Abstract
By means of joint sparse representation (JSR) and kernel representation, kernel joint sparse representation (KJSR) models can effectively model the intrinsic nonlinear relations of hyperspectral data and better exploit spatial neighborhood structure to improve the classification performance of hyperspectral images. However, due to the presence of noisy or inhomogeneous pixels around the central testing pixel in the spatial domain, the performance of KJSR is greatly affected. Motivated by the idea of self-paced learning (SPL), this paper proposes a self-paced KJSR (SPKJSR) model to adaptively learn weights and sparse coefficient vectors for different neighboring pixels in the kernel-based feature space. SPL strateges can learn a weight to indicate the difficulty of feature pixels within a spatial neighborhood. By assigning small weights for unimportant or complex pixels, the negative effect of inhomogeneous or noisy neighboring pixels can be suppressed. Hence, SPKJSR is usually much more robust. Experimental results on Indian Pines and Salinas hyperspectral data sets demonstrate that SPKJSR is much more effective than traditional JSR and KJSR models.
Highlights
Hyperspectral sensors simultaneously acquire digital images in many narrow and contiguous spectral bands across a wide range of the spectrum
To evaluate the performance of the proposed method for hyperspectral image (HSI) classification, we use the following two benchmark hyperspectral data sets: (1) Indian Pines: This image scene has a size of 145 × 145 pixels and 220 spectral bands, where 200 spectral bands are used in the experiments by removing 20 noisy bands from the original data
Compared with the original kernel joint sparse representation (KJSR), self-paced KJSR (SPKJSR) improves the overall accuracy (OA) and κ coefficient by about 4% in Indian Pines, and by about 2% in Salinas. This demonstrates that the self-paced learning strategy can eliminate the negative effect of inhomogeneous pixels and select effective feature neighboring pixels for the joint sparse representation, which helps to improve the classification performance
Summary
Hyperspectral sensors simultaneously acquire digital images in many narrow and contiguous spectral bands across a wide range of the spectrum. The adaptive neighborhood can be constructed based on traditional image segmentation techniques [12], the superpixel segmentation method [13], and the shape-adaptive region extraction method [14] Both the weighting method and adaptive neighborhood method improve the consistency of neighborhood pixel sets such that the joint representation framework is effective in most cases. Compared with the original JSR, the use of kernel methods yields a significant performance improvement [24] These kernel-based JSR methods assume that neighboring pixels have equal importance and do not considered the differences of neighboring pixels in the feature space. This is obviously unreasonable when pixels in the spatial neighborhood are inhomogeneous.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.