaled side overlap loss L SSO And the corner distance loss L CD , which takes into account both the degree of boundary overlap and the precise positioning of corner points. Therefore, according to the bounding box regression loss function L in this embodiment, SSC A can comprehensively evaluate the quality of the predicted box and improve the overall performance of SAR image target detection.

[0076] The following is a description of the scaled side overlap loss L SSO And the corner distance loss L CD for further discussion.

[0077] The side overlap ratio SO is defined as follows:

[0078] SO=(w min / w max )+(h min / h max ),

[0079] Among them, w min and h min are the overlapped parts of the predicted box and the real box in width and height, respectively, ma x and h max are the union of the width and height of the two boxes respectively. For example, suppose the predicted box is (x 1 ,x 2 ,y 1 ,y 2 ), the real box is (x′ 1 ,x′ 2 ,y′ 1 ,y′ 2 ), define w min =min(x 2 ,x′ 2 )-max(x 1 ,x′ 1 ), w max =max(x 2 ,x′ 2 )-min(x 1 ,x′ 1 ), h min =min(y 2 ,y′ 2 )-max(y 1 ,y′ 1 ), h max =max(y 2 ,y′ 2 )-min(y 1 ,y′ 1 ).

[0080] The commonly used side overlap loss function definition L SO As shown below:

[0081] L SO =2–SO.

[0082] After in-depth research, the inventors have taken into account the range of classic I oU functions (such as GioU) and conventional L SO Therefore, in some embodiments, the scaled side overlap loss L is obtained based on the following formula: SSO As improved SSO loss:

[0083] L SSO =-0.5*SO,

[0084] Among them, SO is the overlap ratio between the predicted box and the real box in width and height.

[0085] The improved L defined in this embodiment SSO The value range of is [-1,0], which is easier to combine with other loss functions.

[0086] In some embodiments, the corner distance loss L is obtained based on the following formula: CD :

[0087] L CD =(D corner1 / D enclose )+(D corner2 / D enclose ),

[0088] Among them, D corner1 is the distance between the upper left corner of the predicted box and the real box, D corner2 is the distance between the predicted box and the lower right corner of the real box, D enclose is the diagonal distance of the minimum bounding box that contains the predicted box and the true box.

[0089] Through the above corner point distance loss L CD , which can effectively measure the degree of alignment between the predicted box and the real box corner points.

[0090] Improved bounding box regression loss function L SSCA The weight coefficient α in L is used to balance SSO and L CD Contribution of the two loss terms. In some embodiments, the weight coefficient α ranges from 0<α≤1. In some examples, the value of α can be set to 0.5.

[0091] Step 4: Input the denoised SAR image and text prompt into the trained open vocabulary target detection model for feature extraction and target detection, and output the detection result, which includes the bounding box and category of the identified target.

[0092] The trained open vocabulary object detection model receives two key inputs: the SAR image processed by the denoising network and a textual hint describing the object, for example, the textual hint can be provided by the user. The denoised SAR image has higher quality and the noise is significantly reduced. In some embodiments, the textual hint can include at least one word describing the category, such as "ship", "aircraft", etc.

[0093] The open vocabulary object detection model performs feature extraction on the input it receives. For the image part, the open vocabulary object detection model uses an image encoder (e.g., Swin Transformer) to extract rich visual features from the denoised SAR image, and uses a text encoder (e.g., BERT model) to extract semantic features from the textual cues. These two features are then fused through a feature enhancement module, which can include self-attention and text-image cross-attention mechanisms to generate comprehensive cross-modal features.

[0094] After feature fusion, the query selection module of the model uses these cross-modal features to generate queries and input them into the decoder. The decoder can contain self-attention, cross-attention, and feed-forward layers to perform object detection based on these queries. This process can make full use of the combination of visual and semantic information to more accurately locate and identify targets in complex SAR images.

[0095] Finally, the model outputs the detection results, including the bounding box of each detected object (indicating the location and size of the object) and the corresponding category. In some embodiments, the model also outputs the confidence between the identified object and the corresponding category, and detection results with different confidence intervals can be obtained by changing the confidence threshold.

