A Gas-Spectral Bimodal Information Fusion Method Combining Electronic Nose and Hyperspectral System to Identify the Rice Quality in Different Storage Periods

Abstract

Rice quality tends to decline with the increase in storage period. In rice production, it is common to pass off poor-quality rice with a long storage period as fresh rice. In this work, we designed a self-selection convolution neural network (SS-Net) combined with nondestructive detection techniques of electronic nose (e-nose) and hyperspectral to identify the rice quality in different storage periods. First, apply the e-nose and hyperspectral system to detect the gas and spectral information of two rice brands, Dao Huaxiang and Xiao Yuanli, in six storage periods, with three humidity levels. Second, a self-selection convolution (SSConv) is proposed to concern essential features affecting the classification performance after fusing the gas and spectral information. Finally, SS-Net is designed to achieve the adaptive classification of gas and spectral information, realizing rice quality discrimination. Compared with other classification methods, SS-Net obtains the best classification performance and provides an effective method for rice quality monitoring.