A sparse unmixing model based on NMF and its application in Raman image

Abstract

Spectral unmixing is a critical issue in multi-spectral data processing, which has the ability to determine the composition and the structural characteristics of the Raman image. Most of current unmixing methods work well to explore the materials in an ideal scenario. However, both the noise and the requirement of the prior knowledge limit their practical application. Thus, we propose a sparse method called to unmix spectra and apply it to explore the elucidate structural and spatial distribution of the plant cell well. GRSRNMF utilizes the blind source separation technology based on NMF to determine the basis elements (abundances) and their corresponding components (endmembers) without prior knowledge. GRSRNMF incorporates graph relationship and L1/2 regularizer to improve the robustness and effectiveness. Besides, two proper indicators are designed to assess the unmixing method for Raman image when the standard spectrum library does not exist. Experiments are conducted on simulated datasets and the real-world Raman image to evaluate the performance of the proposed methods from various aspects. Experimental results illustrate that the proposed method favors sparsity and offers improved estimation accuracy compared to other methods.