第期杨萌等遥感图像变化区域的无监督压缩感知遥感图像变化区域.docVIP

遥感图像变化区域的无监督压缩感知 杨萌1） 张弓2） 1）,2）（南京航空航天大学信息科学与技术学院，南京 210016） 摘 要 传统的基于结构特征的遥感图像变化检测方法，易受成像稳定性的影响而误差很大。本文针对图像内在的稀疏性结构信息，提出了基于压缩感知（compressive sensing, CS）的遥感图像变化检测方法。通过自适应构造超完备字典将图像局部信息投影到高维空间中，实现图像的稀疏表示，并运用随机矩阵得到了数据在高维空间中的低维特征子空间。最后利用模糊C均值（fuzzy c means, FCM）聚类算法进行无监督聚类，实现遥感图像变化区域信息的重构。实验结果表明，本文方法不仅能够很好的检测出图像的轮廓变化和图像的区域变化，而且对噪声具有很好的鲁棒性。 关键词 变化检测 遥感图像 压缩感知(CS) 模糊C均值(FCM)聚类 中图法分类号：TP751.1 文献标识码：A 文章编号：1006-8961 Unsupervised Compressive Sensing of Change Area in Remote Sensing Images YANG Meng1), ZHANG Gong2) (College of Information Science Technology, Nanjing University of Aeronautics and Astronautics, Nanjing, 210016) Abstract Traditional remote sensing image change detection approaches based on structure features are usually limited by imaging stability. In this paper, we introduce a new method for change detection in remote sensing images using compressive sensing (CS) based on the image inherent sparse structures. For this algorithm, a large collection of image patches is projected onto high dimensional spaces through redundant dictionary, giving an adaptive sparse representation per each image patch. Random matrix which meets RIP with high probability is taken as measurement matrix to realize feature space dimension reduction. And then, the final change detection is realized by clustering the feature vector space using the fuzzy c-mean clustering algorithm, achieving the reconstruction of change regional information. The experimental results demonstrate that the proposed algorithm has a good change detection results both in contour and region and also has a good robustness on the noise. Keywords change detection, remote sensing image, compressive sensing (CS), fuzzy c means (FCM) clustering 0 引 言 遥感图像变化检测是指利用多时相获取的同一地表区域的遥感图像来确定和分析地表变化，提供地物的空间展布及其变化的定性与定量信息，它在地物地层分析、国土资源普查和军事、灾害监测等方面拥有广泛的应用价值和商业价值。 随着遥感图像变化检测技术的不断发展，研究者提出了多种变化检测算法。图像差值方法[1]是将两幅图像的灰度值相减，通过差值图像与所选取的阈值进行比较，来区分变化与不变化的区域。对数比值方法[2]是在得到两幅图像对应像素的比值后，再对其取对数。相比于差值方法，对数比值方法对图像的乘性噪声较