KAN-GLNet: An enhanced PointNet++ model for canola silique segmentation and counting

Area of Science:

• Agricultural Science
• Computer Vision
• Plant Biology

Background:

• Accurate plant phenotypic trait analysis is vital for crop breeding and precision agriculture.
• Existing methods for plant phenotyping often lack efficiency and precision.

Purpose of the Study:

• To develop a lightweight yet accurate semantic segmentation model for canola siliques.
• To enable high-throughput plant phenotyping through automated analysis.

Main Methods:

• Proposed KAN-GLNet (Kolmogorov-Arnold Network with Global-Local Feature Modulation), an enhanced PointNet++ architecture.
• Integrated an optimized Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm for counting.
• Utilized Neural Radiance Fields (NeRF) for high-fidelity point cloud reconstruction.

Main Results:

• KAN-GLNet achieved 94.50% mIoU, 96.72% mAcc, and 97.77% OAcc in semantic segmentation, outperforming baseline models.
• Optimized DBSCAN achieved 97.45% accuracy in instance segmentation (counting).
• The model has only 5.72M parameters, balancing accuracy and complexity.

Conclusions:

• KAN-GLNet provides an efficient and accurate solution for canola silique segmentation and counting.
• This method advances high-throughput plant phenotyping capabilities.
• The developed model and dataset are publicly available for research.