Pixel Unmixing for Urban Environment Monitoring Using Multi-Temporal Satellite Images

Urban environment monitoring is one of the most important applications of Remote Sensing. In this paper, two images, acquired by Landsat-7 ETM+ on 14 September 2000 and Landsat-5 TM on 12 August 2005 respectively, are used to learn land-cover changes. The study area is within the round-city highway of Xuzhou city, China. Firstly, image registration and haze removal are performed. Then four endmembers, including vegetation, impervious surface, soil and water, are determined in the maximum noise fraction feature space. The spectral mixture analysis is conducted to the pre-processed images of two periods by means of Back-Propagation Neural Network algorithm, and the corresponding fraction images for each endmember are generated. Finally, the image differencing method is applied to the multi-temporal fraction images for monitoring urban land-cover changes according to defined suitable threshold values. The experimental results indicate that spectral mixture analysis algorithm is great potential for urban land-cover change detection.