A novel method to detect stem and fruit dynamically for apricot posture estimation and adjustment

Separating pulp and core is critical for apricot processing, but faces labor shortages. To address this challenge, a fully automated pitting machine (FAPM) based on automatic apricot orientation device (AAOD) was proposed to achieve mechanized pitting by apricot automatic orientation. The designed and constructed AAOD adopt with dynamic visual detection and mechanical orientation for apricot posture adjustment. YOLOv8 series models were applied for apricot and stem detection, and then estimating their three-dimensional posture. Compared with other YOLOv8 series models, YOLOv8n was selected as the preferred detection model with a detection speed of 10.3 ms and a size of 6.1 MB to meet the need of real-time detection and lightweight deployment. YOLOv8n achieved precision ( P ), recall ( R ), and mean average precision ( mAP ) values of 82.0 %, 90.9 %, and 90.1 %, respectively. Moreover, new indicators, namely positional offsets in the image coordinate system ( Offset img ), positional offsets ( Offset 3D ), angular offsets in the 3D coordinate system ( Offset ang ), and the ratio of intersection to manual bounding box areas ( Ratio im ), were proposed to validate the performance of AAOD for position estimation in three varieties of apricot. The best performance was obtained in Saimaiti apricot and achieved Offset img of 2.9 pixels, Offset 3D of 1.2 mm, and Offset ang of 0.9°, with Ratio im for apricot and stem were 99.3 % and 97.3 %. Experimental show that the optimal operating parameters for AAOD are 20 rps for alignment wheel rotation speed and the distance of 22.5 mm from apricot base to alignment wheel axis, which presented the best successful orientation rate of 91.5 % with an Offset 3D of 1.8 mm. Result demonstrated that the dynamic detection-based orientation approach proposed in this study has great potential for automatic apricot pitting.