Single-view complex SAR (Synthetic Aperture Radar) image target detection method and system

Abstract

The invention discloses a single-view complex SAR image target detection method and system, and the method comprises the following steps: S1, SAR data reading: carrying out the multi-view processing and image conversion of input SLC data; s2, SAR image preprocessing: slicing an input large-size SAR image, and preliminarily screening out slices which do not contain a target; s3, performing SAR image target detection, and selecting an optimal algorithm model according to a specific application scene; s4, SAR image post-processing, including slice detection result integration, multi-path detection result fusion and prior knowledge-based false alarm suppression. According to the method and the system, noise and details can be balanced, the false alarm is suppressed by using priori knowledge, the detection performance and the robustness are improved, and the method and the system are enabled to be more suitable for complex and changeable application environments.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing technology and image processing technology, and in particular to a single-view complex SAR image target detection method and system based on prior knowledge. Background Art

[0002] With the rapid development of remote sensing technology, Synthetic Aperture Radar (SAR), as an important remote sensing detection method, has shown broad application prospects in target detection, military reconnaissance, terrain mapping, environmental monitoring and other fields. SAR radar generates high-resolution radar images by emitting microwave signals and receiving the echoes reflected from the ground. Compared with optical images, SAR radar is not affected by weather conditions and lighting, and can penetrate obstructions such as clouds and haze to obtain clear images of ground targets. Therefore, SAR radar has unique advantages in complex environments such as cities, forests, oceans and mountains. However, SAR images also have their unique characteristics and challenges. First, the target features in complex scenes are not obvious, and there are many interference factors, such as terrain undulations, vegetation coverage, building shadows, etc. These factors will seriously affect the accuracy and robustness of target detection. Secondly, the speckle noise, signal attenuation and Doppler shift phenomena unique to SAR images will further increase the difficulty of target detection.

[0003] Traditional SAR target detection methods are usually based on image processing techniques, such as filters, edge detection, and feature extraction. These methods require manual design and feature extraction, which can make full use of human prior knowledge and domain knowledge to a certain extent and improve the pertinence of the model. However, when faced with the challenges of complex scenes and diverse targets, manually designed features are often difficult to fully and accurately describe complex targets in SAR images, which limits the detection performance.

[0004] The application of deep learning, especially convolutional neural networks (CNN), makes SAR image target detection more accurate and efficient. Its advantage lies in its powerful feature extraction capability. It no longer needs to manually design features, can process complex image information, learn the implicit internal connections in the data, and has strong generalization ability. However, the decision-making process of deep learning models is relatively complex, and it is difficult to intuitively explain its decision basis, which to a certain extent limits the performance of the model in scenarios with clear decision basis. In addition, the performance of deep learning models is highly dependent on the quantity and quality of training data. If the training data is insufficient or biased, it may lead to degraded model performance or poor generalization ability.

[0005] Single Look Complex (SLC) data is an important output of SAR imaging processing. It contains the phase and amplitude information of the SAR echo signal and is the basis for subsequent target detection and other tasks. In SAR target detection applications based on deep learning, models that directly use SLC data as input are not common, because the SLC format has a large amount of data, requires high computing resources, and has limited processing speed, making it difficult to apply to scenarios with high real-time requirements. However, if target detection is performed only based on the compressed amplitude image, some target features will be lost, thus affecting the accuracy of target detection. Summary of the invention

[0006] In order to solve the problems of high noise and false alarm in SAR target detection in complex scenes, the present invention proposes a single-view complex SAR image target detection method and system based on prior knowledge. The method can balance noise and details, and use prior knowledge to suppress false alarms, improve detection performance and robustness, and make it more adaptable to complex and changeable application environments.

[0007] To achieve the above object, the present invention provides a single-view complex SAR image target detection method, the method comprising the following steps:

[0008] S1: SAR data reading, multi-view processing and image conversion of input SLC data;

[0009] S2: SAR image preprocessing: slicing the input large-size SAR image and preliminarily filtering out the slices that do not contain the target;

[0010] S3: SAR image target detection, select the optimal algorithm model according to the specific application scenario;

[0011] S4: SAR image post-processing, including the integration of slice detection results, the fusion of multi-channel detection results, and false alarm suppression based on prior knowledge.

[0012] Further, step S1 comprises:

[0013] S1-1: Read SAR SLC data, including two channels of real part and imaginary part. The data is two-dimensional complex matrix data, and each pixel includes a real part and an imaginary part;

[0014] S1-2: Perform multi-view processing on the SLC complex data, perform window sliding average on the real part and the imaginary part respectively, and obtain the output matrix;

[0015] S1-3: converting the complex data after multi-view processing into a jpg format image, and obtaining a plurality of jpg format images based on different conversion thresholds.

