Enhancing the performance of lotus seed kernel and shell sorting after dehulling: A real-time approach based on deep learning and image processing (YOLOv8)

Lotus seeds represent a high-value agricultural commodity, widely utilised in the food and pharmaceutical industries for their nutritional and medicinal benefits. However, post-harvest processing, specifically the removal of shells, often relies on manual labour or rudimentary equipment that results in low productivity. This study proposes a computer vision system based on the YOLOv8 deep learning framework to automate and enhance the efficiency of post-decortication sorting. The model is designed to detect and classify three distinct components: unshelled seeds, kernels, and shell fragments. Based on these predictions, the system controls pneumatic actuators to perform precise, real-time separation. Trained on a dataset of more than 30 000 images, the model achieved a mean average precision (mAP@0.5) of 92.6%. These results validate the model's capability to accurately identify lotus seed components under high-speed processing conditions. Consequently, this research presents a feasible technological solution to mitigate the labour dependence and optimise the shell residue removal stage in industrial lotus seed processing.