Analysis of moldy peanuts by Raman hyperspectral imaging

Abstract

Peanuts are susceptible to mildew which are harmful to human health. In this research, Raman hyperspectral imaging technique was used to analyze moldy peanuts of mixed varieties. One-dimensional spectral data and two-dimensional spatial information were extracted separately: spectral variable selection for the spectral data and gray-level co-occurrence matrix (GLCM) analysis for the spatial information, enabling the identification of mold-related features. Both traditional machine learning and deep learning methods were employed to train the model using the one-dimensional spectral data and two-dimensional spatial data, establishing a prediction model for moldy peanuts detection. Finally, the optimal classification model was achieved using s-GoogLeNet with one-dimensional Raman spectral data, optimized by the Adam optimizer (learning rate = 0.0005). The model demonstrated 90.25 % accuracy on the testing set, with F1 scores of 98.73 %, 96.20 %, 78.87 %, 81.16 %, 91.57 %, for the five peanut varieties (Baisha, Boke, Hongsha, Huahong, and Huayihonghuo), respectively. This research provides a powerful and innovative approach for detecting moldy peanuts of mixed varieties, which will be of great significance for improving grain safety in the industry.