Adaptive Symmetry Self-Matching for 3D Point Cloud Completion of Occluded Tomato Fruits in Complex Canopy Environments.

Abstract

As a globally important cash crop, the optimization of tomato yield and quality is strategically significant for food security and sustainable agricultural development. In order to address the problem of missing point cloud data on fruits in a facility agriculture environment due to complex canopy structure, leaf shading and limited collection viewpoints, the traditional geometric fitting method makes it difficult to restore the real morphology of fruits due to the dependence on data integrity. This study proposes an adaptive symmetry self-matching (ASSM) algorithm. It dynamically adjusts symmetry planes by detecting defect region characteristics in real time, implements point cloud completion under multi-symmetry constraints and constructs a triple-orthogonal symmetry plane system to adapt to multi-directional heterogeneous structures under complex occlusion. Experiments conducted on 150 tomato fruits with 5-70% occlusion rates demonstrate that ASSM achieved coefficient of determination (R 2 ) values of 0.9914 (length), 0.9880 (width) and 0.9349 (height) under high occlusion, reducing the root mean square error (RMSE) by 23.51-56.10% compared with traditional ellipsoid fitting. Further validation on eggplant fruits confirmed the cross-crop adaptability of the method. The proposed ASSM method overcomes conventional techniques' data integrity dependency, providing high-precision three-dimensional (3D) data for monitoring plant growth and enabling accurate phenotyping in smart agricultural systems.

Plant phenotyping relevance

トマト果実の遮蔽点群を補完し、果実の長さ・幅・高さを推定する新規アルゴリズムを開発・検証しており、植物形質取得が研究の中心である。

Code and data availability

The paper's Data Availability Statement deposits the study's tomato/eggplant fruit point cloud data on ScienceDB, a public repository, making the paper-specific phenotyping data (3D point clouds of 150 tomato fruits used for completion and trait measurement) publicly actionable.