Non-Destructive Detection of Mold in Maize Using Near-Infrared Spectral Fingerprinting

Abstract

The spectral raw data are initially acquired using a handheld near-infrared spectrometer. To enhance the signal quality, preprocessing is conducted, and a classification model is developed for full-band spectral data. In order to further optimize the model and enhance the classification accuracy, the feature wavelengths were extracted from the spectral data with effective preprocessing techniques in the full-band model. Finally, the maize kernel mold classification model is constructed. The experimental results show that the classification accuracy of SG+BC-RF full band model can reach up to 94.93%, and the accuracy for the identification of asymptomatic moldy maize is 91.43%. The classification accuracy of SG+SNV-SVM-ISFLA feature selection band model can reach up to 97.81%, and the accuracy for the identification of asymptomatic moldy maize is 94.29%, which can realize the accurate grading of moldy accurate classification of maize and can well distinguish asymptomatic moldy maize.

Keywords: Maize mold, Near-infrared spectroscopic, Non-destructive testing, Early identification, Spectral analysis