Research On Change Detection Technology Of Remote Sensing Images Using Multi-feature Fusion

Change detection of remote sensing images is an important part of remote sensing data processing and applications. With the rapid development of aerospace technology and sensor technology, sources of remote sensing data are getting richer and the resolution is getting higher, thus, traditional change detection methods have been unable to meet the needs of practical applications. To solve this problem, the change detection technology of remote sensing images using multi-feature fusion is systematically studied. On the basis of image segmentation and multi-feature extraction, both supervised and unsupervised change detection method using multi-feature fusion are studied. The major innovations of this dissertation are listed as follows:1. The research status, general principles, methods classification and major processes of the change detection method of remote sensing images using multi-feature fusion are systematically summarized. Principles of supervised change detection method using multi-feature fusion and unsupervised change detection method using multi-feature fusion are introduced, and their characteristics are compared.2. An adaptive edge detection method based on quaternion and histogram is proposed. For the difficulties of adaptive image edge detection and the limitations of conventional edge detection method for multi-spectral remote sensing image, in the quaternion space, the multi-spectral image edge detection is completed using vector rotation, and the adaptive threshold is got through histogram statistics, which results the edge image. Experiments using general multi-spectral remote sensing images and waters multi-spectral remote sensing images are conducted, which verifies the effectiveness of this method.3. The change detection method using multi-feature fusion based on BP neural network and the change detection method using multi-feature fusion based on Support Vector Machine(SVM) are studied. Under the support of a number of samples, the sets of multi-feature difference vector are divided into two categories using BP neural network or SVM, thus the changed and unchanged classes are distinguished. Through theoretical analysis and experimental comparison, these two methods are verified to be better than the traditional difference method and the method using objects spectral feature only, and the change detection method using multi-feature fusion based on SVM is better than that based on BP neural network.4. The change detection method using multi-feature fusion based on Iterated Slow Feature Analysis(ISFA) is proposed. Change of the original multiple features is converted into a new feature space using Slow Feature Analysis(SFA), thus the separability of changed pixels and unchanged pixels is increased. In order to improve the adaptive ability of this algorithm, the idea of iterative weights is introduced and the automatic integration and conversion of multi-feature is achieved. Experiments show that ISFA method can effectively achieve change detection using multi-feature fusion and perform higher accuracy than Principal Component Analysis(PCA) method. Comparison with the supervised methods, ISFA method performs a better ability for change detection of multi-spectral images, which usually have richer spectral information.