A non-destructive and highly efficient model for detecting the genuineness of maize variety 'JINGKE 968′ using machine vision combined with deep learning

Seed genuineness and varietal purity are key indicators of seed quality. Detecting the genuineness of a single seed can simultaneously determine seed purity. The traditional methods for detecting seed genuineness or identifying a variety are time-consuming, costly, and destructive. This study intends to establish a low-cost, efficient, and non-destructive method to detect the genuineness of single maize seeds, based on RGB images combined with deep learning. Eight hundred maize seeds of JINGKE 968 from different lots in different years and 800 seeds of other varieties were selected. Scanned images of both the germ and non-germ surfaces of the seeds were collected. The images were divided into a training set and a validation set according to the ratio of 7:3. A total of 17,600 images were obtained after data augmentation. The VGG16 network was used for transfer learning after fine-tuning, to identify and classify the seed images, and then to establish the model to detect the genuineness of the maize variety 'JINGKE 968'. The results show that the optimal detection accuracy was over 99%, and the model loss was maintained at about 0.05. Another 100 suspected samples were tested, and the recognition accuracy was as high as 98%. In summary, this study provided a non-destructive, highly efficient, fairly reliable, simple and cost saving method to identify true and false individuals of JINGKE 968. These results can serve as a reference to identify the genuineness of seeds for other crops.

Automated summary

This study established a low-cost, efficient, and non-destructive method using deep learning and RGB images to detect the genuineness of single maize seeds. The model achieved an optimal detection accuracy of over 99% for the maize variety 'JINGKE 968' and showed a recognition accuracy as high as 98% for other varieties, providing a simple, cost-saving approach for seed identification.