An Enhanced YOLOv7-Based Framework for Wheat Kernel Classification

Contamination during durum wheat processing presents critical challenges to product quality, food safety, and equipment reliability. Conventional rule-based inspection techniques often exhibit limited effectiveness in distinguishing wheat kernels from visually similar impurities, particularly under complex and dynamic industrial environments. This study proposes a conceptual framework for real-time classification in wheat sorting applications by incorporating the YOLOv7 object detection architecture with a gamma correction-based contrast enhancement module. The gamma correction step serves as a preprocessing mechanism that improves the visual separability of wheat kernels and non-grain materials by selectively enhancing local luminance variations. This facilitates more accurate downstream feature extraction and object classification. The proposed framework defines its recognition scope using a three-class dataset consisting of durum wheat kernels, solid impurities, and plant-based impurities. Although empirical validation has not yet been conducted, the approach establishes a methodological basis for future research into efficient and scalable grain inspection systems. Potential applications include high-throughput sorting, automated quality control, and real-time monitoring solutions for agricultural processing environments.