The aim of this study was to address the issues of inaccurate picking point localization caused by overlapping and occlusion of ridge-planted strawberry fruits, as well as storage quality deterioration induced by postharvest perishability. A lightweight object detection model was developed to rapidly and precisely localize strawberry fruits and picking points, thereby providing technical support for intelligent harvesting and postharvest supply chain quality control. An improved YOLOv8-SCE model was developed by replacing the C2f module in the YOLOv8 backbone with the C2f-fasternet module to reduce computational redundancy, introducing an Efficient Multi-Scale Attention (EMA) mechanism to enhance multi-scale feature representation and adopting the SIoU loss function for bounding-box regression optimization. Additionally, a novel picking point prediction algorithm was proposed based on fruit-pedicel spatial relationships. The experimental results demonstrate that the improved model achieved a precision (P), recall (R), and mean average precision (mAP@0.5) of 99.23%, 98.11%, and 99.02%, respectively, on the test set, with a 25.22% reduction in parameter count and an 8.93 FPS improvement in inference speed. The accuracy of picking point prediction reached 91.11%, with single-fruit prediction time of only 96 ms. Practical deployment tests show that both recognition and localization algorithms maintained excellent performance, achieving accuracy rates above 92.07%. Compared with traditional machine vision methods and mainstream deep learning models, the proposed method exhibited superior precision and enhanced time efficiency. These findings confirm that the YOLOv8-SCE model and its associated picking point prediction method provide a solid theoretical foundation and technological innovation pathway for achieving intelligent, precise harvesting operations in strawberry-harvesting robots.