The Research And Application Of Change Detection In Remote Sensing Images Based On Semi-Nonnegative Matrix Factorization

With the development of remote sensing (RS) technology as well as the availability of multitemporal observed data, change detection has become an important research hotspot in remote sensing image processing and applications. It tends to process and analyze two or more coregistered remote sensing images acquired at different time for identifying the changes on the ground. Thus, such tasks like environment monitoring, military reconnaissance and disaster monitoring have been performed through the change detection.With the rapid development of urbanization, the data of urban planning obtained by traditional field investigation can’t meet the actual requirements, because it is time-consuming and can’t dynamically extract change information. According to the characteristic of remote sensing, utilizing remote sensing images to acquire the changes on the ground will be a new means for urban planning. At the same time, the difficulty in the visualization of the field data and the change detection results is circumvented with geographic information system (GIS) technology.Research works focus on change detection approach and its application in urban planning. The main contribution is stated as the following two parts:1. As a new paradigm of factorization, the non-negative matrix factorization (NMF) will be exploited. Thanks to the capability of targeting some of the specific features of the data analysis to some extent, the NMF has been extensively used in analyzing non-negative data. Moreover, the semi-NMF (SNMF) was proposed in literatures and possesses the ability to process both the non-negative and negative matrices through the relaxation of the constrains on the data. Therefore, a SNMF-based change detection method of remote sensing images is proposed. Specifically, the difference image is generated by the treatment of two coregistered remote sensing images. Then, the principal component analysis (PCA) is adopted for feature extraction to construct the feature-by-item data matrix. Finally, the base matrix and coefficient matrix are obtained by iteratively performing the SNMF and the change map is determined by assigning each element of the coefficient matrix to a proper class. The effectiveness is verified through the experiments with both optical remote sensing images and the synthetic aperture radar (SAR) images.2. Based on the integration of the ArcGIS Engine and IDL, a visualization system for change detection of the multi-temporal remote sensing images is developed. This system chooses to use static map as background map, utilizes the coordinate information of remote sensing image to generate the coordinate file, and transforms the change map to the layer, then displays the results after classifying and rendering with the degree of change. Moreover, the system provides to users with online map services to browse more geography. Applying the change detection method to the multitemporal remote sensing images, combined with GIS technology, is competent to provide more evidence for urban planning. It is useful for promoting the sustainable development of urbanization, and is also helpful for environment monitoring, national defense, etc.