Benchmark for Hyperspectral Unmixing Algorithm Evaluation

Over the past decades, many methods have been proposed to solve linear or nonlinear mixing of spectra inside hyperspectral data. Due a relatively low spatial resolution imaging, each image pixel may contain from multiple materials. In turn, unmixing is finding these materials and their abundances. A few main approaches performing emerged, such as nonnegative matrix factorization (NMF), mixture modelling (LMM), and, most recently, autoencoder networks. These use different in endmember abundance information images. However, due huge variation data being used, it difficult determine which perform sufficiently on datasets if they can generalize any input problems. By trying mitigate this problem, we propose algorithm testing methodology create standard benchmark test already available newly created algorithms. experiments were created, variety used compare openly algorithms best-performing ones.