NRLC-YOLO for lightweight detection and grasp positioning of latex cups in rubber plantations

Natural rubber harvesting remains highly dependent on manual labor, particularly during latex cup collection, which limits efficiency and increases operational costs. Intelligent robotic harvesting systems require accurate visual perception and reliable grasp point positioning under rubber plantation environments. However, latex cups are typically small, visually diverse, and often affected by adjacent latex drains, making detection and manipulation challenging for existing vision models. To address these challenges, this study proposed a lightweight vision-based framework for latex cup detection and grasp-point positioning in automated rubber harvesting. The proposed NRLC-YOLO is developed based on YOLO11n-seg by integrating a lightweight backbone, a local–global attention enhancement module, and a dynamic convolution strategy. In addition, a fast grasp-point positioning method is designed to determine the gripping center of latex cups, enabling stable robotic manipulation. As validated by experimental results, our model yields a mAP @50–95 of 78.5%, which surpasses the baseline approach with fewer parameters and lower computational costs. The average error of the proposed grasp-point positioning scheme is measured at 8.08 pixels. Field experiments further demonstrate a grasping success rate of up to 93.3% under real plantation conditions. The proposed framework provides an efficient AI-enabled perception solution for automated latex cup harvesting and offers practical support for the development of intelligent rubber plantation management systems.