Application of terahertz spectroscopy combined with feature improvement algorithm for the identification of adulterated rice seeds

Abstract

In order to combat the adulteration of rice seeds for earning illegal huge profits, this paper presents a terahertz time-domain spectroscopy-based pattern recognition method for adulterated rice seeds. Based on the collected terahertz time-domain spectral data, the Relief algorithm, random forest (RF) algorithm and maximum correlation minimum redundancy (mRMR) algorithm are developed to select the characteristic frequencies, followed by two types of signal processing methods, Hilbert transform, Butterworth Low-Pass Filter, to process the spectral data and fuse them with the original spectral data. Finally, two machine learning models, support vector machine (SVM) model and extreme learning machine (ELM) model, are used to classify the spectral data samples after the feature processing. The results show that the sample spectral data processed by mRMR feature selection algorithm and Hilbert transform have the best recognition effect on the ELM model with an accuracy of 100%. This study has some reference significance for detecting adulterated rice seeds.