Spectral-spatial hyperspectral classification with spatial filtering and minimum spanning forest

The combination of spectral and spatial information in the classification of hyperspectral images is known to be a suitable way in improving classification accuracy. In this paper, a novel spectral-spatial classification scheme is presented based on weighted mean filtering (WMF) and construction of minimum spanning forest (MSF). At first, WMF is conducted on a given hyperspectral image. Then, the first eight principal components are regarded as reference images and support vector machine (SVM) classification is performed. M marker pixels are selected randomly from the obtained classification map and the MSF is constructed. Finally, the segmentation map is post-processed by using a majority voting technique within connected components. The experimental results are illustrated on a hyperspectral image indicating that the proposed scheme increases the classification accuracy, compared to previously classification techniques. Therefore, it is attractive for hyper-spectral images classification.