The Research Of Change Detection Method Using High-Resol-ution Remote Sensing Images And Vector Data

With the rapid development of Chinese economy, the negative problems of the imbalance between urban and rural areas, the unreasonable industrial structure and the contradiction between resources and environment are becoming more extrusive. Carrying out the comprehensive monitoring of the national geographical condition can provide the basis for the formulation of government and policy decisions, and provide the basic data for the relevant departments. The main technical methods of national geographic conditions census include comprehensive monitoring of remote sensing technology, multi-source data fusion and change detection processing techniques and technologies such as geographic features detection.Standing on the perspective of the national geographical conditions monitoring, we studied the geographic features change detection technology by integrating of multi-source data. The main contents of this article are shown as follows:(1) Integration of multi-source data can help us to get more information for change detection, which provides a prior knowledge for image analysis. A new method of image segments extraction based on the integration of remote sensing images and vector data is put forward in this article, that is, the multi-scale segmentation is based on the boundaries and categories of the vector data, and the image segments with better spectral homogeneity will be chosen as basic unit for change detection. The image segments obtained by this method match the goals of "same spectrum within classes" and " different spectrum among classes".(2) We present the ranges of image segment feature extraction in two cases:the first one is that the pixels inside the image segment will be chosen to compute the characteristic value when the homogeneity of segments gotten in segmentation is poor; the second one is that the whole image segment (including the marginal pixels) will be chosen to compute the characteristic value when the homogeneity of segments gotten in segmentation is good. Then, we choose spectral features, shape, texture and other features of the segment, building the feature spaces. The feature spaces need to be optimized, due to the uncertainty of feature choosing for the image. In this article, we analyzed and studied the commonly used optimization methods and the optimization principle of the features.(3) The existing detection methods of change detection based on remote sensing images are summarized and classified. It will divide into two methods (these are the post-processing and pre-comparison), according to the technique processes of change detection. The main technical routes of the post-processing method are image classification and change detection, whose principle is to use the K-nearest neighbors approach for classification. The pre-comparison of change detection is represented by the image segments differential entropy method, which can obtain the result of change detection by getting the corresponding image segments entropy of two images in different periods and choosing the appropriate thresholds.