Mechanics of rice grain breakage and YOLOV4-based damage quantification

• Critical impact velocity of rice grains was determined experimentally. • Physical resilience was linked to internal damage using machine learning. • Real-time damage detection system was developed for threshing processes. • Model classified damaged grains with 87% precision and healthy grains with 99%. The mechanical properties of paddy grains and their role in grain damage during threshing remain insufficiently explored, limiting equipment optimisation. This study investigates four rice varieties, Taiyou 398, Taiyou 390, Shenyou 5428, and Yueyou 3700, harvested at full maturity. Controlled compression tests (4–10 mm s −1 deformation rates) were used to measure rupture force, toughness, yield strength, and elastic modulus. Increasing deformation rate decreased rupture force and toughness but increased yield strength and stiffness. Taiyou 398 consistently exhibited the highest mechanical resilience, while Yueyou 3700 was the weakest. Using Hertzian contact theory and energy balance principles, critical impact velocities were estimated as 30.9, 29.6, 28.8, and 28.4 m s −1 for Taiyou 398, Taiyou 390, Shenyou 5428, and Yueyou 3700, respectively. These thresholds aligned with observed damage under threshing speeds of 700–1100 rpm: internal damage for Taiyou 398 rose from ∼10 % to 30 %, and Yueyou 3700 from ∼20 % to 50 %. A YOLOv4-based video classifier quantified internal damage in real time, achieving a mean Average Precision of 92.78 %, with precision scores of 99 % (healthy) and 87 % (damaged), and F1 scores of 0.95 and 0.82. This integrated approach offers actionable insights for reducing post-harvest losses.