Tomato Flower Recognition Method at the Maturity Stage Based on Improved YOLOv8n

摘要

To meet the automation and intelligence requirements for tomato flower pollination tasks in plant factories, and to address the challenge of low detection accuracy caused by the small size, dense distribution, and varying orientations of tomato flowers, this study proposes an improved object detection method based on the YOLOv8n network. Specifically, the original C2f module in the YOLOv8n Backbone is optimized by integrating the iRMB (Inverted Residual Mobile Block) and SWC (Shift-Wise Convolution) modules, forming a novel C2f-iRMB-SWC structure. The iRMB module enhances feature representation through dimension expansion in the intermediate layers, while the SWC module effectively enlarges the receptive field with minimal computational cost, enabling the model to better capture multi-scale and complex feature information of tomato flowers. The improved model YOLOv8n-C2f-iRMB-SWC achieves a recall of 89.55% and a mAP@50 of 91.46% on the experimental dataset. The results demonstrate that the proposed model significantly enhances detection performance of tomato flowers at the maturity stage, providing reliable technical support for the practical deployment of pollination robots in smart agriculture. In addition, the study discusses the model’s limitations regarding generalizability to other crop types and highlights future research directions, including robustness enhancement under variable lighting conditions and lightweight optimization for real-time deployment on embedded devices.