A 3D stem diameter measurement method for field maize at jointing stage: combining RLRSA-PointNet++ and structural feature fitting

Abstract

Introduction In precision agriculture, accurate measurement of maize stem diameter during the jointing stage is crucial for lodging resistance assessment and yield prediction. However, existing methods have certain limitations: manual measurement is time-consuming and highly subjective, while two-dimensional image recognition can only capture local features and fails to reconstruct the true three-dimensional structure of the stem. Therefore, there is a critical need for an accurate and automated three-dimensional stem diameter measurement approach. Methods This study proposes a three-dimensional stem diameter measurement method that integrates an improved PointNet++ segmentation network with structural feature fitting, focusing on the position of the second above-ground internode of maize plants. Specifically, multi-view image reconstruction is employed to generate three-dimensional point clouds of maize stems, and Relative Position Encoding, the Local Group Rearrangement Module, and the Local Region Self-Attention mechanism are incorporated into the PointNet++ network to achieve precise segmentation of stems from the ground. On this basis, a structural feature fitting strategy is applied, where principal axis analysis and ellipse fitting are utilized to extract cross-sectional features, thereby obtaining the major axis and minor axis parameters for stem diameter estimation. Results Experimental results demonstrate that the proposed method maintains high accuracy under complex field conditions, achieving a mean absolute error (MAE) of 1.27 mm (R² = 0.87) for major-axis stem diameter and 1.38 mm (R² = 0.82) for minor-axis stem diameter. Discussion The proposed method effectively overcomes the limitations of traditional manual and two-dimensional measurement techniques. It provides a robust and accurate solution for maize stem diameter measurement during the jointing stage. This approach offers technical support for intelligent maize growth monitoring, lodging resistance analysis, and three-dimensional phenotypic trait extraction.

Code and data availability

The supplied blocks describe maize stem point cloud collection, annotation in CloudCompare, and an RLRSA-PointNet++ segmentation method, but contain no public phenotype dataset, image/point cloud deposit, author analysis code, or trained model with an availability statement or authors' URL. CloudCompare and OpenMVS are

No evidence-backed public reproduction asset is currently recorded.