Classifying the remote sensing image with complex terrain based on geographical districting method

Since a high degree of heterogeneity within the large region with mountains, plains, rivers and other complex terrains coexistence, automatically classifying the remote sensing image of this region as a whole will result in serious misclassification and low classification accuracy. To solve this problem, this paper presents a new classification method that the regional image is divided into several geographical homogenous sub-images which then is classified independently. The goal of this method is to increase the homogeneity of the districted sub-images to be classified. The Landsat ETM+ image covering the whole region of Jiaozuo city is processed as an example in the experiment. In contrast with the traditional classification method, the result of this method has higher classification accuracy with two evaluated indices of overall accuracy and KAPPA coefficients increased by 12.7% and 0.1508.