Detección inteligente de puntos de poda en viñedos mediante visión por computador e integración embebida

Abstract

This work develops an embedded computer vision system for the automatic detection of pruning points in grapevine plants, aiming to improve the accuracy and efficiency of pruning processes in precision viticulture. The study hypothesizes that integrating 2D object detection techniques with 3D geometric perception enables reliable estimation of the agronomic cutting location. The project was developed using Intel RealSense RGB-D cameras, NVIDIA Jetson Orin platforms, and YOLO11 detection models trained with optimized data augmentation strategies. Image processing allowed the identification of buds and cane segments, which were then projected into three-dimensional space to compute the optimal pruning point. The obtained results showed that the YOLO11m model achieved the highest performance in terms of precision and mAP50, validating its suitability for the final system implementation. In conclusion, the proposed system represents a functional technological solution for automating grapevine pruning and establishes the foundation for its future integration into autonomous agricultural robots. Keywords: computer vision, smart pruning, embedded models, precision viticulture