There are many spectral bands or band functions developed for land-cover feature measurements. When the ratio of the number of training samples to the number of feature measurements is small, the traditional land-cover classification is not accurate. To solve this kind of problem, decision-theoretic rough set model (DTRSM) can be introduced. This model is linked with distinguishing different types of samples in the image. The samples in the minority classes will be misclassified based on the model. To minimize the misclassification, the improved feature selection algorithm with comprehensive criteria is proposed on processing remotely sensed data. By comparing with other feature selection algorithms, we find that the proposed method can effectively select key features for different data sets and the accuracy of classifiers can be ensured.
