Hyperspectral Classification of Clustering SVM Based on Modified Spatial Information

Abstract

The difference in local spectral bands of the same species and the local spectral similarity between different species can easily lead to the occurrence of noise points in the region in the traditional classification results. The multi-spectral gray image weighting and the overall gray image weighting filtering algorithm are used to improve the image texture feature, and the modified image is used to perform small window clustering classification based on the high confidence class pixel. The results show that the classification model of the improved algorithm has a certain improvement in classification accuracy: it is 12.3% higher than the traditional SVM classification, and the image noise phenomenon is obviously improved.