Radiometer and radar fusion detection method based on multi-scale segmentation

Abstract

The invention provides a radiometer and radar fusion detection method based on multi-scale segmentation, and belongs to the technical field of target detection, and the method comprises the steps: carrying out the image registration of an SAIR image and an SAR image based on a salient region, and carrying out the denoising processing through employing a BM3D filtering algorithm introducing Canny edge detection; performing multi-scale super-pixel segmentation operation on the image, and performing weighted fusion based on a segmentation result in combination with a super-pixel scale and evidence credibility; based on a homologous multi-scale fusion result, in combination with an image imaging mechanism difference and a conflict source analysis result, a D-S fusion rule is improved according to evidence distance and data source reliability, and fusion stability and target detection accuracy under a high conflict condition are improved. According to the method, the high-resolution characteristic of the radar and the anti-interference capability of the radiometer can be fully fused, the target feature expression is enhanced, the background interference is inhibited, the fused image quality and the target detection performance are improved, and the method has good application value.

Description

Technical Field

[0001] This invention belongs to the field of target detection technology, and more specifically, relates to a multi-scale segmentation-based radiometer and radar fusion detection method. Background Technology

[0002] As detection technology continues to evolve towards informatization, the stability and accuracy of target detection in complex environments face challenges. Traditional imaging methods relying on single sensors are prone to problems such as inaccurate identification and high false alarm rates under interference or adverse weather conditions, making it difficult to meet the needs of practical applications.

[0003] Radar, as a commonly used active imaging sensor, possesses strong penetration capabilities and all-weather operation. However, its imaging relies on electromagnetic wave reflection, making it highly susceptible to influences from target materials and electromagnetic properties. Furthermore, it is easily interfered with in complex electromagnetic environments, leading to imaging distortion and the risk of misjudgment. In contrast, radiometers are passive imaging devices that receive thermal radiation signals from targets and the environment without emitting electromagnetic waves, offering excellent concealment and anti-interference capabilities. In particular, the SAIR (Synthetic Aperture Radiometer) can achieve efficient instantaneous imaging through antenna array interferometry, exhibiting superior stability in complex scenarios. However, radiometer imaging resolution is relatively low, making it difficult to provide clear texture and distance information. Radar and radiometers are highly complementary in terms of imaging mechanisms and characteristics. Fusing them together promises to balance resolution and anti-interference performance, improving the robustness of target detection.

[0004] Currently, due to the differences in data characteristics, resolution, and information dimensions between the two types of sensors, there are still technical challenges in achieving effective fusion. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a radiometer and radar fusion detection method based on multi-scale segmentation, thereby solving the technical problem that it is difficult to effectively fuse image information from radar and radiometer.

[0006] To achieve the above objectives, according to one aspect of the present invention, a radiometer and radar fusion detection method based on multi-scale segmentation is provided, comprising the following steps: Step 1: Perform image registration based on salient regions on the synthetic aperture radiometer (SAIR) image and the synthetic aperture radar (SAR) image respectively to obtain spatially aligned image pairs; Step 2: Denoising is performed on the registered SAIR and SAR images using the BM3D filtering algorithm incorporating Canny edge detection. Step 3: Perform multi-scale superpixel segmentation on the denoised image to obtain image segmentation results at different scales. Then, based on the segmentation results, perform weighted fusion by combining the superpixel scale and the credibility of the evidence to obtain the homogeneous multi-scale fusion result. Step four: Based on the same source multi-scale fusion results, construct the quality functions corresponding to the SAIR image and SAR image respectively, and calculate their conflict factors; and according to the magnitude of the conflict factors and the conflict source analysis results, use the improved Dempster-Shafer evidence fusion rule to perform multi-source image fusion to obtain the final fusion detection result.

[0007] Preferably, step one specifically includes: Significant regions are extracted from synthetic aperture radar (SAR) and synthetic aperture radiometer (SAIR) images based on grayscale thresholding to generate corresponding binary saliency maps. Morphological processing is then performed on these saliency maps to extract the centroid of each enclosed region as an initial feature point. The centroid coordinate expression is as follows:

[0008] Calculate the similarity between all centroid pairs in the SAR and SAIR images, and select the centroid pair with the highest similarity as the initial candidate matching points. Set a search window around the candidate matching points, and compare the saliency maps of local regions using a sliding window approach. Flatten the image patches within the window into one-dimensional binary vectors, and calculate the Hamming distance using the following formula to measure similarity:

[0009] in, This represents the Hamming distance between the saliency maps of two regions. To match the final image size after segmentation, The XOR operation is performed. The matching point with the smallest Hamming distance is selected as the optimal registration point, and the affine transformation parameters between the images are calculated based on this matching point to achieve image registration.

