Unmixing of Hyperspectral Data Using Robust Statistics-based NMF

Roozbeh Rajabi,Hassan Ghassemian
DOI: https://doi.org/10.48550/arXiv.1212.0888
2012-12-05
Abstract:Mixed pixels are presented in hyperspectral images due to low spatial resolution of hyperspectral sensors. Spectral unmixing decomposes mixed pixels spectra into endmembers spectra and abundance fractions. In this paper using of robust statistics-based nonnegative matrix factorization (RNMF) for spectral unmixing of hyperspectral data is investigated. RNMF uses a robust cost function and iterative updating procedure, so is not sensitive to outliers. This method has been applied to simulated data using USGS spectral library, AVIRIS and ROSIS datasets. Unmixing results are compared to traditional NMF method based on SAD and AAD measures. Results demonstrate that this method can be used efficiently for hyperspectral unmixing purposes.
Computer Vision and Pattern Recognition
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the mixed pixels problem in hyperspectral images due to the low spatial resolution of sensors. Specifically, the paper explores the use of the Robust Statistics - based Nonnegative Matrix Factorization (RNMF) method for spectral unmixing of hyperspectral data. The goal of spectral unmixing is to decompose the spectra of mixed pixels into endmember spectra and abundance fractions, so as to be able to identify and quantify different substances in the scene more accurately. In the paper, the authors propose the RNMF method. This method uses a robust cost function and an iterative update process, making it insensitive to outliers and thus improving the accuracy of unmixing. The paper also reports the experimental results on simulated data and compares them with the traditional Nonnegative Matrix Factorization (NMF) method. The results show that RNMF performs better in two metrics, namely Spectral Angle Distance (SAD) and Abundance Angle Distance (AAD), demonstrating the effectiveness of this method.