Research Publications

Hyperspectral remote sensing imagery is a three-dimensional data cube with the width and length corresponding to spatial dimensions and the spectral bands as the third dimension, which has rich information about spectral and spatial distributions of distinct surface materials. The main advantage of using hyperspectral imagery is that the materials in the scene can be identified more accurately and reliably than using panchromatic or multispectral imagery. Feature selection and pixel classification are two main associated tasks of hyperspectral imagery processing and analysis. In this paper, we address the following problems. 1) Structured sparse logistic regression models are applied for hyperspectral feature selection and pixel classification, which benefit both from efficient algorithms and well-developed theory for generalization properties and variable selection consistency. 2) Structured features of hyperspectral imagery are explored by spectral band clustering and three-dimensional discrete wavelet transform (3D-DWT). 3) Various structured sparse constraints such as lasso, group lasso, sparse group lasso are used with structured features. Compared with other feature extraction and classification methods for hyperspectral imagery, the proposed approach has some important advantages