An Optimized YOLO-ViT Hybrid Model for Enhanced Precision in Rice Classification and Quality Assessment

Abstract

Rice holds strategic importance in economic and nutritional value, making accurate classification and quality assessment essential for agricultural production and market supply chains. This study introduces an innovative approach that combines the strengths of the You Only Look Once (YOLOv8) and Vision Transformer (ViT) models to enhance the classification of five key rice varieties like Tarom, Shiroodi, Fajr, Neda, and Basmati. That provides a comprehensive quality assessment. YOLOv8 enables rapid and precise detection of rice grains in images, while ViT captures complex spatial relationships and dependencies among image features, improving the model's ability to handle intricate patterns and contextual information. Three scenarios are explored: Scenario I employs a standalone YOLOv8 model; Scenario II implements a YOLO-ViT hybrid model for extracting spatial and relational features, and Scenario III integrates YOLO-ViT for combined detection and quality evaluation. The results demonstrate that the hybrid YOLO-ViT model significantly enhances classification accuracy and quality assessment, highlighting its effectiveness for agricultural quality control and food supply chain management. This approach innovatively leverages YOLO's fast object classification capabilities and ViT's ability to model complex relationships, providing a high-precision and efficient solution for rice quality evaluation., The proposed model has the potential to improve food safety and facilitate effective market management.The model is widely applicable in automated agricultural systems.