[0010] Preferably, step two specifically includes: Within a predefined neighborhood of the image, the smoothed image is calculated using finite difference based on the first-order partial derivative. gradient magnitude G and gradient direction .Right now:

[0011]

[0012] Based on the gradient direction, non-maximum suppression is performed on the gradient magnitude of each pixel in the image: if the gradient magnitude of the current pixel is not a local maximum along the gradient direction, then the pixel is suppressed; the gradient magnitude of each pixel is checked along the gradient direction, and if the gradient magnitude of the current pixel is not a local maximum along the gradient direction, then it is suppressed. Finally, double threshold detection and edge connection are performed.

[0013] More preferably, pixel estimation of the image is performed according to the following formula of BM3D filtering:

[0014] in The final estimate for each reference block. For SAR images, since the main noise is multiplicative, a logarithmic transformation is required to convert it into additive noise. After filtering, an inverse logarithmic transformation is performed. SAIR images, however, can be directly applied. After edge detection processing, morphological operations are used to expand the boundary region to generate an edge mask. This mask is used for subsequent edge information protection. The processed image is shown below:

[0015] in For the original image, This is the filtered image. For the strength of edge protection, This indicates that edge information is fully preserved.

[0016] Preferably, step three specifically includes: The results of step two for both the SAR and SAIR images are subjected to superpixel segmentation at multiple scales; that is, the SLIC algorithm is applied to each image for segmentation at different scale parameters. Let the original image size be *x*, and the corresponding initial superpixel block size be *x*.

[0017] In multi-scale segmentation results of images of the same type, superpixels at each scale can be regarded as relatively independent sources of evidence information because they express different levels of image features and are complementary. The mapping relationship between pixels and the basic probability at each scale is defined based on the Sigmoid function.

[0018]

[0019] In order to eliminate the cause k and To mitigate the output range differences caused by parameter variations, and ensure that the mapped values ​​lie within [0,1], the basic mapping function of this algorithm is defined as follows:

[0020] Based on the above formula, the probability mapping of the multi-scale segmentation results is completed, then the evidence support is calculated, and the evidence correction coefficient is further calculated. Based on the correction coefficient, the weighted fusion of the homologous multi-scale segmentation is performed.

[0021] Preferably, step four specifically includes: The homogeneous multi-scale fusion result obtained in step three is further processed. In the previous homogeneous multi-scale fusion process, a basic probability mapping was performed on the pixels. The resulting fusion value can represent the credible probability of the evidence source. Therefore, the basic probability assignment function is defined as follows:

[0022] in , The initial fusion result image as input I exist The value of the point. Calculate the conflict factor K:

[0023] Based on different values ​​of the threshold K, the fusion process is divided into three cases: low conflict, medium conflict, and high conflict. Under the low conflict condition, the following rules are directly applied:

[0024] In cases of moderate conflict, a certain degree of conflict exists, which can be addressed by adjusting the evidence sources in step three before fusion processing. In cases of high conflict, it indicates that a certain imaging result from the radiometer or radar is unreliable. If evidence fusion is still forcibly performed, it may lead to distortion of the fusion result and obscure useful information. Therefore, prior knowledge should be used to identify the sources of conflict, and higher credibility should be assigned to relatively reliable data sources to improve the accuracy of fusion decisions.

[0025] More preferably, the fused image is reconstructed. Based on the fusion result, the obtained fused image is:

[0026] in, Indicates the position of pixels in the merged image. Indicates the final fused and reconstructed image The mid-pixel represents the target's basic confidence level. , These represent the number of rows and columns in the fused and reconstructed image, respectively.

[0027] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: 1. The radiometer and radar fusion detection method based on multi-scale segmentation proposed in this invention performs multi-scale superpixel segmentation on the basis of filtering results to obtain image segmentation results at different scales. A fusion framework is then constructed based on these segmentation results, combining superpixel scale and evidence credibility to achieve fusion of source-based multi-scale images. The proposed multi-scale superpixel segmentation strategy can fully express multi-level image features, effectively improving the accuracy of segmentation and fusion. Since the segmentation results after multi-scale superpixel segmentation do not differ in resolution or information dimension, the fusion framework proposed in this invention can achieve effective fusion of synthetic aperture radiometer (SAIR) images and synthetic aperture radar (SAR) images.