[0016] Further, step S2 comprises:

[0017] S2-1: input the multiple jpg format images generated in step S1, and slice each image using a slicing algorithm to obtain a slice set;

[0018] S2-2: Filter the slice set through the screening algorithm, retain the slices containing the target, and delete the slices that are all background;

[0019] S2-3: Use a denoising algorithm to denoise the retained target slices to obtain a denoised slice set.

[0020] Further, step S3 includes:

[0021] S3-1: Deploy appropriate target detection algorithms and models according to application scenarios;

[0022] S3-2: Input the multi-channel slice set outputted in step S2, and perform inference to obtain a multi-channel detection result.

[0023] Furthermore, in step S3-2, each detection result includes the coordinates of the upper left corner of the target box, the width and height of the target box, the target category and the confidence level.

[0024] Further, step S4 includes:

[0025] S4-1: input the multi-channel detection results generated in step S3, integrate the slice detection results of each channel, and restore the detection frame to the original large image;

[0026] S4-2: Fuse the detection results of multiple channels, perform non-maximum suppression on the target frame, filter out overlapping frames, and finally obtain a set of detection results;

[0027] S4-3: Perform false alarm suppression on the detection result obtained in S4-2.

[0028] Further, step S4-3 includes: when the physical size of the target is known, screening is performed according to the pixel size and image resolution of the target, and false targets that are too large or too small are filtered out to obtain the final detection result.

[0029] On the other hand, the present invention provides a single-view complex SAR image target detection system, which includes a data reading module, a preprocessing module, a detection module and a post-processing module.

[0030] Further, the data reading module is used to perform multi-view processing and image conversion on the input SLC data;

[0031] The preprocessing module includes image slicing, denoising and screening algorithms, which slice the input SAR image and screen out the slices that do not contain the target;

[0032] The detection module includes a variety of detection models, and the appropriate algorithm model is selected according to actual needs;

[0033] The post-processing module includes a slice integration algorithm, a multivariate fusion algorithm and a false alarm suppression algorithm based on prior knowledge.

[0034] Beneficial effects:

[0035] The present invention first performs multi-view processing on the input SLC data, and utilizes the statistical characteristics of SLC complex data to effectively reduce speckle noise. Then, the SLC data is converted based on multiple thresholds to obtain multi-channel images, which can retain more image details while suppressing image noise, and provide diversified input for the detection module. Finally, in the post-processing module, in addition to fusing multi-channel detection results, false alarm suppression can be performed on the detection results based on prior knowledge. The problem of high noise and high false alarm in SAR target detection under complex scenes is effectively solved, and the accuracy and reliability of detection are significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flow chart of the single-view complex SAR image target detection method and system architecture according to the present invention;

[0037] Figure 2 is a flow chart of an image conversion algorithm in an image reading module according to the present invention;

[0038] Figure 3 It is a flow chart of a priori knowledge false alarm suppression algorithm in a data post-processing module according to the present invention;

[0039] Figure 4 is a grayscale image of SLC data according to the present invention;

[0040] Figure 5 is the multi-view output result I according to the present invention m Amplitude image;

[0041] Figure 6 An example of an image normalized according to different cutoff values ​​of the present invention;

[0042] Figure 7 is an example of the detection result according to the present invention;

[0043] Figure 8 is a horizontal frame schematic diagram according to the present invention;

[0044] Fig. 9 Schematic diagram of a rotating frame according to the present invention. DETAILED DESCRIPTION

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

[0046] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0047] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0048] The following combination Figures 1 to 9 The specific embodiments of the present invention are described in detail. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0049] Aiming at SAR target detection in complex scenes, this paper proposes a single-view complex SAR image target detection method and system based on prior knowledge. The framework of the method and system takes SLC format data as input, integrates traditional machine learning algorithms and deep learning algorithms, and aims to improve the performance and robustness of SAR target detection.

[0050] like Figure 1 As shown, the single-view complex SAR image target detection system according to the present invention includes a data reading module, a preprocessing module, a detection module and a post-processing module.

