Comparative Study of Three Feature Selection Methods for Regional Land Cover Classification Using MODIS Data

Selecting suitable features is very crucial for achieving successful classification of land cover types. This paper presents a comparative study of three typical feature selection methods for the task of regional land cover classification using MODIS data. Comparison results have shown that Branch and Bound is the best for land cover classification with MODIS data, while ReliefF and mRMR achieve nearly the same accuracies on the target application. The experimental results also show that it is necessary to conduct feature selection, which can reduce the computation cost largely, while the accuracy remains the same or even better.