Identification of maize seed varieties based on near infrared reflectance spectroscopy and chemometrics

• Published in: International journal of agricultural and biological engineering 2018-03, Vol.11 (2), p.177-183
• Main Authors: Cui, Yongjin, Xu, Lanjun, An, Dong, Liu, Zhe, Gu, Jiancheng, Li, Shaoming, Zhang, Xiaodong, Zhu, Dehai
• Format: Article
• Language: English
• Subjects: Algorithms; Authenticity; Biomimetics; Chemometrics; Corn; Crop yield; Discriminant analysis; Environment models; Environmental factors; Identification; Identification methods; Infrared spectroscopy; Markets; Methods; Near infrared radiation; Neural networks; Nondestructive testing; Pattern recognition; Performance indices; Principal components analysis; Proteins; Reflectance; Rice; Seeds; Spectroscopy; Wheat
• ISSN: 1934-6344; 1934-6352
• DOI: 10.25165/j.ijabe.20181102.2815

False seeds can often be seen in the maize seed market, leading to a serious decline in maize yield. Those existing variety identification methods are expensive, time consuming, and destructive to seeds. The aim of this study is to develop a cheap, fast and non-destructive method which can robustly identify large amounts of maize seed varieties based on near-infrared reflectance spectroscopy (NIRS) and chemometrics. Because it is difficult to establish models for every variety in the market, this study mainly investigated the performance of models based on a large number of samples (more than 40 major varieties in the market). The reflectance spectra of maize seeds were collected by two modes (bulk kernels mode and single kernel mode). Both collection modes can be applied to identification, but only the single kernel mode can be applied to purity sorting. The spectra were pretreated with smoothing, the first derivative and vector normalization; and then principal component analysis (PCA), linear discriminant analysis (LDA) and biomimetic pattern recognition (BPR) were applied to establish identification models. The environmental factors such as producing areas and years have a significant influence on the performance of the models. Therefore, the method to improve the robustness of the models was investigated in this study. New indexes (correct acceptance degree (CAD), correct rejection degree (CRD) and correct degree (CD)) were defined to analyze the performance of the models more accurately. Finally, the models obtained a mean correct discrimination rate of over 90%, and exhibited robust properties for samples harvested from different areas and years. The results showed that NIR technology combined with chemometrics methods such as PCA, LDA, and BPR could be a suitable and alternative technique to identify the authenticity of maize seed varieties.