[0028] 2. The radiometer and radar fusion detection method based on multi-scale segmentation proposed in this invention performs image registration based on salient regions for synthetic aperture radiometer (SAIR) images and synthetic aperture radar (SAR) images respectively. The image alignment accuracy is improved by introducing a salient region-based registration method.

[0029] 3. The radiometer and radar fusion detection method based on multi-scale segmentation proposed in this invention uses the BM3D filtering algorithm with Canny edge detection to denoise the registered SAIR and SAR images. By combining the BM3D filtering algorithm with Canny edge detection, efficient denoising and edge structure preservation are achieved.

[0030] 4. The radiometer and radar fusion detection method based on multi-scale segmentation proposed in this invention calculates the evidence distance between multi-source images based on the fused image results, utilizing the differences in imaging mechanisms between radiometers and radars and the analysis results of conflict sources. Furthermore, it improves the Dempster-Shafer evidence fusion rule based on the reliability of the data source. The improved DS fusion method, which is based on superpixel scale and evidence credibility weighting, significantly enhances the robustness of multi-source image fusion, especially showing stronger stability and detection capability in high-conflict scenarios, thereby improving the overall accuracy and reliability of target detection in complex environments.

[0031] Overall, this invention, by fusing image information acquired from different types of sensors, can fully integrate the high-resolution characteristics of radar with the anti-interference capability of radiometers, enhance target feature representation, suppress background interference, improve the quality of fused images and target detection performance, and has good application value. Attached Figure Description

[0032] Figure 1 This is a flowchart of the radiometer and radar fusion detection method based on multi-scale segmentation of the present invention.

[0033] Figure 2This is the original image to be registered in an embodiment of the radiometer and radar fusion detection method based on multi-scale segmentation of the present invention.

[0034] Figure 3 This is a schematic diagram of the registration process in an embodiment of the radiometer and radar fusion detection method based on multi-scale segmentation of the present invention.

[0035] Figure 4 This is a schematic diagram of the filtering effect of the radiometer image in an embodiment of the radiometer and radar fusion detection method based on multi-scale segmentation of the present invention.

[0036] Figure 5 This is a schematic diagram of the multi-scale segmentation and homogeneous multi-scale fusion results in an embodiment of the radiometer and radar fusion detection method based on multi-scale segmentation of the present invention.

[0037] Figure 6 This is a schematic diagram of a dual-target angle-adding anti-interference scene fusion in an embodiment of the multi-scale segmentation radiometer and radar fusion detection method of the present invention.

[0038] Figure 7 This is a schematic diagram of a three-target anti-interference scene fusion based on the multi-scale segmentation radiometer and radar fusion detection method of the present invention.

[0039] Figure 8 This is a schematic diagram of dual-target scene fusion in an embodiment of the radiometer and radar fusion detection method based on multi-scale segmentation of the present invention.

[0040] Figure 9 This is a comparison diagram of the fusion detection results of a dual-target anti-interference scene with added angle in an embodiment of the multi-scale segmentation radiometer and radar fusion detection method of the present invention.

[0041] Figure 10 This is a comparison diagram of the fusion detection results of three targets with added angle anti-interference scene in an embodiment of the multi-scale segmentation radiometer and radar fusion detection method of the present invention.

[0042] Figure 11 This is a comparison diagram of dual-target scene fusion detection results in an embodiment of the multi-scale segmentation radiometer and radar fusion detection method of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0044] like Figure 1As shown, this invention proposes a multi-scale segmentation-based radiometer and radar fusion detection method, which specifically includes the following steps: S1. First, based on grayscale thresholds, salient regions are extracted from both the radar and radiometer images, generating corresponding binary saliency maps. Then, morphological processing is performed on the binary saliency maps to enhance structural features, and closed regions are extracted. For each closed region, its geometric centroid position is calculated, and this centroid is used as the initial feature point. Next, the similarity index between each pair of centroids in the two images is calculated, and the pair of centroids with the highest similarity is selected as the initial candidate matching point. Within the neighborhood of this candidate point, a local region is extracted using a sliding window approach. The extracted binary image regions are expanded and quantized, and Hamming distance is used to measure the similarity between regions, thereby achieving accurate determination of the initial image registration point.

[0045] S2. After registration, the radar and radiometer images are denoised using the BM3D filtering algorithm that incorporates edge information. Specifically, first, the Canny edge detection operator is executed on the image to extract edge contour information and construct an edge mask. Then, the edge mask is introduced into the BM3D result to preserve edge information of image blocks, thus avoiding edge blurring or structural distortion caused by collaborative filtering in traditional BM3D processing.

