Identification of Wheat Cultivars Based on the Hyperspectral Image of Single Seed

Traditional methods for identification of wheat cultivars involve human inspection, Polyacrylamide gel electrophoresis (PAGE), and reversed phase high performance liquid chromatography (RP-HPLC), and these methods all had some limitations. Therefore, new technologies for identify wheat cultivars for single seed is needed, especially for wheat breeders that usually have small amount of seeds. In this study, hyperspectral imaging was applied to identify wheat cultivars of single seed. hyperspectral image were collected at two seed positions of ventral groove up and ventral groove down, and the mean spectra of this two positions was also calculated. Using image mosaic methods, the average spectra of the single seed region was extracted for analysis. Principle component analysis (PCA) was used to see the characteristic of wheat cultivars in latent space. Soft independent modeling of class analogy (SIMCA) was used to construct the classification model. Results showed that hyperspectral image could characterize the detail feature of single seed, and the data collection position had influence on the hyperspectral image. From the results of PCA and SIMCA model, it indicated that hyperspectral image could differentiate three types of wheat including strong gluten wheat, medium gluten wheat, and weak gluten wheat, the classification accuracy achieved 96%. The classification accuracy of six wheat cultivars including NONGDA195 and ZHONGYOU206 (strong gluten wheat), JINGDONG12 and JINGDONG17 (medium gluten wheat), YANGMAI13 and YANGMAI15 (weak gluten wheat) achieved 93%. For the binary classification of every two cultivars within the same type, the classification accuracy achieved 100%. And all the binary classification accuracy were over 90%. It could be concluded that hyperspectral image was feasible to be used to identify wheat cultivars from single seed.