End-to-end SAR image target detection method and system

By constructing an encoder-decoder architecture SAR image detection model using an end-to-end deep learning approach, the problems of low efficiency in speckle denoising and target detection in SAR image target detection are solved, achieving efficient and accurate image processing.

CN119006776APending Publication Date: 2024-11-22SUN YAT SEN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410969236.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing SAR image target detection methods cannot simultaneously perform speckle denoising and target detection, resulting in low processing efficiency. Furthermore, traditional methods are highly dependent on sea clutter statistical models, leading to severe error accumulation.

Method used

An end-to-end deep learning approach is adopted to construct a detection model with an encoder-decoder architecture. Combined with an information extraction layer, a speckle removal module and a target detection module are used, and the model is trained using an adaptive weight loss function to achieve speckle removal and target detection of SAR images.

Benefits of technology

It improves the speed and accuracy of SAR image processing, simplifies the model structure, and has good transferability and processing efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006776A_ABST
    Figure CN119006776A_ABST
Patent Text Reader

Abstract

The invention discloses an end-to-end SAR (Synthetic Aperture Radar) image target detection method and system. The method comprises the following steps: acquiring image data and constructing a data set; based on a coding-decoding architecture, an information extraction layer is introduced, and a speckle removal module is constructed; based on a deep learning network framework, combining the speckle removing module to construct a detection model; training the detection model based on the data set to obtain a trained detection model; and inputting a to-be-detected image into the trained detection model, and outputting a detection result. The system comprises a data set construction module, a model construction module, a model training module and a detection module. According to the invention, the speed and precision of SAR image target detection can be improved. The method can be widely applied to the field of image processing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to an end-to-end SAR image target detection method and system. Background Art

[0002] Compared with natural images, microwave remote sensing images are often accompanied by a large amount of coherent speckle noise, which greatly reduces the resolution of microwave remote sensing images and creates certain difficulties for further applications such as edge detection, image segmentation, and target recognition. Microwave remote sensing images, especially SAR (Synthetic Aperture Radar) images, coherent speckle noise is a typical multiplicative noise, which exists with the generation of SAR images. Compared with additive noise, multiplicative noise is more difficult to handle.

[0003] In SAR image ship detection, traditional methods mainly include methods based on contrast information and template matching. The Constant False Alarm Rate (CFAR) method is the most commonly used SAR image target detection technology. The CFAR method has achieved good performance in some simple scenarios, but the performance of this method depends largely on the construction of the sea clutter statistical model and the parameter estimation of the constructed model. Secondly, previous studies have processed SAR image decorrelation and target detection separately. The separate processing method requires multiple data transmissions and independent processing, resulting in low processing efficiency. In addition, there is a lack of information interaction and utilization between each link, and the processing results are limited by the local information processed at each stage, resulting in error accumulation in downstream tasks, and unable to provide fast and high-precision technology. Summary of the invention

[0004] In view of this, in order to solve the technical problem that the existing SAR image target detection method cannot simultaneously perform speckle denoising and target detection, thereby resulting in low processing efficiency, the present invention proposes an end-to-end SAR image target detection method, the method comprising the following steps:

[0005] Obtain image data and build a dataset;

[0006] Based on the encoding-decoding architecture, an information extraction layer is introduced to construct a despeckle module;

[0007] Based on the deep learning network framework, combined with the de-speckle module, a detection model is constructed

[0008] Training the detection model based on the data set to obtain a trained detection model;

[0009] The image to be tested is input into the trained detection model, and the detection result is output.

[0010] In this embodiment, SAR image despeckle and target detection are implemented in an end-to-end manner based on a deep learning method.

[0011] In some embodiments, the step of collecting the self-coherent hologram and constructing the training set, and the step of acquiring the image data and constructing the data set specifically include:

[0012] Obtain image data based on high-resolution remote sensing images;

[0013] Annotating and formatting the image data, and constructing a data set;

[0014] The data set includes original SAR images, despeckled SAR images and target labels.

