Research On Feature Extraction And Classification Of Hyperspectral Remote Sensing Image

This paper focuses on the feature extraction and classification of hyperspectral images.The main work is as follows: A graph convolutional subspace clustering algorithm based on PCA and a similarity graph model is proposed.Firstly,the PCA algorithm is used to reduce the dimension of the high-dimensional dataset.Then use image convolution subspace clustering algorithm for feature extraction and cluster analysis.The research shows that,compared with the conventional clustering algorithms,the graph convolutional subspace clustering algorithm has better higher clustering accuracy on high-dimensional small sample data.A hyperspectral image classification method based on multiscale feature fusion is proposed.Firstly,using octave convolution to extract the low-level features of the hyperspectral images,secondly,the attention mechanism is introduced to focus on learning the focus area in both spatial and spectral dimensions,finally,the fused spatial-spectral features are extracted for the classification task.The results show that the classification accuracy of this method on the WHU-Hi datasets reaches 99.63%,97.90%,and 98.69%,which are the highest overall accuracy of these datasets.