Research on high-throughput crop authenticity identification method based on near-infrared spectroscopy and InResSpectra model

Abstract

The authenticity identification of crop varieties is of great significance for crop variety protection and variety breeding. In this paper, a new convolutional neural network (CNN) based on Inception and residual module (InResSpectra) was proposed to realize the authenticity identification of wheat and rice seeds by near infrared spectroscopy (NIRS). Based on a self-developed near-infrared crop high-throughput acquisition system, 24 and 21 varieties of wheat and rice seeds were identified, respectively. The accuracy (ACC), precision (PRE), recall (REC), and F1 measure (F1) of the near-infrared model constructed based on InResSpectra for the identification of wheat sample set were 95.35%, 95.42%, 95.42%, and 95.42%, which were higher than those based on CNN, Inception model and partial least squares discriminant analysis(PLS-DA). The ACC, PRE, REC, and F1 of the near-infrared model constructed based on InResSpectra for the identification of rice sample set were 93.07%, 93.10%, 93.10%, and 93.10%, which were better than those based on the other three algorithms. The InResSpectra model also obtained lower standard deviation and better complexity analysis results compared with the CNN and Inception models. The results showed that NIRS combined with InResSpectra model can accurately and robustly identify the authenticity of the seeds of multiple wheat and rice varieties. The proposed model has the potential to be used in the identification of more crop seed varieties to help improve the efficiency of crop breeding.