[0015] In some embodiments, it also includes:

[0016] Based on the target label, the original SAR image is divided to obtain background data;

[0017] The original SAR image is enhanced according to the background data.

[0018] Through this preferred embodiment, the background data is segmented using labels, and then the image is enhanced using the background data, thereby expanding the data set.

[0019] In some embodiments, the information extraction layer includes a channel attention module and a spatial attention module.

[0020] In some embodiments, the encoding layer includes at least four convolution modules, and data is processed by maximum pooling and residual blocks when being transferred between the convolution modules in the encoding layer.

[0021] In some embodiments, the decoding layer includes at least two convolution modules, and data is subjected to inverse convolution processing when being transferred between the convolution modules in the decoding layer.

[0022] In this embodiment, an encoding-decoding network model with an information extraction layer is proposed for SAR image despeckle.

[0023] In some embodiments, during training, the loss function is formulated as follows:

[0024]

[0025] Among them, L1 is the SAR decoherence loss, L2 is the target detection loss, is the variance of the prediction result of the despeckle module, The variance of the prediction results of the target detection module.

[0026] Through this preferred embodiment, an adaptive weighted loss function is proposed to balance the two tasks of SAR image despeckle and target detection.

[0027] In some embodiments, it also includes:

[0028] The trained detection model is deployed to the edge.

[0029] The present invention also proposes an end-to-end SAR image target detection system, the system comprising:

[0030] A dataset construction module, used to obtain image data and construct a dataset;

[0031] Model building module, based on the encoding-decoding architecture, introduces the information extraction layer to build a de-coherence module; based on the deep learning network framework, combined with the de-coherence module, builds a detection model

[0032] A model training module is used to train the detection model based on the data set to obtain a trained detection model;

[0033] The detection module is used to input the image to be tested into the trained detection model and output the detection result.

[0034] The present invention also proposes an end-to-end SAR image target detection device, comprising:

[0035] at least one processor;

[0036] at least one memory for storing at least one program;

[0037] When the at least one program is executed by the at least one processor, the at least one processor implements the above-mentioned end-to-end SAR image target detection method.

[0038] Based on the above scheme, the present invention provides an end-to-end SAR image target detection method and system, which realizes SAR image de-coherence speckle and target detection by constructing a suitable network and using a training network, wherein the network includes a de-coherence speckle module and a target detection module, and a SAR image containing coherence speckle is first input through the de-coherence speckle module to obtain a clean SAR image, and then the image is input into the target detection module to realize target detection. The present invention can improve the speed and accuracy of SAR processing technology, and at the same time, the model structure is simple and portable. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flowchart of the steps of an end-to-end SAR image target detection method of the present invention;

[0040] Figure 2It is a structural block diagram of an end-to-end SAR image target detection system of the present invention;

[0041] Figure 3 It is a schematic diagram of the structure of a speckle removal module according to a specific embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0043] It should be noted that, for the convenience of description, only the parts related to the relevant invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0044] It should be understood that the "system", "device", "unit" and / or "module" used in this application is a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the word can be replaced by other expressions.

[0045] Unless the context clearly indicates an exception, the words "a", "an", "a kind" and / or "the" do not refer to the singular, but also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements. The elements defined by the sentence "includes a..." do not exclude the existence of other identical elements in the process, method, commodity or device that includes the elements.

[0046] In the description of the embodiments of the present application, "plurality" means two or more than two. The following terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features.

[0047] In addition, flow charts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, the various steps may be processed in reverse order or simultaneously. At the same time, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0048] Reference Figure 1 , is a flow chart of an optional example of an end-to-end SAR image target detection method proposed in the present invention. The method can be applied to a computer device. The detection method proposed in this embodiment may include but is not limited to the following steps:

[0049] Step S1, acquiring image data and constructing a data set;

[0050] Step S2: Based on the encoding-decoding architecture, an information extraction layer is introduced to construct a speckle removal module;

[0051] Step S3: constructing a detection model based on a deep learning network framework and in combination with the speckle removal module;

[0052] Step S4: training the detection model based on the data set to obtain a trained detection model;

[0053] Step S5: input the image to be tested into the trained detection model and output the detection result.

