Abstract:
The existence of mixed pixels in hyperspectral data makes their classification very challenging. In this paper, we propose a new spectral-spatial Kernelized Sparse Bands ...Show MoreMetadata
Abstract:
The existence of mixed pixels in hyperspectral data makes their classification very challenging. In this paper, we propose a new spectral-spatial Kernelized Sparse Bands (SS-KerSparseBands) selector to extract discriminative features for accurate hyperspectral data classification (HDC). In our method, both the intrinsic cube structure of data and the sparse characteristics of features are explored, by formulating spectra as a hyperspectral tensor and reducing feature selection to the spectral-spatial joint sparse coding (SC) of labels. Moreover, a new tensor-multiple measurement vector (TMMV) optimization algorithm is advanced to identify the most significant KerSparseBands for the subsequent Fisher discriminant classification. Some experiments are performed on several synthetic dataset and real Indian Pines, Salinas-A, and Pavia datasets to investigate the performance of SS-KerSparseBands, and the results show that it can accurately classify mixed pixels, and is competitive in terms of classification accuracy and computational complexity when compared to its counterparts.
Published in: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing ( Volume: 9, Issue: 9, September 2016)