[0051] The data reading module performs multi-view processing and image conversion on the input SLC data. It effectively reduces half-point noise, improves the image signal-to-noise ratio, and lays the foundation for target detection in complex scenes. The conversion algorithm clips and normalizes pixels based on multiple thresholds to generate multi-channel SAR images. Selecting multiple thresholds can balance the needs of noise suppression and detail retention, providing diversified inputs for the detection module. The data reading module converts the original SLC data with different thresholds to provide multiple inputs for the detection module.

[0052] The preprocessing module includes algorithms such as image slicing, denoising, and screening, which slice the input large-size SAR image and preliminarily screen out slices that do not contain targets, which can not only significantly improve processing efficiency but also reduce false alarms.

[0053] The detection module includes a variety of detection models, which can be selected according to actual needs. The detection module is an algorithm library containing a variety of models, and the appropriate algorithm model can be flexibly selected according to the specific application scenario.

[0054] The data post-processing module includes a slice integration algorithm, a multi-channel fusion algorithm and a false alarm suppression algorithm based on prior knowledge. The data post-processing module includes the integration of slice detection results, the fusion of multi-channel detection results and the false alarm suppression based on prior knowledge, which improves the detection performance in complex scenes and reduces false alarms caused by misidentification.

[0055] The single-view complex SAR image target detection method according to the present invention comprises the following steps:

[0056] Step S1: SAR data reading, multi-view processing and image conversion of input SLC data.

[0057] S1-1: Read SAR SLC data, including two channels: real part and imaginary part.

[0058] S1-2: Perform multi-view processing on SLC complex data.

[0059] S1-3: converting the complex data after multi-view processing into a jpg format image, and obtaining a plurality of jpg format images based on different conversion thresholds.

[0060] Step S2: SAR image preprocessing: slicing the input large-size SAR image and preliminarily filtering out the slices that do not contain the target.

[0061] The multiple data outputted in step S1 are processed as follows:

[0062] S2-1: Input the multiple jpg format images generated in step S1, and slice each image using a slicing algorithm to obtain a slice set.

[0063] S2-2: Roughly screen the slice set through the screening algorithm, retain the slices containing the target, and delete the slices that are all background.

[0064] S2-3: Use a denoising algorithm to denoise the retained target slices to obtain a denoised slice set.

[0065] Step S3: SAR image target detection, flexibly select the appropriate algorithm model according to the specific application scenario.

[0066] S3-1: Deploy appropriate target detection algorithms and models based on application scenarios.

[0067] S3-2: Input the multi-channel slice set output by step S2, perform inference to obtain multi-channel detection results, where each channel of detection results includes the upper left corner coordinates of the target box, the width and height of the target box, the target category, and the confidence level.

[0068] Step S4: SAR image post-processing, including the integration of slice detection results, the fusion of multi-channel detection results, and false alarm suppression based on prior knowledge.

[0069] S4-1: Input the multi-channel detection results generated in step S3, integrate the slice detection results of each channel, and restore the detection frame to the original large image.

[0070] S4-2: Fuse the detection results of multiple channels, perform non-maximum suppression on the target box, filter out overlapping boxes, and finally obtain a set of detection results.

[0071] S4-3: False alarm suppression is performed on the detection results obtained in S4-2 based on prior knowledge. For example, when the physical size of the target is known, the target is screened according to its pixel size and image resolution, and false targets that are too large or too small are filtered to obtain the final detection result.

[0072] The present invention first performs multi-view processing on the input SLC data, and utilizes the statistical characteristics of SLC complex data to effectively reduce speckle noise. Then, the SLC data is converted based on multiple thresholds to obtain multi-channel images, which can retain more image details while suppressing image noise, and provide diversified input for the detection module. Finally, in the post-processing module, in addition to fusing multi-channel detection results, false alarm suppression can be performed on the detection results based on prior knowledge. The problem of high noise and high false alarm in SAR target detection under complex scenes is effectively solved, and the accuracy and reliability of detection are significantly improved.

[0073] The implementation process of the present invention is described in detail below in conjunction with specific embodiments.

[0074] Specifically, step S1: SAR data reading

[0075] S1-1: Read the SLC data of a SAR image I s As the input of the entire framework, due to the SAR coherent imaging mechanism, its image data is naturally a two-dimensional complex matrix data, and each pixel contains the real part I Re and the imaginary part I Im Two parts, that is, each pixel I = I Re +JI Im , where j is an imaginary unit, such as Figure 4 Shown are (4a) amplitude images (4b) Real part image (4c) Imaginary image

[0076] S1-2: Multi-view processing is performed on the SLC complex data, and the real part and the imaginary part are windowed and averaged respectively. The value is generally 3×3 window or 5×5 window, which can effectively reduce the phase noise and obtain the output matrix I m ,like Figure 5 shown.