[0054] In some feasible embodiments, the step S1 specifically includes:

[0055] S1.1. Obtain image data based on high-resolution remote sensing images;

[0056] S1.2, annotating and formatting the image data, and constructing a data set;

[0057] The data set includes original SAR images, de-speckled SAR images and target labels.

[0058] In some feasible embodiments, the step S1 further includes:

[0059] Based on the target label, the original SAR image is divided to obtain background data;

[0060] The original SAR image is enhanced according to the background data.

[0061] Among them, the target label is used to quantify the maximum pixel value, the minimum pixel value, the mean pixel value and the number of pixels to distinguish the background SAR image. The contrast of the SAR imaging image is improved through key steps such as statistics of environmental background SAR image parameters, layered brightness enhancement of target SAR images, numerical range constraints of target SAR images, multi-layer brightness enhancement image fusion, and SAR image pixel value adjustment. The original SAR image is enhanced to supplement the data set, that is, one target label corresponds to multiple SAR images.

[0062] In some feasible embodiments, the detection model includes:

[0063] Despeckle module:

[0064] The overall architecture of the SAR despeckle module adopts a coding-decoding architecture, and adds an information extraction layer to the traditional coding-decoding architecture, such as Figure 3 shown.

[0065] The coding layer includes four modules, and the resolutions of the four modules are 256×256×64, 128×128×64, 64×64×128, and 32×32×256 respectively; the four modules are arranged in sequence, and the data transmitted between the two templates needs to be processed by maximum pooling and residual blocks.

[0066] The information extraction layer includes a channel attention module and a spatial attention module; the resolutions of the channel attention module and the spatial attention module are 32×32×512 respectively. The channel attention module is based on the correlation between channels. By learning the importance weight of each channel, it adaptively adjusts the feature responses of different channels, so that the model pays more attention to the feature channels related to the current task, thereby enhancing the feature expression ability and the generalization ability of the model. This mechanism is similar to the human visual system paying attention to the importance of different feature channels when perceiving information. The spatial attention module focuses on capturing the spatial information in the SAR data. It adaptively adjusts the feature responses of different positions by learning the correlation and weights between different positions, thereby improving the model's perception of the spatial structure and correlation in the SAR data. This mechanism is similar to the degree of attention paid by the human visual system to different positions when perceiving spatial information.

[0067] The decoding layer includes two modules, and the resolutions of the two modules are 128×128×64 and 512×512×64 respectively; the two modules are arranged in sequence, and the information extraction layer passes data to the decoding layer and the two modules pass data through inverse convolution; the modules with the same resolution in the decoding layer and the encoding layer are used as the input of the next submodule in the decoding layer through jump connections; after processing in the decoding layer, the output result, i.e., the SAR imaging result, can be obtained. When there are many types of data sets, the decoding layer can also be set to four modules, which correspond to the resolutions of the four modules in the decoding layer.

[0068] All modules use 3*3 convolution for feature extraction. All our basic modules are 3*3 convolution. The main purpose of doing this is to adapt to existing hardware. 3*3 convolution has undergone a lot of underlying optimizations on ARM processors or other mobile platform processors, which can effectively improve the efficiency of algorithm execution.

[0069] In some feasible embodiments, during the training process, the formula of the loss function is expressed as follows:

[0070]

[0071] Among them, L1 is the SAR decoherence loss, L2 is the target detection loss, is the variance of the prediction result of the despeckle module, The variance of the prediction results of the target detection module.

