Wheat varieties identification based on a deep learning approach

Wheat variety recognition and authentication are essential tasks of the quality assessment in the grain chain industry, especially for seed testing and certification processes. Recognition and authentication by direct visual analysis of grains are still achieved manually. The automatic approach, based on computer vision and machine learning classification, provided rapid and high throughput methods. Even thus, the classification task stays a complex and challenging case at the varietal level. The present work proposes a deep learning-based approach that provides an accurate classification for wheat varietal level classification (VLC). Particularly, the Convolutional Neural network (CNN) was used to classify the wheat grain image into four varieties (Simeto, Vitron, ARZ, and HD). Furthermore, five standard CNN architectures were trained based on Transfer Learning to boost the classification performance. To assess the proposed models' quality, we used a dataset of 31,606 single-grain images collected from different Algeria regions, and their images were captured using different scanners. The results showed that the test accuracy ranged from 85% to 95.68% for varietal level classification. The best test accuracy was obtained with the DensNet201 architecture (95.68%), Inception V3 (95.62%), and MobileNet (95.49%). Hence, the proposed approach results are accurate and reliable, encouraging the deployment of such an approach in practice.