Abstract
Sparse unmixing is an important technique for hyperspectral data analysis. Most sparse unmixing algorithms underutilize the spatial and spectral information of the hyperspectral image, which is unfavourable for the accuracy of endmember identification and abundance estimation. We propose a new spectral unmixing method based on the low‐rank representation model and spatial‐weighted collaborative sparsity, aiming to exploit structural information in both the spatial and spectral domains for unmixing. The spatial weights are incorporated into the collaborative sparse regularization term to enhance the spatial continuity of the image. Meanwhile, the global low‐rank constraint is employed to maintain the spatial low‐dimensional structure of the image. The model is solved by the well‐known alternating direction method of multiplier, in which the abundance coefficients and the spatial weights are updated iteratively in the inner and outer loops, respectively. Experimental results obtained from simulation and real data reveal the superior performance of the proposed algorithm on spectral unmixing.
Highlights
Hyperspectral remote sensing is considered to be a major breakthrough in remote sensing technology due to its high spectral resolutions and simultaneous acquisition of both images and spectra of objects [1, 2]
We propose a new sparse reduced-rank regression model for hyperspectral unmixing, which merges spatial and spectral information into the regularizer to enhance the interpretation of hyperspectral data
A local spatial weighting factor is introduced in the collaborative sparse unmixing model, which promotes the spatial smoothness of the image
Summary
Hyperspectral remote sensing is considered to be a major breakthrough in remote sensing technology due to its high spectral resolutions and simultaneous acquisition of both images and spectra of objects [1, 2]. The double reweighted sparse unmixing algorithm (DRSU) employs the double weighted l1 norm to simultaneously reduce the nonzero rows in the abundance matrix corresponding to the actual endmembers and the nonzero values in each abundance vector [13]. The local collaborative sparse unmixing algorithm (LCSU) focuses on the local collaborative characteristics of the image, which imposes different spatial weights on each local region of the abundance map [21]. Low rankness is another inherent characteristic of the abundance matrix, which provides a new perspective for exploring the spatial structure of the data [22, 23].
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