[0072] In the process of model training, in order to balance the two tasks of SAR despeckle and target detection, we set an adaptive weight loss function. The loss function includes two parts: SAR despeckle L1 and target detection L2. This loss function is based on the principle of homoscedastic uncertainty, that is, the uncertainty of the input data is the same, that is, the variance is equal. When the despeckle loss is large, it will have a large homoscedasticity. therefore The value of this item will become smaller; and when the target detection is small, it will have a smaller homoscedasticity therefore The value of this item will become larger. The same is true in reverse. In this way, the two tasks of SAR despeckle and target detection can be balanced during the training of the model without adding a balancing factor.

[0073] like Figure 2 As shown, an end-to-end SAR image target detection system includes:

[0074] A dataset construction module, used to obtain image data and construct a dataset;

[0075] Model building module, based on the encoding-decoding architecture, introduces the information extraction layer to build a de-coherence module; based on the deep learning network framework, combined with the de-coherence module, builds a detection model

[0076] A model training module is used to train the detection model based on the data set to obtain a trained detection model;

[0077] The detection module is used to input the image to be tested into the trained detection model and output the detection result.

[0078] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0079] An end-to-end SAR image target detection device:

[0080] at least one processor;

[0081] at least one memory for storing at least one program;

[0082] When the at least one program is executed by the at least one processor, the at least one processor implements the above-mentioned end-to-end SAR image target detection method.

[0083] The contents of the above method embodiments are all applicable to the present device embodiments. The functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0084] A storage medium stores processor-executable instructions, which are used to implement the above-mentioned end-to-end SAR image target detection method when executed by the processor.

[0085] The contents of the above method embodiments are all applicable to the present storage medium embodiments. The functions specifically implemented by the present storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0086] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. An end-to-end SAR image target detection method, characterized in that: The following steps are involved: Obtain image data and build a dataset; Based on the encoding-decoding architecture, an information extraction layer is introduced to construct a despeckle module; Based on the deep learning network framework, combined with the speckle decorrelation module, a detection model is constructed; Training the detection model based on the data set to obtain a trained detection model; The image to be tested is input into the trained detection model, and the detection result is output.

2. The end-to-end SAR image target detection method according to claim 1, characterized in that: The step of acquiring image data and constructing a data set specifically includes: Obtain image data based on high-resolution remote sensing images; Annotating and formatting the image data, and constructing a data set; The data set includes original SAR images, despeckled SAR images and target labels.

3. The end-to-end SAR image target detection method according to claim 2, characterized in that: Also includes: Based on the target label, the original SAR image is divided to obtain background data; The original SAR image is enhanced according to the background data.

4. The end-to-end SAR image target detection method according to claim 1, characterized in that: The information extraction layer includes a channel attention module and a spatial attention module.

5. The end-to-end SAR image target detection method according to claim 1, characterized in that: The coding layer includes at least four convolution modules, and data is processed by maximum pooling and residual blocks when being transmitted between the convolution modules in the coding layer.

6. The end-to-end SAR image target detection method according to claim 1, characterized in that: The decoding layer includes at least two convolution modules, and data is subjected to inverse convolution processing when being transmitted between the convolution modules in the decoding layer.

7. The end-to-end SAR image target detection method according to claim 1, characterized in that: During the training process, the formula of the loss function is expressed as follows: Among them, L1 is the SAR decoherence loss, L2 is the target detection loss, is the variance of the prediction result of the despeckle module, The variance of the prediction results of the target detection module.

8. The end-to-end SAR image target detection method according to claim 1, characterized in that: Also includes: The trained detection model is deployed to the edge.

9. An end-to-end SAR image target detection system, characterized in that: include: A dataset construction module, used to obtain image data and construct a dataset; The model building module introduces the information extraction layer based on the encoding-decoding architecture and builds the de-speckle module; Based on the deep learning network framework, combined with the de-speckle module, a detection model is constructed A model training module is used to train the detection model based on the data set to obtain a trained detection model; The detection module is used to input the image to be tested into the trained detection model and output the detection result.

10. An end-to-end SAR image target detection device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the end-to-end SAR image target detection method as described in any one of claims 1-8.