A longan-picking sequence planning method for UAV system based on multi-target tracking
Manual harvesting of tall fruit trees is plagued by substantial labor intensity and significant operational risks, while ground-based manipulators encounter substantial challenges in achieving the picking of high-altitude fruits. A picking UAV system presents a viable solution to these issues. Howev...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-12-01
|
| Series: | Smart Agricultural Technology |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2772375525004423 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Manual harvesting of tall fruit trees is plagued by substantial labor intensity and significant operational risks, while ground-based manipulators encounter substantial challenges in achieving the picking of high-altitude fruits. A picking UAV system presents a viable solution to these issues. However, due to limited computational capabilities and inherent flight instability, the UAV cannot reliably generate stable and efficient picking sequence plans. To address this, we propose a lightweight and high-accuracy sequence planning method based on multi-target tracking, designed for UAV-based longan picking. First, a specialized lightweight target detection algorithm, YOLOv8s-Longan, is developed to achieve high-precision target localization and facilitate lightweight model deployment. Second, the YOLOv8-LonTrack multi-target tracking algorithm is designed and proposed, which integrates the detection capabilities of YOLOv8-Longan with BotSort tracking technology, providing stable positional data for sequence planning. Aiming at the scattered distribution of longan string fruit, a spatial-characteristic-based partitioning strategy is designed, complemented by the Corner-Kmeans clustering algorithm, which enhances clustering stability and partitioning accuracy. A comparative analysis shows that the proposed method achieves an average improvement of 71. 7% in silhouette coefficient, a 67% reduction in the average distance within the cluster and a 30. 1% increase in the average distance between groups, allowing effective subdivision of subareas of picking. A LonTsp planning algorithm is used for path planning, optimizing picking paths to reduce total picking time by 19.1% and lowering UAV energy consumption and time costs. Field tests show that the UAV system, integrated with the proposed sequence planning model, reduces overall picking time by 5.4% (104 seconds) in orchards with varying fruit densities, validating the algorithm's effectiveness in real applications. |
|---|---|
| ISSN: | 2772-3755 |