分割
生物
点云
特质
象形文字
多叶的
图像分割
杠杆(统计)
植物茎
性状
农学
纤维作物
空间分布
植物
数学
种质资源
统计
空间分析
空间变异性
作者
Lizhi Jiang,Javier Rodriguez-Sanchez,Peng Wah Chee,Changying Li,Longsheng Fu
标识
DOI:10.1016/j.aiia.2026.07.006
摘要
High-resolution point cloud provides detailed three-dimensional (3D) spatial information about cotton plants, making them valuable for analyzing complex architectural traits. However, segmentation of cotton bolls and other plant components under dense foliage remains underexplored, although accurately measuring boll-related and architectural traits before defoliation reflects cotton’s natural growth conditions and is valuable for breeders. The goal of this study was to leverage a state-of-the-art transformer-based point cloud segmentation model, Point Transformer V3 (PTv3), to segment cotton plants with dense foliage into distinct organs and derive architectural traits. PTv3 was compared with three other models: PointNet++, PointNeXt, and PTv2, to segment cotton plants into distinct organs with up to four classes on two datasets with different levels of challenge. Post-processing methods were developed to count bolls and derive other architectural traits. The experimental results showed that PTv3 achieved the best segmentation performance across all tested segmentation scenarios. The mean absolute percentage error (MAPE) for cotton boll counting, main stem length, main stem diameter, branch angle, and internode distance were 9.55%, 4.09%, 9.23%, 7.04%, and 14.16%, respectively. Moreover, the spatial distribution of cotton bolls and foliage along the vertical direction in different genotypes was successfully obtained and analyzed. This method provides a promising approach for analyzing leafy cotton plants by enabling accurate extraction of structural and yield-related traits under natural growth conditions. The phenotyping information supports the evaluation of architectural differences prior to harvest, thereby advancing cotton phenotyping research and improving breeding efficiency.
科研通智能强力驱动
Strongly Powered by AbleSci AI