Tiny target detection method based on vehicle-mounted SAR image

Abstract

The invention discloses a tiny target detection method based on a vehicle-mounted SAR (Synthetic Aperture Radar) image. And extracting multiple types of feature parameters of candidate targets detected by CFAR for representing candidate features, judging the candidate targets based on a multi-feature weighted score judgment mechanism, calculating a comprehensive score value for each candidate target, and comparing the comprehensive score value with a preset judgment threshold. And when the comprehensive score value of the candidate target meets a judgment condition, judging the candidate target as an effective target. According to the invention, reliable detection and effective discrimination of a tiny target can be realized in a complex clutter environment, and while the detection sensitivity is ensured, the false detection number is significantly reduced, so that the detection accuracy and stability are improved.

Description

Technical Field

[0001] This invention relates to a method for detecting small targets based on vehicle-mounted SAR images, belonging to the field of synthetic aperture radar technology. Background Technology

[0002] Synthetic Aperture Radar (SAR) is an active microwave remote sensing technology that can acquire the two-dimensional microwave scattering characteristics of a target area regardless of lighting conditions, day and night, and has been widely used in surface imaging, target detection, and monitoring. Vehicle-mounted SAR is a radar imaging method that mounts the SAR system on a ground vehicle platform. It features flexible deployment, high repetition rate, and suitability for short-range high-resolution imaging, making it particularly suitable for applications such as the detection of small targets on road surfaces. However, due to the short imaging distance and large variation in observation angle of vehicle-mounted SAR, radar echoes are easily affected by factors such as surface structures, markings, and seams in the area to be detected. This results in the presence of many scattering structures outside the target area in the imaging results, easily generating false alarms and affecting the detection of actual small targets. Existing target detection methods based on vehicle-mounted SAR mostly employ constant false alarm rate (CFAR) detection algorithms to perform target detection on radar imaging data.

[0003] However, in real-world environments, due to the complex surface scattering characteristics of the target area, the presence of sidelobe interference and noise points, relying solely on CFAR detection can easily lead to a large number of false detections, resulting in insufficient reliability of the detection results and increasing the burden of manual verification. Therefore, there is an urgent need for a target detection method that can effectively suppress false detections while ensuring detection sensitivity. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes a method for detecting small targets based on vehicle-mounted SAR images. This method can reliably detect and effectively distinguish small targets in complex cluttered environments, significantly reducing the number of false detections while maintaining detection sensitivity, thereby improving the accuracy and stability of detection.

[0005] This invention is achieved through the following technical solution: A method for detecting small targets based on vehicle-mounted SAR images, the method comprising the following steps: Step 1: The vehicle-mounted platform acquires SAR echo data and completes basic imaging processing. The constant false alarm rate (CFAR) detection method is used to perform initial detection on the imaging results to obtain candidate points or candidate regions of suspected small targets. Step 2: Spatial correlation and agglomeration fusion are performed on the candidate targets obtained from the initial detection to eliminate over-detection, and the corresponding local image regions are extracted with the agglomerated candidate targets as the center. Step 3: Extract multidimensional features representing the morphology and salience of the target within the local region of the candidate target, including target shape features and brightness contrast features between the target and the background; Step 4: Weighted fusion of the multidimensional features of each candidate target to form a comprehensive score, and combined with threshold decision to achieve refined screening and confirmation of effective small targets.

[0006] Beneficial effects: Effectively reduce false detection rate and improve detection reliability This invention introduces a multi-feature joint decision mechanism combining shape and brightness contrast features, effectively suppressing false targets caused by complex backgrounds and clutter on the basis of constant false alarm initial detection, thus significantly reducing the false alarm rate; 2. Improve testing stability and engineering applicability: This invention employs a feature analysis and comprehensive evaluation method based on local regions, enhancing the algorithm's robustness to different scenarios, clutter conditions, and changes in imaging parameters, reducing reliance on single thresholds or parameter settings, and making it suitable for real-time vehicle applications; 3. Construct a refined framework for detecting small targets: This invention establishes a refined detection process consisting of initial detection, candidate target aggregation, and multi-feature weighted decision, achieving stable and scalable detection of small targets on vehicle-mounted SAR. Attached Figure Description

