On ALIF-WT-GDA-based Terahertz spectral method for identification of transgenic cotton seeds

Abstract

Terahertz (THz) time-domain spectroscopy can interact with organic molecules to produce “resonance” absorption, and it is suitable for qualitative and quantitative analysis on the basis of spectral absorption mechanism. However, there are some problems with it, such as un-obviousness signal peaks and positions, signal oscillations and low recognition accuracy. Therefore, this paper has proposed a method named ALIF-WT-GDA-based THz spectrum for identifying transgenic cotton seeds, which can strengthen and accurately identify features of the THz spectral data. First, this paper has put forward a pretreatment method of THz spectral data based on Adaptive Local Iterative Filtering (ALIF) and Wavelet Transform (WT). Specifically speaking, ALIF is used to decompose THz time-domain spectral signals and extract useful mode components for signal reconstruction in order to improve the signal-to-noise ratio (SNR). Also, WT is employed to perform smoothing filtering of THz signals, which makes THz absorbance curves much smoother and absorption peak positions much more obvious. Then, thanks to the advantages of kernel methods, Generalized Discriminant Analysis (GDA) method of Kernel Fisher Discriminant Analysis (KFDA) is used for dimension reduction and clustering analysis. Finally, the proposed ALIF-WT-GDA method is adopted for pretreatment and clustering of the THz spectral data from four different types of transgenic cotton seeds, with its identification accuracy of 98.33%. The above result has proved that the proposed ALIF-WT-GDA method is effective.