Key Information Retrieval in Hyperspectral Imagery through Spatial-Spectral Data Fusion

Abstract

Hyperspectral (HS) imaging is measuring the radiance of materials within each pixel area at a large number of contiguous spectral wavelength bands. The key spatial information such as small targets and border lines are hard to be precisely detected from HS data due to the technological constraints. Therefore, the need for image processing techniques is an important field of research in HS remote sensing. A novel semisupervised spatial-spectral data fusion method for resolution enhancement of HS images through maximizing the spatial correlation of the endmembers (signature of pure or purest materials in the scene) using a superresolution mapping (SRM) technique is proposed in this paper. The method adopts a linear mixture model and a fully constrained least squares spectral unmixing algorithm to obtain the endmember abundances (fractional images) of HS images. Then, the extracted endmember distribution maps are fused with the spatial information using a spatial-spectral correlation maximizing model and a learning-based SRM technique to exploit the subpixel level data. The obtained results validate the reliability of the technique for key information retrieval. The proposed method is very efficient and is low in terms of computational cost which makes it favorable for real-time applications.