No Supervision Of Multi-channel Remote Sensing Image Change Detection Method

Change detection in multichannel remotely sensed images aims at identifying change information by analyzing a pair of images acquired on the same geographical area in different times.This technique is used in wide range of applications,like studies on land use/land cover dynamic,environmental monitoring, disaster prevention and control,detecting of military target, etc. Unsupervised change detection techniques make a direct comparison of the two multichannel remotely sensed images considered, without additional information. So the unsupervised change detection techniques are used widely. And how to improve the accuracy of unsupervised change detection techniques becomes the hotspot in the remotely sensed images application.Focusing on some problems of unsupervised change detection technique, this article has made a lot of research, and proposes methods to resolve the problems. The main ideas and innovations are as follows:1.A split window-based method for unsupervised change detection in multichannel remotely sensed images is proposed. This method splits difference image into a set of subimages, and determine segmentation threshold of the whole scene by combining the thresholds of subimages.Experimental results demonstrate that the proposed method can detect change information accurately even if percentage of changed area in whole scene is relatively too small or too large, and improve detection accuracy obviously compared to general change detection methods.2.A split window-based method with Semisupervised SVM(SSSVM) algorithm is proposed for unsupervised change detection in multichannel remotely sensed images. It firstly splits difference image into a set of subimages, and then determine the classification hyperplanes of the difference image with the optimal hyperplane of the subimage.Experimental results demonstrate that the proposed method can detect change information accurately even if percentage of changed area in whole scene image is relatively too small or too large, and improve detection accuracy obviously compared to the traditional semisupervised SVM method.3.The Fuzzy C-Mean(FCM) method is introduced into change detection problem of multichannel remotely sensed images.However, the FCM method is sensitive to isolated point, and its clustering results are influenced by noise.To solve this problem, a new method combining FCM with neighborhood analysis for unsupervised change detection in multichannel remote sensing images is proposed. Experimental results demonstrate that the proposed method can remove the influence of noise from the FCM algorithm and detect change information accurately.