Hyperspectral Image Classification Based on Image Segmentation

This paper introduces a novel semi-supervised segment-based classification method for hyperspectral images with a few samples, which consists of following steps. First, the principal component analysis method is adopted to obtain the first principal component for original hyperspectral image. Second, the obtained first principal component was segmented with the region growing segmentation method. Then, some unlabeled samples were labeled according to the labeled samples within the same region. Next, support vector machine (SVM) is used to classify the hyperspectral image. Finally, edge-preserving filtering is performed on the classification result to remove the noise and preserve edges. Experimental results demonstrate that the proposed method can improve the classification accuracy significantly when the number of labeled samples is relatively small.