[0046] To overcome the shortcomings of traditional single-scale superpixel segmentation, such as insufficient segmentation accuracy and limited representation of structural details, S3 proposes an image processing method based on multi-scale superpixel segmentation. Specifically, the preprocessed image undergoes SLIC superpixel segmentation at multiple scales, with different initial superpixel sizes and numbers corresponding to different scales, thereby extracting structural features at different levels from coarse to fine. This multi-scale strategy can enhance the representation of local details while preserving the global structure of the image, effectively alleviating the oversegmentation or undersegmentation problems caused by the fixed scale of single-scale segmentation.

[0047] Furthermore, after obtaining the multi-scale segmentation results, a fusion method based on superpixel scale and evidence credibility weighting is proposed. First, the segmentation results at multiple scales are treated as different pieces of evidence from the same data source, constructing a multi-source evidence fusion framework. Based on the DS evidence theory framework, a basic probability assignment (BPA) is assigned to each superpixel region. The BPA is generated based on a normalized confidence model constructed using the Sigmoid function, which can flexibly adjust the probability distribution according to the local feature information of the superpixels. During the fusion process, the relationship between superpixel scale differences and the credibility of the evidence is fully considered. A weighting mechanism is used to comprehensively evaluate the segmentation evidence at different scales. The weighting calculation process is as follows: The similarity between two sources of evidence is defined based on the Jousselme evidence distance. for: (3.1) In the formula, The distance represents the Jousselme distance between two quality functions. Since the degree of difference between pieces of evidence can be reflected by this distance metric, a larger Jousselme distance indicates greater inconsistency between the two evidence sources, thus lower similarity; conversely, a smaller distance indicates higher consistency between the two pieces of evidence, i.e., greater similarity. By considering the numerical range of the Jousselme distance, a quantitative analysis of the similarity between evidence can be performed. The range of is [0,1], which can be further expressed as: (3.2) Compared to traditional conflict factors, which require enumerating all possible fully conflicting combinations in a multi-source evidence fusion scenario, resulting in high computational complexity and lack of structural interpretability, Jousselme evidence distance achieves a quantitative description of conflict intensity by constructing a similarity matrix between sets. This method can measure the differences between different pieces of evidence in a more intuitive way while preserving the structural features of conflict information. Based on the similarity assessment results between evidence, an overall support measure for a particular piece of evidence can be further defined to help determine its credibility and necessity of retention in the fusion process. According to equation (3.2), the following can be obtained for the evidence... i The support level is: (3.3) Assumption t The superpixel segmentation scale of the evidence satisfies For image segmentation results, a larger number of segmented regions indicates more refined image resolution at a smaller scale, which helps capture local details and improves the fit to target edges and structures. Therefore, it can be considered that the number of segmented blocks is positively correlated with the level of detail in the segmentation result; the finer the segmentation, the higher the reliability of the detailed information it represents. Based on this, for... After weighting, we get: (3.4) (3.5) Then, the weighted evidence support is normalized to obtain the evidence. i The credibility is: (3.6) Equation (3.6) also reflects the importance of different sources of evidence. Based on the above equation, the correction coefficient for the source of evidence is further defined as follows: (3.7) (3.8) The evidence sources are weighted and fused based on the correction coefficients to obtain the corrected quality function: (3.9) This is the weighted fusion result of multi-scale segmentation based on the same origin.

[0048] S4 will identify the framework The focal element is defined as the target ( T ),background( B ), assuming SAIR and SAR image pixel pairs The corresponding mass function is Calculate the conflict factor K of the DS evidence theory: (4.1) Specifically, based on the threshold of K, fusion is divided into three cases: low conflict, medium conflict, and high conflict. 1) When When the fusion process is considered to be conflict-free (low-conflict), Dempster's composition rule can be used, i.e.: (4.2) (4.3) 2) When At this point, it falls under the category of moderate conflict, indicating a certain degree of conflict, but which can still be addressed through fusion adjustments. Analyzing the source of the conflict first requires determining whether it was caused by the superpixel segmentation process. To this end, the Jousselme distance between the aforementioned co-originating multi-scale superpixel evidence is calculated. Assuming a synthetic modified quality function... of k The quality functions corresponding to the evidence sources of superpixel segmentation results at each scale are as follows: The corresponding distance matrix D for: (4.4) Then, according to equation (4.4), the distance of each source of evidence to all other sources of evidence is calculated as follows: (4.5) The greater the distance between the evidence sources, the less reliable the current segmentation. Therefore, the evidence source with the largest distance is removed, and the remaining evidence is resynthesized. If the conflict disappears after this process, the Dempster synthesis rule is used for evidence fusion. If the conflict still exists, the evidence is weighted according to the reliability of the two sensors, with the more reliable radiometer given a higher weight, and finally, DS evidence fusion is performed.