[0096] The SAR image target detection method proposed in this embodiment achieves significant performance improvement by combining a denoising network and an open vocabulary target detection model. First, the introduction of a special denoising network effectively reduces the speckle noise in the SAR image, improves the image quality, and lays a good foundation for subsequent target detection. At the same time, the use of an open vocabulary target detection model greatly expands the range of detectable target categories, allowing the model to flexibly respond to various types of targets, greatly improving the flexibility and applicability of detection. The combination of a denoising network and an open vocabulary target detection model not only overcomes the limitations of traditional methods on fixed categories, but also enhances the model's ability to handle complex scenes. The present disclosure also introduces an improved bounding box regression loss function, which significantly improves the accuracy of target positioning by comprehensively considering the corner distance and side overlap information, especially when processing targets with blurred boundaries in SAR images. It has good performance. In addition, the end-to-end integration solution proposed in the present disclosure not only improves processing efficiency, but also reduces the information loss that may be caused by intermediate steps.

[0097] In some embodiments, the integrated target detection model is trained using the AdamW optimizer, and the learning rate and weight decay of the AdamW optimizer are both in the range of 10 -5 to 10 -3 between.

[0098] In this embodiment, the AdamW optimizer is selected as the core optimization algorithm when training the integrated target detection model. AdamW is a variant of the Adam (Adaptive Moment Estimation) optimizer, which introduces the weight decay (Weight Decay) regularization technology based on the original Adam. Selecting the AdamW optimizer is conducive to improving the parameter update efficiency and model generalization ability during the training process.

[0099] Learning rate and weight decay are two key parameters of the AdamW optimizer. The learning rate controls the step size of each parameter update and has an important impact on the convergence speed and final performance of the model. Weight decay is a regularization technique that prevents model overfitting by adding a parameter norm penalty term to the loss function. In this implementation, the value range of these two parameters is set to 10 -5 to 10 -3 A balance can be struck between ensuring that the model learns effectively and preventing the learning process from being unstable or overfitting.

[0100] For example, a learning rate within this range can enable the model to learn quickly in the early stages of training, and also enable more detailed parameter adjustments in the later stages. -3 ) helps to quickly explore the parameter space in the early stages of training, while a smaller learning rate (e.g. close to 10 -5 ) helps achieve more refined parameter optimization in the later stages of training. Similarly, 10 -5 to 10 -3 The weight decay between θ and θ can provide sufficient regularization to prevent overfitting without excessively restricting the learning ability of the model.

[0101] Figure 2 The overall architecture of the DGD (Denoise Grounding DINO) model according to an exemplary embodiment of the present disclosure is shown, which is an integrated model combining denoising and open vocabulary object detection. The model contains two main input branches: text input and image input.

[0102] The text input branch receives category hints in the format of "car.ship.aircraft." and passes them to the GroundingDINO module after being processed by the text encoder. The image input branch first passes through the Denoise Net to remove speckle noise in the SAR image, and then passes through the Image Encoder to extract features.

[0103] The GroundingDINO module is the core of the model, which includes components such as Feature Enhancer, Query Selection module and Decoder. Feature Enhancer is used to fuse text and image features, Query Selection is used to generate queries based on fused features, and finally the Decoder outputs the detection results. The output of the model includes bounding box (Box), category (Cls) and two losses - SSCA loss (SSCA Loss) and classification loss (Cls Loss). SSCA loss is an improved bounding box regression loss used to optimize target positioning accuracy. Classification loss is used to evaluate the confidence between the identified target and the corresponding category.

[0104] The synthetic aperture radar image target detection scheme according to the present disclosure is verified by using specific examples below.

[0105] Table 1 Dataset description

[0106]

[0107] The SSDD dataset consists mainly of SAR images of small ships, with a total of 1,160 images. Each image contains an average of 2.1 targets, the image size is 500*500 pixels, and there is only one category label for ships. The HRSID dataset consists mainly of high-resolution ship SAR images, with a total of 5,604 images, each image contains an average of 3.0 targets, and there is only one category label for ships. The SARDet-100K dataset collects 10 public SAR datasets and crops each image into 512*512 image blocks, including six types of targets: ships, aircraft, cars, tanks, bridges, and harbors.