[0077] S1-3: Convert the complex data after multi-view processing into jpg format images, and obtain multiple jpg format images based on different conversion thresholds, such as Figure 2 As shown, the specific conversion process is as follows:

[0078] S1-3-1: Statistics I m The standard deviation of all data is obtained by outputting std .

[0079] S1-3-2: Calculation of complex data I m Amplitude Get output I abs .

[0080] S1-3-3: According to different thresholds t1, t2, t3, I abs Truncate, that is, I abs >t i I std The pixels are uniformly taken as t i I std , i=1,2,3,t i According to experience, 4, 6, and 7 are generally used to obtain multi-channel output I s1 ,I s2 ,I s3 The main purpose of truncation is to prepare for the subsequent normalized image, which can obtain images with different contrasts;

[0081] S1-3-4: I s1 ,I s2 ,I s3Normalize to the standard 8-bit grayscale image, i.e. 0-255, and get the final output I of step S1 j1 ,I j2 ,I j3 The three truncated and normalized image output results are as follows Figure 6 As shown, in Figure (6a) I j1 (t1=4), I in Figure (6b) j2 (t2=6, I in Figure (6c) j3 (t3=7).

[0082] Step S2: SAR image preprocessing

[0083] S2-1: According to the specified slice size, j1 ,I j2 ,I j3 The image is sliced ​​to obtain three sets of slices

[0084] S2-2: Use a screening algorithm to roughly screen each set of slices. For example, based on the principle that target pixel values ​​are high and background pixel values ​​are low, compress the slices to a uniform size, and then count the sum of the pixel values ​​of each slice. If it is greater than a manually set threshold (empirical value), it is considered to contain the target slice, and if it is less than a threshold, it is considered to be the background. Keep the slices containing the target and delete the slices that are all background, and get the filtered slice set

[0085] S2-3: Select the appropriate denoising algorithm d in the denoising algorithm library according to actual needs. The denoising algorithm should effectively reduce the noise of the slice and help improve the detection performance. Perform denoising.

[0086] Step S3: SAR image target detection

[0087] S3-1: Select and deploy appropriate target detection algorithms and models m based on whether the actual application scenarios and the data used for model training belong to similar scenarios.

[0088] S3-2: Infer the multi-channel slice set output in step S2, and obtain the multi-channel detection results end-to-end through the target detection network

[0090] is the target detection frame. The detection frame content includes: target center coordinates (x, y), target rectangular frame height and width (h, w). The target can be marked with a rectangular frame in the image through the detection frame. The detection result example is as follows Figure 7 shown.

[0091] Step S4: SAR image post-processing

[0092] S4-1: Integrate the slice detection results of each channel, restore the detection frame to the original large image, and obtain the three-channel detection results T j1 ,T j2 ,T j3 .

[0093] S4-2: Fuse the three detection results, perform non-maximum suppression on the target frame, filter out overlapping frames, and finally obtain a set of detection results T j {t j1 ,t j2 ,t j3 ,…,t jp}.

[0094] S4-3: According to various possible situations, develop a variety of algorithm plug-ins based on prior knowledge, and in the process of use, according to the actual application scenario information, select the appropriate algorithm p in the prior knowledge algorithm library by configuring the system parameter file. For example, if the physical size of the target to be detected is known, manually configure the system parameter file and select the false alarm suppression algorithm based on size, such as Figure 3 As shown in the figure, the specific process of the false alarm suppression algorithm is as follows:

[0095] S4-3-1: Traverse each detection frame and find the outline of the target in the frame;

[0096] S4-3-2: Calculate the minimum circumscribed rectangle that can enclose the contour and obtain the rotation box O of the target j {o j1 ,o j2 ,o j3 ,…,o jp}, so that the detection box can describe the target more accurately. Figure 8 A horizontal box is shown; Fig. 9 A rotating frame is shown.

[0097] S4-3-3: Traversal O j {o j1 ,o j2 ,o j3 ,…,o jp}, count the pixel size of the target. For example, if the height and width of the target rectangle contained in the detection frame are (h, w), then the pixel size of the target is h*w S4-3-4: Calculate the physical size of the detection frame based on the image resolution. For example, if the image resolution is 0.2m*0.2m, it can be understood that the interval between each pixel is 0.2m. At this time, the physical size of the target is h*w*0.2*0.2

[0098] S4-3-5: Filter out the false targets that are too large or too small, and get the final detection result O{o1,o2,o3,…,o r For example, it is known that the maximum physical size of an object in a real scene is Area max Minimum Area min , then h×w×0.2×0.2>Area max or h×w×0.2×0.2 <Area min The target will be filtered out.

