Hyperspectral Image Classification by Fusing Collaborative and Sparse Representations

Abstract

This paper proposes to combine collaborative representation (CR) and sparse representation (SR) for hyperspectral image classification. SR may select too few samples that cannot well reflect within-class variations, while CR generates nonsparse code using all the atoms that may unfortunately include between-class interference. To alleviate these problems, two methods fusing CR and SR are proposed, i.e., a fused representation-based classification (FRC) method and an elastic net representation-based classification (ENRC) method. FRC attempts to achieve the balance between CR and SR in the residual domain, while ENRC uses a convex combination of ℓ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> and ℓ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> penalties. Experimental results on two hyperspectral data demonstrate that the proposed methods outperform the original counterparts, i.e., CR-based classification (CRC) and SR-based classification (SRC).