Beyond NMF: Advanced signal processing and machine learning methodologies for hyperspectral analysis in EELS

摘要: Component separation techniques such as non-negative matrix factorization (NMF) have found a permanent home in the electron energy-loss spectroscopy (EELS) community due to their ability to separate overlapped signals in a hyperspectral dataset. Especially in the field of nanoplasmonic, where NMF components can directly represent physical plasmon modes in simple systems [1]. While NMF has only been extensively used in EELS over the last decade, it has been an established technique in the signal processing community for much longer [2]. We believe there is significant potential application to porting other established signal processing techniques to EELS hyperspectral analysis. Here, we will discuss two such techniques: a data fusion methodology called pan-sharpening and unsupervised component separation methodology using a type of machine-learning network called an autoencoder.Pan …