Extraction of component bases from mixed spectra using non-negative matrix factorization with dissimilarity regularization

摘要: Abstract We propose a method of automatic pattern decomposition for mixed spectra, based on non-negative matrix factorization. This uses regularization term that increases the volume coordinated by decomposed bases. treatment enhances dissimilarities among bases and is suitable expressing natural component which generally differ from each other. shows better accuracy in test using virtual spectrum datasets compared to conventional regularizations. also proposed with high-throughput X-ray diffraction dataset measured synchrotron radiation facility. Our can decompose into bases, are correctly assigned background chemical compounds, while results obtained other terms cannot sufficiently separate peaks meaningful peaks. found combination dissimilarity smoothness detect small These show advantage purpose decompositions.