Irregular seeds DEM parameters prediction based on 3D point cloud and GA-BP-GA optimization

Abstract

Due to the small and irregular shapes of vegetable seeds, modeling them is challenging, and the imprecision of physical parameters hinders the performance of vegetable seeders, impeding simulation development. In this study, seeds of cucumber, pepper, and tomato were seen as examples. A 3D point cloud reconstruction method based on Structure-from-Motion Multi-View Stereo (SfM-MVS) was employed to accurately extract 3D models of small and irregularly shaped seeds. Corresponding discrete element models were established. Combining physical and simulation experiments on seed angle of repose(AOR), significant parameters influencing seed AOR and their ranges were identified through Plackett-Burman Design (PBD) and steepest ascent test. Within this range, the GA-BP-GA algorithm was used to accurately inverse the optimal parameter combination. The results indicate that the SfM-MVS 3D point cloud reconstruction method can extract more detailed shape information of small and irregularly shaped seeds. The GA-BP-GA algorithm achieved an inversion of physical parameters with the smallest relative error of cucumber, pepper, and tomato seeds being 0.26%, 0.98%, and 0.51%, respectively. Through experimental comparative analysis, the feasibility and accuracy of this method in calibrating discrete element parameters for small and irregularly shaped seeds were validated. The established seed models and calibrated parameters in this study can be implemented to the simulation optimization design of vegetable seeders, enhancing development efficiency and operational performance.

Plant phenotyping relevance

SfM-MVSによる種子の3D形状取得と、形状情報を用いたDEMパラメータ校正が中心的な技術貢献であり、種子という植物器官の形態計測を扱っている。

Code and data availability

The paper describes seed 3D point cloud reconstruction (SfM-MVS), DEM simulations, and GA-BP-GA parameter calibration for cucumber, pepper, and tomato seeds, but the supplied blocks contain no data availability statement, no public deposit of seed images, point clouds, models, or analysis code, and no author-provided公共

No evidence-backed public reproduction asset is currently recorded.