Three-Dimensional Surface Feature for Hyperspectral Imagery Classification

摘要

Gabor surface feature (GSF) uses the first order and second order derivatives of Gabor magnitude pictures (GMPs) to jointly represent image. However, GSF can not excavate the contextual information that hides in the spectral-spatial structure of three-dimensional hyperspectral imagery since GSF can only deal with spatial relationships. Meanwhile, GSF runs on GMPs with multi-scale and multi-orientation, which leads to dimensional explosion problem. Aiming at these two problems, three-dimensional surface feature (3DSF) approach is proposed for hyperspectral imagery in this paper. 3DSF directly deals with the raw hyperspectral imagery data and utilizes its first order derivative magnitude to jointly represent hyperspectral imagery. Experiments on three real hyperspectral datasets, including Pavia University, Houston University and Indian Pines, verify the effectiveness of the proposed 3DSF approach.