Band selection based on genetic algorithms for classification of hyperspectral data

As an effective approach for reducing hyperspectral data dimension, band selection has been widely used in the fields of classification and identification of specific objects. Combining genetic algorithm and the k-means clustering method, we propose a new band selection method for hyperspectral data in this paper. It is composed of three main steps. First, we apply the K-means classification method to cluster bands of hyperspectral data into several band sets. Second, the cost function criterion is constructed from the classification accuracy. Finally, the band selection method is developed based on genetic algorithms and k-means clustering supervised classifications. In result, the specific target of obtaining the least bands without loss of classification accuracy can be obtained. The proposed method is verified on real hyperspectral data for classification of plant leaves and compared with classical methods. It is proved that the presented method can generate results with higher accuracy and fewer bands at the same time.