基于稀疏表示的SAR图像去噪和变化检测方法-模式识别与智能系统专业论文.docxVIP

万方数据 万方数据 摘 要 摘 要 SAR 可以穿透云、雾和降水等覆盖物，拥有全天时、全天候的监测能力。在 环境监测、水文观测、地形测绘、资源勘察和军事侦查等方面，这些特点使它具 有独天得厚的优势。但是由于 SAR 图像中固有的乘性散斑噪声，使得后期变化检 测、目标识别等变的困难。所以相干斑抑制是 SAR 图像处理中必不可少的过程。 本文针对 SAR 图像散斑抑制以及 SAR 图像变化检测进行了探索和研究，主要工 作如下： （1）分析了现有的同质区域分割的方法，阐述了 KSVD 字典学习方法，提 出了基于同质区域分割的 SAR 图像去噪方法，对比试验表明，该方法能有效去除 散斑噪声同时又能保持弱目标点和边缘细节信息； （2）分析了传统的 SAR 图像变化检测去噪预处理的问题，阐述了小波域的 隐马尔科夫模型，提出了基于变化检测 SAR 图像去噪方法，对比试验表明，该方 法抑斑效果明显，特别是对后期 SAR 图像变化检测性能有较大提高； （3）分析了基于融合的变化检测方法，提出了一种新的基于稀疏系数融合 的 SAR 图像变化检测方法，对比试验表明，该方法能大幅降低变化检测漏检率， 提高了 SAR 图像变化检测准确率。 关键词：合成孔径雷达 散斑抑制 同质区域分割 稀疏表示 变化检测 基于稀疏表示的 SAR 图像去噪和变化检测 ABSTRACT ABSTRACT SAR has capabilities that can penetrate clouds, fog and precipitation coverings, and all-day, all-weather monitoring, these features make it have a unique advantage in disaster monitoring, ocean observations, resource exploration, surveying and military reconnaissance, etc. However, the inherent SAR image multiplicative noise makes later changes detection and target recognition become difficult. So speckle suppression in SAR image processing is an essential process. In this paper, for SAR image speckle suppression and change detection, we carried out research and exploration. The main work is as follows: Analysis of existing methods of homogeneous region segmentation and elaborating KSVD dictionary learning method. SAR image speckle suppression based on homogeneous region segmentation is proposed .Comparative tests show that the method can effectively remove speckle while maintain the weak point target and edge details; Analysis of the traditional de-noising preprocessing problem in SAR image change detection and elaborating hidden Markov models in wavelet domain. SAR image speckle suppression based on change detection method is proposed. Comparative tests show that speckle suppression effect of the method is obvious. Especially for later SAR image change detection, the performance has improved greatly; Analysis of change detection method based on fusion and giving a description of the im