Trinocular Vision-Driven Robotic Fertilization: Enhanced YOLOv8n for Precision Mulberry Growth Synchronization

This study focused on addressing the issue of delayed root system development in mulberry trees during aerosol cultivation, which is attributed to the asynchronous growth of branches and buds. To tackle this challenge, we propose an intelligent foliar fertilizer spraying system based on deep learning. The system incorporates a parallel robotic arm spraying device and employs trinocular vision to capture image datasets of mulberry tree branches. After comparing YOLOv8n with other YOLO versions, we made several enhancements to the YOLOv8n model. These improvements included the introduction of the Asymptotic Feature Pyramid Network (AFPN), the optimization of feature extraction using the MSBlock module, the adoption of a dynamic ATSS label assignment strategy, and the replacement of the CIoU loss function with the Focal_XIoU loss function. Furthermore, an artificial neural network was utilized to calculate the coordinates of the robotic arm. The experimental results demonstrate that the enhanced YOLOv8n model achieved an average precision of 94.48%, representing a 6.05% improvement over the original model. Additionally, the prediction error for the robotic arm coordinates was maintained at ≤1.3%. This system effectively enables the precise location and directional fertilization of mulberry branches exhibiting lagging growth, thereby significantly promoting the synchronous development of mulberry seedlings.