[0007] Figure 1 The diagram below shows the overall process of a small target detection method based on vehicle-mounted SAR images in an embodiment of the present invention. Figure 2 (a) A schematic diagram of radar imaging results obtained by collecting data from a vehicle-mounted SAR system and processing the images; (b) A schematic diagram of the location of the corresponding real target, where the real foreign object is marked with a red circle. Figure 3 This is a schematic diagram of the distribution of candidate targets obtained after initial detection and aggregation processing of radar imaging results in an embodiment of the present invention, wherein candidate foreign objects are marked with red circles; Figure 4 The above are schematic diagrams of local images of candidate targets extracted from SAR imaging results in embodiments of the present invention, and schematic diagrams of the results of binarization and connected component segmentation of the corresponding local images of candidate targets. Figure 5 The diagram below is a schematic diagram of the multidimensional feature distribution of candidate targets in an embodiment of the present invention, showing the area feature distribution, eccentricity feature distribution, solidity feature distribution and brightness contrast feature distribution of candidate targets, respectively. Figure 6The diagram below illustrates the target detection results obtained after judging candidate targets based on a multi-feature weighted scoring decision mechanism in an embodiment of the present invention, wherein the judged targets are marked with red circles. Detailed Implementation

[0008] The specific implementation of the present invention will be described below with reference to the accompanying drawings and embodiments. A schematic diagram of the overall process of the vehicle-mounted SAR-based small target detection method is shown below. Figure 1 As shown.

[0009] Step 1: The vehicle-mounted platform acquires SAR echo data and completes basic imaging processing. The constant false alarm rate (CFAR) detection method is used to perform initial detection on the imaging results to obtain candidate points or candidate regions of suspected small targets. The vehicle-mounted SAR system is used to scan the area to be detected, acquire radar echo data, and perform imaging processing on the radar echo data to obtain radar imaging results.

[0010] Let the radar imaging result be represented as a two-dimensional matrix: (1) in, Indicates the azimuth sampling index. This represents the distance-oriented sampling index.

[0011] The radar imaging results are subjected to initial target detection processing. The initial detection can be implemented using the constant false alarm rate (CFAR) detection method to obtain an initial set of detection points that meet the detection conditions. (2) in, Indicates the first The location of a suspected target detection point in radar imaging This represents the initial number of detection points. The initial detection is used to acquire potential targets as completely as possible while ensuring detection sensitivity.

[0012] Step 2: Spatial correlation and agglomeration fusion are performed on the candidate targets obtained from the initial detection to eliminate over-detection, and the corresponding local image regions are extracted with the agglomerated candidate targets as the center. For the initial set of detection points obtained in step one For detection points that are spatially adjacent or related, agglomeration processing is performed to merge multiple adjacent detection points into a single candidate target, resulting in a candidate target set: (3) in, Indicates the first The central location of each candidate target .

[0013] For each candidate target, in the radar imaging results Extracting local radar image regions based on their center position: (4) in, This represents a local window region surrounding the center of the candidate target, used for subsequent feature analysis.

[0014] Step 3: Extract multidimensional features representing the morphology and salience of the target within the local region of the candidate target, including target shape features and brightness contrast features between the target and the background; The local radar image region is a window area containing the candidate target and its surrounding background, used to limit the calculation range. Since the window area contains both target scattering and background scattering, the local image needs to be segmented to obtain the region corresponding to the target. The local radar image region obtained in step two... Segmentation is performed to obtain the target regions corresponding to the candidate targets. Let the... Each candidate target region is represented as: (5) Based on the target region, multiple feature parameters are extracted to characterize the properties of the candidate targets, and feature vectors of the candidate targets are constructed: (6) The feature parameters include at least the following: Shape features: used to describe the geometric shape of the target region; Brightness or contrast features: used to describe the scattering differences between a target area and its surrounding background; Size-related features: used to constrain the spatial scale of candidate targets.

[0015] By extracting multiple features, candidate foreign objects can be fully characterized in terms of geometric structure and radar scattering characteristics.

[0016] Step 4: Weighted fusion of the multidimensional features of each candidate target to form a comprehensive score, and combined with threshold decision to achieve refined screening and confirmation of effective small targets; Since different feature parameters have different dimensions and value ranges, the feature parameters extracted in step three are normalized to obtain normalized feature vectors: (7) in, The normalization method can be implemented using linear mapping, piecewise mapping, or saturated mapping. Based on this, a multi-feature weighted scoring model for candidate targets is constructed: (8) in, For the first The overall score of each candidate target For the first The weight coefficients corresponding to the class features, and .

[0017] The overall score With preset decision threshold Compare and construct the target decision rule: (9) The multi-feature weighted scoring model described above enables a unified quantitative evaluation of candidate targets, thereby effectively improving the accuracy and stability of target detection in complex backgrounds and cluttered environments.

[0018] Example This embodiment uses a vehicle-mounted SAR platform to detect the area to be detected as an example to verify the effectiveness of the small target detection method based on vehicle-mounted SAR proposed in this invention.

[0019] In this embodiment, the vehicle-mounted SAR platform travels along a predetermined route, continuously scans the area to be detected with radar, acquires radar echo data, and obtains radar imaging results through imaging processing, such as... Figure 2 As shown in (a). The radar imaging scene contains several real targets. In this embodiment, there are four targets, and their spatial positions are as follows. Figure 2 As shown in (b). The radar imaging scene also includes background components such as scattering from the surface structure of the area to be detected, sidelobe interference, and noise scattering.

