LBD-PointNet++: a point cloud segmentation network for phenotypic trait extraction of broccoli seedlings

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Breeding

Background:

• Climate change and soil salinization threaten global broccoli production.
• Accurate phenotyping of broccoli seedlings is crucial for identifying salt-tolerant varieties.
• Traditional phenotyping methods are inefficient and prone to errors.

Purpose of the Study:

• To develop an automated 3D point cloud semantic segmentation model (LBD-PointNet++) for broccoli seedling phenotyping.
• To extract phenotypic parameters of broccoli seedlings under salt stress conditions.
• To enhance the selection of salt-tolerant germplasm for improved crop resilience.

Main Methods:

• Reconstruction of high-fidelity 3D point clouds using Structure from Motion (SfM) from precision three-view imaging.
• Development of LBD-PointNet++, incorporating Large Kernel Attention (LKA), Dual Uncertainty and Shape-Adaptive Sampling (DUSAS), and Boundary-Aware Nested Contrastive and Adaptive Varifocal Joint Loss (BNCV-Loss).
• Evaluation of model performance using mean Intersection over Union (mIoU) and Mean F1-score across different salt (NaCl) concentrations.

Main Results:

• LBD-PointNet++ achieved high accuracy with mIoU of 88.07% and Mean F1-score of 93.48% for Leaf, Stem, and Pot segmentation.
• The model demonstrated superior efficiency compared to state-of-the-art methods, with significantly fewer parameters and faster inference speeds.
• A preliminary salt stress phenotypic threshold of 100 mmol/L NaCl was identified, beyond which growth inhibition intensified significantly.

Conclusions:

• LBD-PointNet++ offers an efficient and accurate automated solution for phenotypic analysis of broccoli seedlings under salt stress.
• The findings support the digital breeding of salt-tolerant Brassicaceae crops in response to environmental challenges.
• The study highlights the potential of advanced AI in accelerating crop improvement for sustainable agriculture.