High-performance 3D morphometrics via deep learning and tabular foundation models: a case study on complex cereal grain classification

Key Points

• The aim is to improve grain classification accuracy using advanced 3D morphometric techniques and deep learning.
• Utilized 3D Gaussian Mixture Models (GMM) and deep learning for grain analysis.
• Employed spherical harmonics coefficients to represent 3D shapes effectively.
• Implemented transformer-based Tabular Foundation Models for classification tasks.
• Binary classification accuracy ranged from 87.76% to 96.47%, with multiclass accuracy at 82.2%.
• Demonstrated that 3D methods outperform traditional 2D approaches in grain classification.
• Provided open data and code for reproducible workflows in 3D model classification.