[0099] Any process or method description in the flowchart of the present invention or described in other ways herein can be understood as a module, segment or part of a code including one or more executable instructions for implementing the steps of a specific logical function or process, which can be implemented in any computer-readable medium for use by an instruction execution system, device or equipment, and the computer-readable medium can be any medium containing storage, communication, propagation or transmission programs for use by execution systems, devices or equipment, including read-only memories, magnetic disks or optical disks, etc.

[0100] In the description of this specification, the description with reference to the terms "embodiment", "example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. In addition, those skilled in the art can combine or combine different embodiments or examples described in this specification and the features therein without causing any contradiction.

[0101] Although the above content has shown and described the embodiments of the present invention, it can be understood that the above embodiments are exemplary and cannot be understood as limitations of the present invention. Ordinary technicians in the field can change, modify, replace, modify and other update operations on the above embodiments within the scope of the present invention.

Claims

1. A single-view complex SAR image target detection method, characterized in that: The method comprises the following steps: S1: SAR data reading, multi-view processing and image conversion of input SLC data; S2: SAR image preprocessing: slicing the input large-size SAR image and preliminarily filtering out the slices that do not contain the target; S3: SAR image target detection, select the optimal algorithm model according to the specific application scenario; S4: SAR image post-processing, including the integration of slice detection results, the fusion of multi-channel detection results, and false alarm suppression based on prior knowledge.

2. The single-view complex SAR image target detection method according to claim 1, characterized in that: Step S1 includes: S1-1: Read SAR SLC data, including two channels of real part and imaginary part. The data is two-dimensional complex matrix data, and each pixel includes a real part and an imaginary part; S1-2: Perform multi-view processing on the SLC complex data, perform window sliding average on the real part and the imaginary part respectively, and obtain the output matrix; S1-3: converting the complex data after multi-view processing into a jpg format image, and obtaining a plurality of jpg format images based on different conversion thresholds.

3. The single-view complex SAR image target detection method according to claim 1, characterized in that: Step S2 includes: S2-1: input the multiple jpg format images generated in step S1, and slice each image using a slicing algorithm to obtain a slice set; S2-2: Filter the slice set through the screening algorithm, retain the slices containing the target, and delete the slices that are all background; S2-3: Use a denoising algorithm to denoise the retained target slices to obtain a denoised slice set.

4. The single-view complex SAR image target detection method according to claim 1, characterized in that: Step S3 includes: S3-1: Deploy appropriate target detection algorithms and models according to application scenarios; S3-2: Input the multi-channel slice set outputted in step S2, and perform inference to obtain a multi-channel detection result.

5. The single-view complex SAR image target detection method according to claim 4, characterized in that: In step S3-2, each detection result includes the coordinates of the upper left corner of the target box, the width and height of the target box, the target category and the confidence level.

6. The single-view complex SAR image target detection method according to claim 1, characterized in that: Step S4 includes: S4-1: input the multi-channel detection results generated in step S3, integrate the slice detection results of each channel, and restore the detection frame to the original large image; S4-2: Fuse the detection results of multiple channels, perform non-maximum suppression on the target frame, filter out overlapping frames, and finally obtain a set of detection results; S4-3: Perform false alarm suppression on the detection result obtained in S4-2.

7. The single-view complex SAR image target detection method according to claim 6, characterized in that: Step S4-3 includes: when the physical size of the target is known, screening is performed according to the pixel size and image resolution of the target, filtering out false targets that are too large or too small, and obtaining the final detection result.

8. A single-view complex SAR image target detection system, characterized in that: The system is used to implement the single-look complex SAR image target detection method according to any one of claims 1-7.

9. The single-view complex SAR image target detection system according to claim 8, characterized in that: The system includes a data reading module, a pre-processing module, a detection module and a post-processing module.

10. The single-view complex SAR image target detection system according to claim 9, characterized in that: The data reading module is used to perform multi-view processing and image conversion on the input SLC data; The preprocessing module includes image slicing, denoising and screening algorithms, which slice the input SAR image and screen out the slices that do not contain the target; The detection module includes a variety of detection models, and the appropriate algorithm model is selected according to actual needs; The post-processing module includes a slice integration algorithm, a multivariate fusion algorithm and a false alarm suppression algorithm based on prior knowledge.

Images