Ping-Yi ChouChen-Kang Huang2025-12-012025-12-012025-12https://scholars.lib.ntu.edu.tw/handle/123456789/734262Ozone seed cleaning is a promising non-chemical method for microgreen production, yet its effects on germination require precise, real-time evaluation. This study addresses this need by developing an innovative framework that integrates a YOLOv8-BiFPN model and an adjusted germination index (AGI). Hourly imaging of ozone-treated red cabbage and broccoli seeds over 48 h yielded a mean average precision (mAP50-95) of 0.86. Low-dose ozone (96–100 min·mg·m−3) increased broccoli germination efficiency by 29 %, whereas high-dose ozone (224–232 min·mg·m−3) suppressed it. The AGI effectively normalized seeding density variations, enhancing cross-treatment comparisons. This resource-efficient framework supports sustainable seed cleaning evaluation and advances precision agriculture in resource-constrained environments.enDeep learningSeed germination analysisMicrogreensSmart agricultureMicrogreens production[SDGs]SDG2[SDGs]SDG3[SDGs]SDG12journal article10.1016/j.compag.2025.111073