An improved method based on a new wavelet transform for overlapped peak detection on spectrum obtained by portable Raman system

摘要

Peak detection is a particularly important pre-processing step in chemical identification using Raman spectra. At present, most peak detection methods have limited applicability when there are overlapping peaks, especially the spectrum measured by portable spectrometers with low resolution. In this paper, an improved method is proposed based on the application of continuous wavelet transform (CWT) peak detection using a new wavelet to the deconvolved Raman spectrum. The new wavelet has a smaller linewidth and is more similar to the intrinsic Lorentz line profile of the Raman spectrum. It, therefore, has several advantages with regard to the detection of overlapping peaks. The proposed method was evaluated using the Raman spectrum of solid amino acid mixtures, and the results show that it is better at detecting overlapping peaks than the two other investigated wavelets. The receiver operating characteristic curves show that this method can detect more true peaks while maintaining a low false discovery rate. Moreover, the maximum of the true positive rate is the largest in the new approach, which indicates better performance for overlapping peak detection.