[0049] 3) When In cases of high conflict, it indicates that a particular imaging result from a radiometer or radar is unreliable. Forcing the application of evidence fusion techniques in such situations not only fails to improve the quality of the fused results but may also obscure correct information. Therefore, the source of conflict should be analyzed based on prior knowledge, assigning higher credibility to relatively reliable data sources to ensure accurate decision-making.

[0050] Specifically: First, for interference from corner reflectors or sea clutter caused by differences in imaging mechanisms, SAIR, with its stronger anti-interference capabilities, is preferred. Second, for target edge contour conflicts caused by resolution differences, SAR with higher resolution is preferred.

[0051] In this embodiment of the invention, to address the problem that traditional radar is significantly affected by corner reflectors and clutter interference in complex environments, easily leading to false alarms, this embodiment provides a fusion detection method based on radar and radiometer images. This method effectively improves the target representation capability and anti-interference performance of the fused image by introducing image filtering, edge protection, multi-scale superpixel segmentation, and an improved evidence fusion strategy. The specific implementation process is as follows: (1) This embodiment uses simulated images from a synthetic aperture radiometer (SAIR) and synthetic aperture radar, such as... Figure 2 As shown, the results obtained further based on salient region registration are as follows: Figure 3 As shown, the size of the registered region is 700x700.

[0052] (2) Perform the preprocessing step (2) based on the obtained registered image. Figure 4 Image (a) is the SAIR image before processing. Figure 4 (b) in the image shows the processed SAIR image. After processing, the image exhibits higher uniformity and lower noise interference in the background area, resulting in a significantly better overall visual effect than the original image.

[0053] (3) Based on the multi-scale segmentation results obtained in step (3), after assigning basic probabilities, the homologous multi-scale fusion is performed according to equations (3.1) to (3.9), and the results are as follows. Figure 5 (a) Figure 5 (b) Figure 5 (c) and Figure 5 As shown in (d), image segmentation at a single scale cannot achieve complete extraction of the target region, resulting in problems such as missing information and discontinuous boundaries. In contrast, the method described in this invention integrates effective segmented regions at multiple scales to comprehensively extract target structural features, which not only improves the completeness of target extraction but also significantly suppresses background noise interference, demonstrating a clear advantage in accuracy compared to single-scale segmentation results.

[0054] (4) Based on step (4), multi-source images are fused, and the final improved evidence fusion of radiometer and radar is performed according to equations (4.1) to (4.5). The original image and the fusion result are as follows: Figure 6 , Figure 7 and Figure 8 As shown. The fusion result exhibits excellent performance in enhancing the target signal, while significantly suppressing background noise and non-target echo interference caused by corner reflectors. Compared with single-sensor observations, the fusion method described in this invention can more clearly highlight target features and improve the accuracy of target detection. To further demonstrate the advantages of this invention, a comparison of the target detection performance of this invention with other similar fusion methods is specifically provided. That is, the data of this embodiment is used uniformly, and the PR curve of target detection based on the fusion result in this embodiment is shown in Figure 1. Figure 9 , Figure 10 and Figure 11 As shown, IEF-MSPF is the proposed Improved Evidence Fusion with Multi-Scale Pre-Fusion (IEF-MSPF) based on multi-scale fusion of SAR and SAIR images, as well as multiplicative fusion, additive fusion, discrete wavelet transform (DWT) fusion, principal component analysis (PCA) fusion, and traditional DS evidence fusion.

[0055] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A radiometer and radar fusion detection method based on multi-scale segmentation, characterized in that, Includes the following steps: Image registration based on salient regions was performed on the synthetic aperture radiometer (SAIR) image and the synthetic aperture radar (SAR) image to obtain spatially aligned image pairs. For the registered SAIR and SAR images, edge detection and filtering are performed sequentially to denoise them. Multi-scale superpixel segmentation is performed on the denoised image to obtain image segmentation results at different scales. Based on the segmentation results, a weighted fusion is performed by combining the superpixel scale and the credibility of the evidence to obtain the homogeneous multi-scale fusion result. Based on the aforementioned homogeneous multi-scale fusion results, quality functions corresponding to SAIR and SAR images are constructed respectively, and their conflict factors are calculated. Based on the magnitude of the conflict factor and the analysis results of the conflict source, multi-source image fusion is performed to obtain the final fusion detection result.

