Enhancing Grapevine Node Detection to Support Pruning Automation: Leveraging State-of-the-Art YOLO Detection Models for 2D Image Analysis

• DOI: 10.20944/preprints202409.0322.v1
• Publication Date: 2024-09-05T02:09:34Z
• Authors: Francisco Oliveira; Daniel Queirós da...; Vítor Filipe; Tatiana M. Pinho; Mário Cunha; José Boaventura C...; Filipe N. Santos
• Keywords: Pruning

Automating pruning tasks entails overcoming several challenges, encompassing not only robotic manipulation but also environment perception and detection. To achieve efficient pruning, systems must accurately identify the correct cutting points. A possible method to define these points is choose location based on number of nodes present targeted cane. For this purpose, in grapevine it required correctly primary canes grapevines. In paper, a novel node detection grapevines proposed with four distinct state-of-the-art versions YOLO model: YOLOv7, YOLOv8, YOLOv9 YOLOv10. These models were trained public dataset images containing artificial backgrounds afterwards validated different cultivars from two Portuguese viticulture regions cluttered backgrounds. This allowed evaluate robustness algorithms diverse environments, compare performance used, as well create publicly available obtained vineyards for Overall, all used capable achieving three datasets. Considering trade-off between accuracy inference speed, YOLOv7 model demonstrated be most robust detecting 2D grapevines, F1-Score values 70 % 86.5 times around 89 ms an input size 1280×1280 px. results, work contributes approach real-time further implementation autonomous system.