A Wavelet Frequency Domain Enhanced Network for Hyperspectral Image Classification

Aiming at the problems of spectral redundancy and under-utilization of spatial and frequency domain information in hyperspectral image classification, this paper proposes a wavelet frequency domain enhanced network (WFDE-Net) for hyperspectral image classification. A spectral feature extractor (SpeFE) and a channel shift spatial awareness module (CSSAM) are designed to extract key features from the spectral and spatial dimensions; a wavelet frequency domain feature enhancement module (WFDFEM) is designed to enhance texture and semantic information through decomposition and reconstruction; and ultimately, multi-domain feature fusion is realized to further increase the classification accuracy. The experimental results for two public hyperspectral datasets show that the proposed method outperforms several existing methods.