SRF-YOLO: A Lightweight Detection Model and Robotic System Integration for Protected Tomato Harvesting

Aiming at the problems of false detection and missed detection caused by complex backgrounds, variable fruit postures, and branch/leaf occlusion in the mechanized harvesting of tomatoes in protected orchard environments, this study proposes a lightweight object detection model named SRF-YOLO, an improved version of YOLOv11n . Firstly, the C2PSA-SHSA module is designed to reduce computational complexity while enhancing the model's perception of occluded and small-scale tomatoes. Secondly, the C3k2-RSCM module is introduced to strengthen multi-scale feature fusion through a rectangular spatial self-calibration mechanism, effectively suppressing background interference. Finally, the F-CIoU loss function is proposed to dynamically adjust the weights of hard and easy samples, accelerating model convergence and improving regression accuracy. Experimental results show that SRF-YOLO achieves a recognition precision of 92.4% and a mean average precision (mAP@0.5) of 93.2% on the tomato dataset, representing improvements of 4.7% and 0.9%, respectively, compared to the baseline model. The model was further deployed on a self-developed multi-arm parallel picking-transporting-collecting integrated robotic system. Based on the ROS framework, full-process automation from visual recognition to coordinated robotic arm operation was achieved. Prototype tests demonstrate that SRF-YOLO exhibits good real-time detection performance, robustness, and engineering applicability in real greenhouse environments, providing a feasible technical solution for intelligent harvesting in protected agriculture.