[0108] Since the SARDet-100K dataset is very large and testing is very time-consuming, in this example, a smaller SARDet-100K (partial) dataset is obtained by extracting it from the original dataset at a ratio of 10:1. First, the category with the most instances in an image can be defined as the category of the image, and then 1,000 images are extracted for each category in the training dataset. Since there are only 837 images in the car category, a total of 5,837 training set images are finally obtained. The test datasets are all obtained in the same way.

[0109] In this example, Grounding-DINO-T pre-trained on the visible light dataset is used as the initial model, and the SAR pre-trained AGSDNet version is used to initialize the denoising model, keep the text encoder and denoising model frozen, and fine-tune other structures of the detection model.

[0110] The text input of SSDD and HRSID is "ship.", while the text input of SARDet-100K (part) is "ship.aircraft.car.tank.bridge.harbor.". SCA is somewhat robust to α. For simplicity, this example does not consider the impact of different hyperparameters in the loss function. Both SCA loss and SSCA loss take α as 0.5.

[0111] In this example, RandomFlip with prob=0.5 and multi-scale RandomChoiceResize are set for training image preprocessing, while a fixed 800*1333 resize is used for test image preprocessing. AdamW is used as the optimizer, with a learning rate of 1e-4 and a weight decay of 1e-4; gradient clipping is also used. To maintain fairness, all models are trained for 12 rounds in each dataset experiment. This example conducts all experiments on the MMDetection framework.

[0112] In order to evaluate the effectiveness of the model, a series of comparative experiments and ablation experiments were conducted on the SSDD, HRSID and SARDet-100K (partial) datasets to demonstrate the important role of the denoising network, SCA loss function and improved SSCA loss function. The detailed results are shown in Table 2. In the experiment, AP50 and mAP, which are commonly used indicators in object detection, are compared for each experiment.

[0113] Table 2 Performance of different detection models on three SAR target detection datasets

[0114]

[0115] The conclusion is as follows:

[0116] (1) Comparison with SOTA models: The performance of the target detection model GroundingDINO according to the present disclosure is much better than the current SOTA and MSFA (Faster R-CNN+VAN-B), especially on the HRSID dataset, where the mAP is improved by about 8.0 and the AP50 is improved by about 9.0, which fully verifies the excellent fine-tuning ability of the open vocabulary detection model GroundingDINO in SAR image detection tasks;

[0117] (2) Comparison of different detection box loss functions: As can be seen from Table 2, compared with GIoU, there is a significant improvement after using SCA loss, and there is further improvement after using SSCA loss. After combining with the denoising network, the improvement is even more obvious;

[0118] (3) Comparison with and without adding the denoising network: After adding the denoising network, the improvement of mAP is more obvious, indicating that the detection box regression result becomes more accurate.

[0119] In summary, the DGD model according to the present disclosure achieves the best performance on both simple SSDD and HRSD datasets and challenging SAR datasets.

[0120] The detection results of the GroundingDINO (GD) model and the Denoise GroundingDINO (DGD) model can be further compared through visualization. Figure 3 The left side is the image before denoising, and the right side is the image after denoising. Figure 4 The left side shows the ship detection results obtained by the GD model, and the right side shows the ship detection results obtained by the DGD model of the present disclosure. The red box indicates the predicted result, and the green box indicates the actual annotation. Figure 3 and Figure 4 It can be seen that after denoising, the spots of the ship are significantly reduced and the bounding box is more accurate.

[0121] Figure 5 The middle left picture is the car detection result obtained by the GD model, and the right picture is the car detection result obtained according to the DGD model disclosed in the present invention. Figure 6 The middle left picture is the tank detection result obtained by the GD model, and the right picture is the tank detection result obtained according to the DGD model disclosed in the present invention. Figure 5 and Figure 6 The red box represents the prediction result, and the green box represents the true annotation.

