New methods have been proposed in this paper to improve the classification accuracy in addition to unsupervised and supervised classification procedures. In land cover classification, only the comparability of pixels is considered, while the figure and texture characteristics are not involved. In contrast, thematic information extraction can make good use of these characteristics and extract classification features more precisely. The authors therefore integrated the land cover classification with the thematic information extraction to improve the classification accuracy. During the study, three methods were selected, namely the expert system classification method for land cover, the decision tree extraction method for residential area and the iterative mixed analytical method for water extraction. These methods were tested in Shaoxing, Jiangning and Nanjing areas respectively. The gross classification accuracy of these new methods is 87.89%, which is about 7.86% higher than the common supervised classification method. Especially in fragmented plots as well as low mountain and hilly areas south of the Yangtze River, the problem of objects classification confusion has been solved effectively.