Quantitative analysis of starch species based on near-infrared spectroscopy and quaternion convolution neural network

Abstract

In this paper, the near-infrared spectral data of five different types of starch were collected, and the starch species identification model was constructed by using a quaternion convolutional neural network (QCNN), we proved that the qualitative model based on QCNN has obtained higher prediction accuracy than traditional qualitative models. In the experimental results, the classification accuracy of QCNN for five different starches reached 0.996. The results show that the combination of the quaternion spectral fusion method and deep learning is more conducive to extracting and mining the deep information of NIR spectra and has important research significance and application value in the field of nearinfrared spectroscopy technology.