[0122] from Figure 5 It can be seen that there are two false detections in the bottom and lower right corner of the GD detection image. Figure 6 It can be seen that compared with DGD, GD has three missed detections on the right side, which fully reflects the superiority of the DGD model obtained according to the present disclosure.

[0123] The above examples show that the DGD model performs well in SAR target detection tasks. In addition, experimental analysis shows that the integrated target detection model according to the present disclosure is scalable and can be migrated to other detection models, which is conducive to subsequent in-depth research.

[0124] For other details and beneficial effects of this embodiment, please refer to the above related introduction, which will not be repeated here.

[0125] Figure 7 An electronic device provided for at least one embodiment of the present disclosure includes a memory and a processor, wherein the memory is used to store computer instructions that can be executed on the processor, and the processor is used to implement the synthetic aperture radar image target detection method described in any embodiment or implementation of the present disclosure when executing the computer instructions.

[0126] At least one embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, and when the program is executed by a processor, the synthetic aperture radar image target detection method described in any embodiment or implementation of the present disclosure is implemented.

[0127] It should be understood by those skilled in the art that one or more embodiments of the present specification may be provided as a method, system or computer program product. Therefore, one or more embodiments of the present specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, one or more embodiments of the present specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0128] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the data processing device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0129] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0130] The embodiments of the subject matter and functional operations described in this specification may be implemented in the following: digital electronic circuits, tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or a combination of one or more of them. The embodiments of the subject matter described in this specification may be implemented as one or more computer programs, i.e., one or more modules in computer program instructions encoded on a tangible non-temporary program carrier to be executed by a data processing device or to control the operation of the data processing device. Alternatively or additionally, the program instructions may be encoded on an artificially generated propagation signal, such as a machine-generated electrical, optical or electromagnetic signal, which is generated to encode information and transmit it to a suitable receiver device for execution by a data processing device. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.

[0131] The processes and logic flows described in this specification can be performed by one or more programmable computers executing one or more computer programs to perform corresponding functions by operating on input data and generating output. The processes and logic flows can also be performed by special purpose logic circuits, such as FPGAs (field programmable gate arrays) or ASICs (application-specific integrated circuits), and the apparatus can also be implemented as special purpose logic circuits.

[0132] Computers suitable for executing computer programs include, for example, general and / or special microprocessors, or any other type of central processing unit. Typically, the central processing unit will receive instructions and data from a read-only memory and / or a random access memory. The basic components of a computer include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Typically, the computer will also include one or more large-capacity storage devices for storing data, such as magnetic disks, magneto-optical disks, or optical disks, or the computer will be operably coupled to this large-capacity storage device to receive data from it or to transmit data to it, or both. However, the computer does not necessarily have such a device. In addition, the computer can be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, to name a few.

[0133] Computer readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including, for example, semiconductor memory devices (e.g., EPROM, EEPROM and flash memory devices), magnetic disks (e.g., internal hard disks or removable disks), magneto-optical disks, and CD ROM and DVD-ROM disks. The processor and memory may be supplemented by, or incorporated in, special purpose logic circuitry.

[0134] Although this specification includes many specific implementation details, these should not be interpreted as limiting the scope of any invention or the scope of protection claimed, but are mainly used to describe the features of the specific embodiments of specific inventions. Certain features described in multiple embodiments in this specification may also be implemented in combination in a single embodiment. On the other hand, the various features described in a single embodiment may also be implemented separately in multiple embodiments or in any suitable sub-combination. In addition, although features may work in certain combinations as described above and even initially claim protection, one or more features from the claimed combination may be removed from the combination in some cases, and the claimed combination may point to a sub-combination or a variation of a sub-combination.

[0135] Similarly, although operations are depicted in a particular order in the accompanying drawings, this should not be understood as requiring that these operations be performed in the particular order shown or performed sequentially, or requiring that all illustrated operations be performed to achieve the desired results. In some cases, multitasking and parallel processing may be advantageous. In addition, the separation of various system modules and components in the above-described embodiments should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product, or packaged into multiple software products.