2. The radiometer and radar fusion detection method based on multi-scale segmentation according to claim 1, characterized in that, Image registration based on salient regions is performed on the synthetic aperture radiometer (SAIR) images and the synthetic aperture radar (SAR) images, specifically including the following steps: Based on grayscale thresholds, salient regions are extracted from SAIR and SAR images respectively, and binary saliency maps are generated. The binary saliency map is subjected to morphological processing and closed regions are extracted; Calculate the geometric centroid of each closed region as the initial feature point; Calculate the similarity between all centroid pairs in the SAIR and SAR images, and select the pair with the highest similarity as the initial candidate matching point; Within the neighborhood of the initial candidate matching point, a local region is extracted using a sliding window and vectorized. Hamming distance is used to measure the region similarity to complete the determination of the image registration point.

3. The radiometer and radar fusion detection method based on multi-scale segmentation according to claim 2, characterized in that, The Hamming distance is calculated using the following formula to measure region similarity: in, This represents the Hamming distance between the saliency maps of two regions. To match the final image size after segmentation, The XOR operation is performed; the matching point with the minimum Hamming distance is selected as the optimal registration point, and the affine transformation parameters between the images are calculated based on the matching point with the minimum Hamming distance to achieve image registration.

4. The radiometer and radar fusion detection method based on multi-scale segmentation according to claim 3, characterized in that, After registration, edge detection and filtering are performed sequentially on the SAIR and SAR images to denoise them. The specific steps include: Perform Canny edge detection on the image to extract edge contour information and construct an edge mask; BM3D filtering is applied to image blocks with introduced edge masks to achieve noise reduction.

5. The radiometer and radar fusion detection method based on multi-scale segmentation according to claim 4, characterized in that, The step of performing multi-scale superpixel segmentation on the denoised image includes the following steps: The SLIC superpixel segmentation algorithm is used to segment the image at multiple preset scales, with different initial superpixel sizes and numbers corresponding to different scales.

6. The radiometer and radar fusion detection method based on multi-scale segmentation according to claim 5, characterized in that, The weighted fusion based on the segmentation results, superpixel scale, and evidence credibility specifically includes the following steps: The segmentation results at multiple scales are treated as different evidence from the same data source, and a basic probability assignment (BPA) based on the Sigmoid function is assigned to each superpixel region. Calculate the similarity and support between different evidence sources based on Jousselme evidence distance; Based on the number of superpixel segmentation blocks corresponding to each evidence source, the support is weighted and normalized to obtain the credibility of each evidence source. The correction coefficients of the evidence sources are calculated based on the credibility, and the evidence sources are weighted and fused to obtain the homogeneous multi-scale fusion result.

7. The radiometer and radar fusion detection method based on multi-scale segmentation according to claim 6, characterized in that, Based on the magnitude of the conflict factor and the analysis results of the conflict sources, multi-source image fusion is performed, specifically including the following steps: Calculate the conflict factor K between the corresponding quality functions of SAIR and SAR images; When the conflict factor K is less than the first threshold, it is determined to be a low-conflict case, and the Dempster synthesis rule is directly used for fusion. When the conflict factor K is greater than or equal to the first threshold and less than the second threshold, it is judged as a medium conflict situation; calculate the Jousselme evidence distance between the same source multi-scale superpixel evidence, remove the evidence source with the largest evidence distance, and re-synthesize the remaining evidence; if the conflict disappears, the Dempster synthesis rule is adopted; if the conflict still exists, the evidence is weighted according to the reliability of SAIR and SAR, and then DS evidence fusion is performed. When the conflict factor K is greater than or equal to the second threshold, it is determined to be a high-conflict situation. Based on prior knowledge, the source of conflict is analyzed, and information from more reliable data sources is prioritized.

8. The radiometer and radar fusion detection method based on multi-scale segmentation according to claim 7, characterized in that, In the aforementioned high-conflict scenario, when analyzing the source of conflict based on prior knowledge: If the conflict is caused by corner reflector or sea clutter interference, the information from the SAIR image shall be given priority. If the conflict is caused by differences in the target edge contour, the information from the SAR image should be given priority.

Images