Computer Vision for Precision Orchard Inventory: Tree Detection and Planting Density Mapping

alexeykutyrev

alexeykutyrev Federal Scientific Agroengineering Center VIM

Keywords: Automated orchard inventory, Computer vision, YOLO26, Deep learning, Plant-ing density monitoring, Robotic platform, Tree trunk detection, Trellis posts, Or-chard mapping, Precision horticulture

Abstract

This article presents a comprehensive method for the automated inventory of intensive industrial orchards using computer vision. The study aims to develop and validate an integrated software solution based on the state-of-the-art YOLO26 deep learning model for counting tree trunks and trellis posts, as well as for spatial analysis of planting density and distribution uniformity. As part of the research, the YOLO26 architecture was adapted (transfer learning), and a specialized dataset reflecting commercial orchard conditions was assembled and annotated. Primary data were collected using a ground robotic platform equipped with a DJI Action 5 Pro camera and a high-precision RTK-GNSS receiver for positioning and georeferencing. The developed software not only performs object detection but also conducts statistical analysis, visualizes spatial distribution, and generates orchard status maps. Experimental testing on a model plot confirmed the system's effectiveness. The YOLO26 Medium model demonstrated a precision of 0.919 and a recall of 0.846. The software revealed an overall tree deficit of 4.95% compared to the target and a statistically significant non-uniformity in their spatial distribution, automatically identifying zones with reduced planting density. The developed system automates labor-intensive monitoring and provides an objective foundation for agronomic decision-making.