A Hyperspectral Image Feature Extraction Algorithm Combining Spatial-Spectral Information

Abstract

Aiming at the problem of “curse of dimensionality” in hyperspectral image (HSI) classification, this paper proposes an HSI feature extraction algorithm combining spatial-spectral information. The proposed algorithm makes full use of the spatial information and spectral information of HSI. First, the image is reconstructed by introducing the principle of spatial consistency, reduces the redundancy of local information, and incorporates spatial structure information into the spectral feature set automatically. Then, the traditional PCA-LPP manifold learning algorithm is improved to extract spatial-spectral information based on the reconstructed image. The Parzen window density estimation method is used to adapt the neighborhood size, the local spatial features of the image are extracted. At the same time, the neighbor graph is constructed based on the spatial-spectral distance metric, so that the similarity measure between data points is better realized. Finally, the extracted spatial-spectral features are input into a Support Vector Machine (SVM) for HSI classification. Experimental results on Salinas and Pavia U datasets show that the algorithm has better classification accuracy than traditional algorithms.