This paper deals with the land cover classification of the Hongsipu Irrigation Area in Ningxia in 1989 based on remote

sensing techniques, serving as a benchmark for the study of the development zones. Reference end-members were used as the

characteristics of classification with which the classification function was stable and the results of classification were easily

to explain. The results show that the overall accuracy of maximum likelihood classification based on abundance maps derived from

the decomposition of mixed pixels is 77.53%. Compared with the maximum likelihood classification based on original image, the

overall accuracy is raised by 9.8%. Therefore, the combination of the linear spectral mixture decomposition model and the maximum

likelihood classification constitutes a good classification method. In order to improve the classification accuracy, this paper

makes a post-processing on the classification results to meet the actual demand.