The research of near infrared spectral peak detection methods in big data era

Abstract. In Spectroscopy, spectral peaks usually are extracted for a better analysis. Currently, the peak finding algorithms mainly contain signal to noise method, peak width and area method, as well as wavelet transform (WT) method. Compared with traditional methods, WT works much better for it merges three steps (noise removal, baseline removal and peak detection) into one and it can avoid the negative effects of flat-peaks. However, the previous research found that the correct ratio decreased seriously when an inappropriate mother wavelet was adopted. In this research, malpractices (pseudo-peak, flat-peak) of the traditional peak width (PW) algorithm were recognized and analyzed. And ascending-descending trend check (ADT) method was adopted to exclude pseudo peaks, a transformed first derivative and a new symbol of peak position were adopted as the solution of flat peaks. The experimental results demonstrated that the new algorithm could recognize peaks and solve the malpractices of traditional algorithms very well.