Non-negative Matrix Factorization For The Data Of NMR

The NMF algorithm is an Unsupervised pattern recognition. The NMF algorithm is introduced to the NMR data processing. Due to NMF on factor matrix random initialization can cause the answer instability problems. Usally, in the practical application, we must compute several times to get the right result. It is very cumbersome. This paper uses the hierarchical clustering method for clustering and then take the mean matrix initialization, the multiplicative iterative rules to find the optimal solution, according to the coefficient matrix to obtain the final classification results, and to find out the influence of the classification variable.The NMF works with the datas which have high dimensions and small sample size. The method make the number of dimensions of the datas low. So the selection of base number is very important. In the application, people get the base number by trying. In this paper, appropriate modefide for the BIC criterion is to get the base number. It is proved to be reasonable by several datas and obtain right classification results. The paper has four chapters. The first chapter introduces the importance of NMF, the probability model of the algorithm, the application and the short.Chapter two introduces NMF algorithm and some improvments.Chapter three introduces a new initialization way. It uses the hierarchical clustering method to avoid the answer instability problems. In this chapter, AIC and BIC criterion is introduced and the appropriate modefide for the BIC criterion is to get the base number. It is proved to be reasonable by several datas and obtain right classification results.Chapter four mainly uses the new way inferred in the third chapter to analysis the coltsfoot and radic polygalae NMR datas.It is very impossiable. It can also get a good results.