Fruit Recognition and Localization for Robotic Harvesting Based on an Improved YOLOv11 Model

With the advancement of intelligent agriculture, autonomous fruit-picking robots have become an increasingly prominent research focus. To address the trade-off between accuracy and efficiency faced by traditional object detection methods in real orchard environments, this paper proposes a fruit recognition and localization algorithm based on an improved YOLOv11 model. By incorporating the Powerful-IoUv2 (PIoUv2) loss function and the simple parameter-free attention module (SimAM), the improved model achieves mAP50 and mAP50-95 scores of 96.2% and 84.9%, respectively, representing improvements of 0.5% and 0.9% over the original YOLOv11 without increasing the number of parameters. The average processing time per image is 22 ms. Experimental results demonstrate that the proposed algorithm enables fast and accurate recognition and spatial localization of target fruits, effectively handling challenges such as varying natural lighting and partial occlusion. This work provides a strong technical foundation for efficient and intelligent fruit-picking applications.