Abstract
Due to the sparsity of hyperspectral images, the dictionary learning framework has been applied in hyperspectral endmember extraction. However, current endmember extraction methods based on dictionary learning are not robust enough in noisy environments. To solve this problem, this paper proposes a novel endmember extraction approach based on online robust dictionary learning, termed EEORDL. Because of the large scale of the hyperspectral image (HSI) data, an online scheme is introduced to reduce the computational time of dictionary learning. In the proposed algorithm, a new form of the objective function is introduced into the dictionary learning process to improve the robustness for noisy HSI data. The experimental results, conducted with both synthetic and real-world hyperspectral datasets, illustrate that the proposed EEORDL outperforms the state-of-the-art approaches under different signal-to-noise ratio (SNR) conditions, especially for high-level noise.
Highlights
Due to the complexity of natural ground spectra and the low spatial resolution of the remote-sensing hyperspectral imaging process, pixels of these images are by nature, mixtures of several spectral signatures known as endmembers [1]
In many application scenarios such as target detection, a certain level of subpixel accuracy is required in order to improve the image processing output, which means that unmixing the hyperspectral image (HSI) is crucial
Endmember extraction is the key step in the decomposition of mixed pixels, that is, the determination of the basic features of the mixed pixels in the observed image, the result of which greatly affects the accuracy of the estimation of the abundance coefficients
Summary
Due to the complexity of natural ground spectra and the low spatial resolution of the remote-sensing hyperspectral imaging process, pixels of these images are by nature, mixtures of several spectral signatures known as endmembers [1]. Representative algorithms include the pixel purity index (PPI) [10], vertex component Analysis (VCA) [11], N-FINDR [12], the automatic target generation process (ATGP) [13], the simplex growing algorithm (SGA) [14] and so on These kinds of algorithms generally have low computational complexity and high efficiency, and can obtain good endmember extraction results when the observed image satisfies the pure pixel assumption. In order to adapt to the noisy environment, in this paper, a hyperspectral endmember extraction approach based on online robust dictionary learning, termed EEORDL, is proposed, to robustly extract the endmembers under the condition of a high noise level In this method, we perform online updating for the objective function with l1 data fitting the term to enhance the robustness of the endmember extraction process in relation to noise.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
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