Font Size: a A A

Research On Multi-Scale Change Detection In Multi-Temporal SAR Imagery

Posted on:2015-02-21Degree:MasterType:Thesis
Country:ChinaCandidate:Y CuiFull Text:PDF
GTID:2348330509460729Subject:Photogrammetry and Remote Sensing
Abstract/Summary:PDF Full Text Request
Scale is an important characteristic of remote sensing images, under which SAR images exhibit differential information of ground objects. In order to describe the changing details exhaustively with scale information, and furthermore improve the precision and robustness of change detection for SAR images, accordingly this thesis performs automatic registration and change detection based on scale characteristics. The major work includes:(1)To improve the drawbacks of low efficiency and poor accuracy of Scale Invariant Feature Transform(SIFT) method, a fine and fast registration method combined with partition is proposed for coarse matched SAR images. Firstly, the coarse- matched SAR images are separated into blocks, where SIFT algorithm is applied respectively; afterwards exact and equably-distributed potential feature points are obtained. Next, Random Sampling Consensus(RANSAC) method is employed to realize matching step of these points, from which finally registration result can be reached. Experimental results on two pairs of real SAR data show that the approach can sharply shorten running time in matching step, and improve the registration precision effectively at the same time.(2) To further improve the performance of edge pixels detection in most change detection methods, a novel multi-scale change detection method based on structure similarity is proposed. Gaussian transformation is utilized to constitute feature descriptor of each pixel. Structure similarity is used for choosing optimal scale factor, under which corresponding feature descriptors are clustered through Fuzzy C- means Method(FCM) to achieve the final change map. Experimental results demonstrate that this approach obviously increases the performance in locating edge pixels and the accuracy of change detection together.(3) To enhance the precision of change detection, this paper proposes a multi-scale approach through Markov Random field. This method brings forward an independent wavelet reconstruction way to obtain multi-scale images that keep accurate details at different scales. Next a threshold selection based on mean- iterative way is adopted to segment these images into changed and unchanged parts, after which the final change map is achieved in a measure of image fusion based on Markov Random field. The independent wavelet reconstruction way makes full use of high- frequency component of wavelet transformation that details are depicted precisely. Furthermore, combination with Markov Random field takes both spatial information and scale informa tion into consideration, making the change detection result much subtle.
Keywords/Search Tags:Synthetic Aperture Radar(SAR), Multi-scale Transform, Scale Invariant Feature Transform(S IFT), Random Sampling Consensus(RA NSAC), Structure Similarity, Wavelet Independent Reconstruction, Markov Random Field
PDF Full Text Request
Related items