SPGrowing: an unsupervised individual plant segmentation pipeline from UAV-derived point clouds for high-throughput in-field maize phenotyping

In-field plant phenotyping is essential for understanding gene–environment interactions and supporting crop breeding. While unmanned aerial vehicle (UAV)-based photogrammetry enables high-throughput three-dimensional (3D) sensing, accurate plant-level analysis of densely planted crops like maize remains challenging due to morphological complexity, leaf occlusion, and dynamic canopy structures. To address these issues, this study develops an end-to-end phenotyping pipeline based on UAV-derived point clouds, incorporating cross-circling oblique (CCO) flight sampling, point cloud reconstruction and preprocessing, individual plant segmentation, and trait extraction, which is applied to multiple maize varieties (ZD 958, SKT 1506, SBR), growth stages (V3–R2), and planting spacings (0.4–0.8 m). At the core of the pipeline, an unsupervised individual plant segmentation algorithm (SPGrowing) is proposed, which integrates stem core detection via ground-projected density analysis, superpoint-based structural abstraction for efficient homogeneous partition representation, and multi-source shortest-path optimization incorporating stem-centered growth patterns as topological constraints. Validation results showed that SPGrowing consistently outperformed existing segmentation benchmarks, achieving an overall accuracy (OA) of 0.950 and a mean intersection over union (mIoU) of 0.898 at the point level, and an AP 50 of 0.949 at the object level. The extracted plant-level traits closely matched manual measurements, with a mean absolute error (MAE) of 0.068 m and a coefficient of determination ( R 2 ) of 0.975 for plant height. Multi-dimensional architectural traits, including plant height, canopy volume, and height-to-width ratio, were further analyzed across varieties and growth stages, revealing distinct architectural profiles among cultivars. This study provides a scalable tool for acquiring fine-grained phenotypic data under field conditions, supplying valuable data support for precision field management and breeding. The source code and dataset are available at https://github.com/Joyar546/SPGrowing.git .