Research On Change Detection For Remote Sensing Images Based On Discriminant Analysis

Change detection in remote sensing images is defined as the procedure of quantitatively analyzing and identifying changes occurred on the earth's surface from remote sensing images acquired at different time, which is one data analyzing technique aimed to the characteristic of remote sensing images and can recognize the difference of objects and phenomenon. The changed information from change detection can be used in updating and use of resource, detection,evaluation and forecasting of disaster, the analysising of battlefield's situation and evaluation of strike's effect. The research on change detection and its some theory don't only make change detection playing more important role in practice, but also promote the development of theory on change detection.This dissertation mainly studies that how do automatic extract changed information from multitemporal remote sensing images, which discuss deeply some theory and methods on remote sensing images change detection. The major innovations of this dissertation are listed as follows: 1. Propose one approach of remote sensing images based on similarity discriminant, include similarity of sets and similarity of pixels. Aimed to the difficulty of selecting decision threshold from difference image, this dissertation proposes that difference image can be divided into unchanged region and changed region in according to similarity of sets in the foundation of summarizing another methods on selecting decision threshold, which avoids computing the threshold.Then,fliter some pixels of false pixels in according to similarity of pixels, which solve the problem that change detection methods are sensitive to images radiation distortion.2. Fisher rule function is imported to remote sensing images change detection and expand from one dimension to two dimension, and propose the approach based on 2D-Fisher rule function. The approach solve effectively the problem on noisesensitivity of exsiting change detection approaches.3. Because the approach of 2D-Fisher rule function need a mass of time, it can't satisfy the require of practice. This dissertation improve the numeration's way of the approach of 2D-Fisher rule function, transform computing two-dimension threshold to two one-dimension thresholds, and improve greatly the speed of computing decision threshold. Improved approach can not be only used change detection, but also be applied another aspect.4. Study the expression method on the result of remote sensing images change detection. Isolated point and isolated region in binary image can be wiped off in terms of mathematics morphologic; Grid is transformed to vector throughout edge tracking in order to continue to utilize changed information.The proposed methods and algorithms are applied to both real and synthetic multitemporal remote sensing images and the experiment results are satisfied,which is superior to some classic methods.