[0136] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the particular order or sequential order shown to achieve the desired results. In some implementations, multitasking and parallel processing may be advantageous.

[0137] The above description is merely a preferred embodiment of one or more embodiments of the present specification and is not intended to limit one or more embodiments of the present specification. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present specification shall be included in the scope of protection of one or more embodiments of the present specification.

Claims

1. A synthetic aperture radar image target detection method, characterized in that: include: Acquire a synthetic aperture radar (SAR) image to be detected; Performing denoising on the SAR image using a denoising network to obtain a denoised SAR image; Establishing an integrated target detection model, the integrated target detection model includes a denoising network and an open vocabulary target detection model followed by the denoising network, and training the integrated target detection model using an improved bounding box regression loss function constructed based on corner distance and side overlap information; The denoised SAR image and text prompt are input into a trained open vocabulary target detection model for feature extraction and target detection, and a detection result is output, wherein the detection result includes a bounding box and a category of the identified target.

2. The synthetic aperture radar image target detection method according to claim 1, characterized in that: The denoising network is a pre-trained AGSDNet model.

3. The synthetic aperture radar image target detection method according to claim 1, characterized in that: The base model of the open vocabulary object detection model is the GroundingDINO model.

4. The synthetic aperture radar image target detection method according to claim 1 or 3, characterized in that: The open vocabulary object detection model uses a pre-trained Swin Transformer for image feature extraction and / or a pre-trained BERT for text feature extraction.

5. The synthetic aperture radar image target detection method according to claim 1, characterized in that: Construct the following improved bounding box regression loss function L SSCA : L SSCA =L SSO +a*L CD , Among them, L SSO is the scaled side overlap loss, L CD is the corner distance loss, and α is the weight coefficient.

6. The synthetic aperture radar image target detection method according to claim 5, characterized in that: The scaled side overlap loss L is obtained based on the following formula SSO : L SSO =-0.5*SO, Among them, SO is the overlap ratio between the predicted box and the real box in width and height.

7. The synthetic aperture radar image target detection method according to claim 5, characterized in that: The corner distance loss L is obtained based on the following formula CD : L CD =(D corner1 / D enclose )+(D corner2 / D enclose ), Among them, D corner1 is the distance between the upper left corner of the predicted box and the real box, D corner2 is the distance between the predicted box and the lower right corner of the real box, D enclose is the diagonal distance of the minimum bounding box that contains the predicted box and the true box.

8. The synthetic aperture radar image target detection method according to claim 5, characterized in that: The value range of the weight coefficient α is 0<α≤1.

9. The synthetic aperture radar image target detection method according to claim 1, characterized in that: The AdamW optimizer is used to train the integrated target detection model. The learning rate and weight decay value range of the AdamW optimizer are both 10 -5 to 10 -3 between.

10. The synthetic aperture radar image target detection method according to claim 1, characterized in that: The text prompt includes at least one category.

11. The synthetic aperture radar image target detection method according to claim 1, characterized in that: The detection result also includes the confidence between each identified target and the corresponding category.

12. A synthetic aperture radar image target detection device, characterized in that: include: A SAR image acquisition unit, used for acquiring a synthetic aperture radar (SAR) image to be detected; A denoising unit, configured to perform denoising processing on the SAR image using a denoising network to obtain a denoised SAR image; An integrated model building unit, used to build an integrated target detection model, wherein the integrated target detection model includes a denoising network and an open vocabulary target detection model followed by the denoising network, and trains the integrated target detection model using an improved bounding box regression loss function built based on corner point distance and side overlap information; The inference unit is used to input the denoised SAR image and text prompt into the trained open vocabulary target detection model for feature extraction and target detection, and output the detection result, which includes the bounding box and category of the identified target.

13. An electronic device, characterized in that: The device comprises a memory and a processor, wherein the memory is used to store computer instructions executable on the processor, and the processor is used to implement the method according to any one of claims 1 to 11 when executing the computer instructions.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 11 is implemented.

Images