Using convolution neural network and hyperspectral image to identify moldy peanut kernels

Peanut is an important oil raw material. During transportation and storage, aflatoxins will be produced if peanuts become moldy. Identifying and screening peanuts can effectively improve food safety. In this paper, a method of collecting hyperspectral peanut data by spectrometer and identifying peanut by deep learning technology is designed. Firstly, 16 hyperspectral images of healthy, damaged and moldy peanuts were collected by spectrometer with 1066 peanut samples. Then, Deeplab v3+, Segnet, Unet and Hypernet constructed in this paper were used as control model for comparison. The proposed peanut recognition index (PRI) was fused into the hyperspectral image as data feature pre-extraction. The constructed multi-feature fusion block (MF block) was integrated into control model as the model feature enhancement. Experiments show that the average accuracy of the four control models was improved by 0.61-1.15% after feature pre-extraction. Through feature enhancement, it increased again by 0.43-4.96%. This proves that the two proposed methods play a positive role in improving the accuracy of peanut recognition. In addition, this study has reference significance for other hyperspectral object recognition.