基于多层CFAR算法的超高分辨率SAR图像目标检测
发布时间：2019-03-05 13:44
【摘要】：统计建模是合成孔径雷达(SAR)图像解译必不可少的东西,不同的SAR统计模型(如K分布,高斯分布,伽马分布,对数正态分布,混合分布等)对不同的SAR地物类型(如农田,森林,草地,河流等)的建模能力各不相同。本文首先介绍了不同的统计分布及其特点,对不同的SAR图像地物类型进行统计建模,找出各种统计模型所适合的地物类型。目标检测更是SAR图像应用的重中之重,文章针对机载SAR图像中车辆检测问题结合CFAR算法提出了针对角反射器散射特点的目标检测算法,并采用真实的机载P波段和L波段SAR图像数据对算法进行了验证。超高分辨率SAR图像具有数据量大,传统CFAR算法处理时间复杂度高,目标具有一定的形态及细节的特征。针对这些特点我们提出了多层CFAR算法。算法中采用对数正态分布作为图像的统计分布模型。我们通过对整幅SAR图像采用基于对数正态分布的全局CFAR算法滤除强散射点来找出SAR图像背景区域。然后依据提取出的SAR图像背景来进一步检测舰船目标。尽管多层CFAR算法提取出较准确的舰船目标,但是依然存在很多虚警目标。我们根据先验舰船尺寸大小,对多层CFAR算法处理后的图像滤除虚警目标。由于超高SAR图像特点,滤除虚警后的目标有着不完整或者船体出现空洞的现象,我们提出了提取目标轮廓算法,并对目标轮廓进行填充来得到完整的目标。实验中使用两幅TerraSAR-X图像真实数据,分辨率为1米,分别采用多层CFAR算法及传统CFAR算法进行实验比较,结果证明我们的算法有较好的检测结果。论文得到了国家自然科学基金(No.61072106,61271302)的资助和国家“973”计划(No.2013CB329402)的支持。
[Abstract]:Statistical modeling is essential for the interpretation of synthetic aperture radar (SAR) images. Different SAR statistical models (such as K distribution, Gao Si distribution, gamma distribution, lognormal distribution, mixed distribution, etc.) relate to different types of SAR features (such as farmland, etc.). The modeling capabilities of forests, grasslands, rivers, etc., vary. In this paper, we first introduce the different statistical distribution and its characteristics, and make statistical modeling for different SAR image feature types, and find out the suitable feature types for various statistical models. Target detection is the most important in the application of SAR image. In this paper, a target detection algorithm based on the scattering characteristics of corner reflector is proposed to solve the problem of vehicle detection in airborne SAR images combined with CFAR algorithm. The real airborne P-band and L-band SAR image data are used to verify the algorithm. Ultra-high resolution SAR images have a large amount of data, the traditional CFAR algorithm processing time complexity is high, the target has a certain shape and detail characteristics. In view of these characteristics, we propose a multi-layer CFAR algorithm. The lognormal distribution is used as the statistical distribution model of the image in the algorithm. We use the global CFAR algorithm based on lognormal distribution to filter the strong scattering points for the whole SAR image to find out the background region of the SAR image. Then the background of the extracted SAR image is used to detect the ship target. Although the multi-layer CFAR algorithm can extract more accurate ship targets, there are still many false alarm targets. According to the size of a priori ship, we filter the false alarm target from the image processed by multi-layer CFAR algorithm. Due to the characteristics of ultra-high SAR images, the target after filtering false alarm is incomplete or empty in the hull. We propose an algorithm to extract the contour of the target and fill the contour to get the complete target. In the experiment, the real data of two TerraSAR-X images are used, and the resolution is 1 meter. Compared with the traditional CFAR algorithm and the multi-layer CFAR algorithm, the results show that our algorithm has better detection results. The thesis is supported by the National Natural Science Foundation (No.61072106,61271302) and the National 973 Program (No.2013CB329402).
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
[Abstract]:Statistical modeling is essential for the interpretation of synthetic aperture radar (SAR) images. Different SAR statistical models (such as K distribution, Gao Si distribution, gamma distribution, lognormal distribution, mixed distribution, etc.) relate to different types of SAR features (such as farmland, etc.). The modeling capabilities of forests, grasslands, rivers, etc., vary. In this paper, we first introduce the different statistical distribution and its characteristics, and make statistical modeling for different SAR image feature types, and find out the suitable feature types for various statistical models. Target detection is the most important in the application of SAR image. In this paper, a target detection algorithm based on the scattering characteristics of corner reflector is proposed to solve the problem of vehicle detection in airborne SAR images combined with CFAR algorithm. The real airborne P-band and L-band SAR image data are used to verify the algorithm. Ultra-high resolution SAR images have a large amount of data, the traditional CFAR algorithm processing time complexity is high, the target has a certain shape and detail characteristics. In view of these characteristics, we propose a multi-layer CFAR algorithm. The lognormal distribution is used as the statistical distribution model of the image in the algorithm. We use the global CFAR algorithm based on lognormal distribution to filter the strong scattering points for the whole SAR image to find out the background region of the SAR image. Then the background of the extracted SAR image is used to detect the ship target. Although the multi-layer CFAR algorithm can extract more accurate ship targets, there are still many false alarm targets. According to the size of a priori ship, we filter the false alarm target from the image processed by multi-layer CFAR algorithm. Due to the characteristics of ultra-high SAR images, the target after filtering false alarm is incomplete or empty in the hull. We propose an algorithm to extract the contour of the target and fill the contour to get the complete target. In the experiment, the real data of two TerraSAR-X images are used, and the resolution is 1 meter. Compared with the traditional CFAR algorithm and the multi-layer CFAR algorithm, the results show that our algorithm has better detection results. The thesis is supported by the National Natural Science Foundation (No.61072106,61271302) and the National 973 Program (No.2013CB329402).
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN957.52
【相似文献】
相关期刊论文 前10条
1 何友,关键,孟祥伟,陆大，
本文编号：2434965
本文链接：https://www.wllwen.com/kejilunwen/wltx/2434965.html
