Work

Clip-assisted flower detection and wind-compensated precision liquid pollination robot for kiwifruit orchards

Year: 2025

Type: article

Abstract: Pollination in kiwifruit orchards requires both precision and timeliness, and the widely adopted pergola training system provides favorable conditions for automation. In this study, we developed a precision liquid pollination robot for kiwifruit and systematically evaluated its field performance. The robot integrates a liquid spray control system with wind compensation, enabling accurate pollen delivery and significantly reducing pollen consumption. To overcome the bottleneck of labor-intensive data annotation, we proposed a CLIP-based automatic labeling approach to construct a kiwifruit flower detection dataset, which was then used to train a lightweight YOLO model for real-time flower recognition. A modular task control framework, developed entirely in Python, was implemented to coordinate localized spray operations and extend the effective working range of the robotic arm. Field trials conducted in Shaanxi, China, during 2024–2025 demonstrated that the robot achieved a fruit set rate above 85 %, while consuming only 200 g/ha of pollen, which is substantially lower than previous robotic systems, and operating at a speed of 35–40 s/m 2 . To ensure reproducibility, the developed framework and code have been released as open source on GitHub, and datasets have been archived on Google Drive for long-term global access. This research is the first to apply CLIP-assisted flower detection in robotic pollination, and the results confirm its potential for reducing data preparation costs, improving pollination efficiency, and providing a scalable, open robotic platform for intelligent orchard management.

Source: Computers and Electronics in Agriculture

DOI: 10.1016/j.compag.2025.111250

Authors: Wei Hao, Jianwei Zhang, Wang Xu, Xu Fan, Tomás Norton, Yongjie Cui