Research on a Low-Damage Garlic Harvester Based on a Dual-Vision Collaborative System
Year: 2026
Type: conference-paper
Abstract: Aiming at the technical challenge of cumulative damage across multiple stages in the combined harvesting of garlic, this study designs an intelligent harvesting system based on dual-vision collaboration. The system adopts a hierarchical perception architecture, deploying vision subsystems at the pre-excavation stage and the screening stage, respectively. Deeply integrated with low-damage operating mechanisms, it constructs a closed-loop damage control system covering the entire process of “excavation-screening-sorting”. The front-end vision unit, based on an improved semantic segmentation model, achieves real-time localization and pose estimation of garlic root clusters in complex field environments, guiding the excavation device for adaptive operation and reducing excavation damage by 71.2%. The core innovation lies in the integrated “hand-eye” sorting unit inside the vibrating screen. This unit, consisting of a flexible robotic arm with an embedded miniature vision sensor at its end-effector, achieves close-range imaging, online damage recognition, and immediate compliant removal of moving garlic under dynamic vibration conditions. The recognition accuracy for damaged bulbs reaches 95.8%, with a comprehensive sorting success rate of 89.7%, while the additional damage rate inflicted on intact bulbs is controlled below 1.5%. Through system collaborative control, the total mechanical damage rate of garlic harvesting is ultimately reduced from over 20% in traditional methods to below 7%. This research provides an innovative systemic solution for efficient, low-loss harvesting of root and tuber crops.
Language: English
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7204161441