[0020] First, initial target detection processing is performed on the imaging results to extract suspected target points that meet the detection conditions from the radar imaging. Spatial aggregation processing is then applied to the detection points obtained from the initial detection, merging multiple spatially adjacent detection points into a single candidate target, thus forming a candidate target set. After aggregation processing, the number of candidate targets is less than the number of initial detection points, and includes both real targets in this embodiment and pseudo-targets generated by background scattering, as shown in the distribution diagram. Figure 3 As shown.

[0021] For each candidate target, a local radar image region is extracted and segmented to obtain the corresponding target region. A schematic diagram of the local image of the candidate target and the segmented result is shown below. Figure 4As shown in the diagram, based on the target region, multiple feature parameters are extracted to characterize the candidate target's properties. These include shape features describing the target's geometry, such as eccentricity, fill factor, and saturation; brightness or contrast features describing the difference between the target region and the background scattering; and size features constraining the target's spatial scale, such as area. A schematic diagram of the distribution of these multiple feature parameters characterizing the candidate target's properties is shown below. Figure 5 As shown.

[0022] The extracted feature parameters are normalized, and a multi-feature weighted scoring model is constructed based on preset weight coefficients to calculate a comprehensive score for each candidate target. The comprehensive score is compared with a preset decision threshold. If the comprehensive score of a candidate target meets the decision criteria, it is determined to be a valid target; otherwise, it is determined to be a false detection and is eliminated.

[0023] Through the above-described decision-making process, the target in this embodiment is selected and determined from the candidate target set, achieving accurate target detection. The detection result is as follows: Figure 6 As shown.

[0024] The initial detection phase outputs a significant number of false positives among the candidate targets. After multi-feature weighted decision-making, the number of false positives is significantly reduced. A comparative evaluation is conducted between the initial detection method and the method proposed in this invention. Evaluation metrics include the number of correctly detected targets, the number of missed detections, and the number of false positives. The evaluation results are shown in Table 1. With the significant reduction in the number of false positives, visible light, infrared, and laser detection equipment can be used for further effective target detection and determination, improving detection efficiency and ensuring engineering feasibility.

[0025] Table 1 Performance Evaluation Results

[0026] In summary, the effectiveness of the proposed method has been demonstrated through target detection using measured data.

[0027] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the technical framework of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting small targets based on vehicle-mounted SAR images, characterized in that, The method includes the following steps: Step 1: The vehicle-mounted platform acquires SAR echo data and completes basic imaging processing. The constant false alarm rate (CFAR) detection method is used to perform initial detection on the imaging results to obtain candidate points or candidate regions of suspected small targets. Step 2: Spatial correlation and agglomeration fusion are performed on the candidate targets obtained from the initial detection to eliminate over-detection, and the corresponding local image regions are extracted with the agglomerated candidate targets as the center. Step 3: Extract multidimensional features representing the morphology and salience of the target within the local region of the candidate target, including target shape features and brightness contrast features between the target and the background; Step 4: Weighted fusion of the multidimensional features of each candidate target to form a comprehensive score, and combined with threshold decision to achieve refined screening and confirmation of effective small targets.

2. The method for detecting small targets based on vehicle-mounted SAR images as described in claim 1, characterized in that, In step three, the local radar image region obtained in step two is... Perform segmentation processing to obtain the target regions corresponding to the candidate targets; let the first... Each candidate target region is represented as: ； Based on the target region, multiple feature parameters are extracted to characterize the properties of the candidate targets, and feature vectors of the candidate targets are constructed: ； The feature parameters include at least the following: Shape features: used to describe the geometric shape of the target region; Brightness or contrast features: used to describe the scattering differences between a target area and its surrounding background; Size-related features: used to constrain the spatial scale of candidate targets; By extracting multiple features, candidate foreign objects can be fully characterized in terms of geometric structure and radar scattering characteristics.

3. The method for detecting small targets based on vehicle-mounted SAR images as described in claim 1, characterized in that, In step four, the feature parameters extracted in step three are normalized to obtain a normalized feature vector: ； in, The normalization method is implemented using linear mapping, piecewise mapping, or saturation mapping; based on this, a multi-feature weighted scoring model for candidate targets is constructed: ； in, For the first The overall score of each candidate target For the first The weight coefficients corresponding to the class features, and .

4. The method for detecting small targets based on vehicle-mounted SAR images as described in claim 1, characterized in that, In step four, the overall score will be... With preset decision threshold Compare and construct the target decision rule: ； The above multi-feature weighted scoring model enables a unified quantitative evaluation of candidate targets.

Images