Wavelet Transform-Based One Dimensional Manifold Embedding For Hyperspectral Image Classification

Traditional wavelet transform-based methods process decompose coefficient in high-dimensional, which makes computational complicated. In order to address this problem, in this paper, a novel approach named wavelet transform-based one dimensional manifold embedding (WT1DME) is proposed for HSI classification. In the proposed approach, firstly, using wavelet transform decomposes the input signal into an approximate coefficients (ACs). Then, smooth ordering is applied to the ACs which maps the coefficients into one-dimensional (1-D) space. Finally, since the coefficients in the 1-D space, hence, 1-D signal processing tools can be applied to build final classifier(we utilize interpolation in this paper). Our proposed methods can be used to process the decompose coefficients in 1-D space, which can perform efficiently. The proposed scheme is experimentally demonstrated by two HSI data sets: IndianPines, University of Pavia has the state-of-the-art performance of results.