An efficient and low-cost 3D phenotyping framework for tomato seedlings via neural radiance fields and PointNet++

Abstract

Accurate and efficient plant phenotyping is essential for modern precision agriculture. as it provides reliable information for seedling quality evaluation, early detection of plant stress, and data support for crop breeding and yield prediction. Traditional three-dimensional (3D) reconstruction and analysis methods are often costly and time-consuming, because they usually depend on expensive laser scanning devices or require many input images. Even with these resources, they often fail to capture fine plant structures such as leaves and branches, which limits their application in seedling monitoring. To address these challenges, we propose an integrated framework that combines neural radiance fields (NeRFs) for high-fidelity 3D reconstruction, PointNet++ for robust semantic segmentation, and a customized algorithm for extracting key morphological parameters of tomato seedlings. The proposed framework can be used to reconstruct detailed 3D models at a low computational cost using only ordinary cameras and a limited number of 2D images. We validate the framework on the basis of a tomato seedling dataset and show that our approach outperforms traditional multiview stereo scanners and simple commercial 3D scanners in terms of both detail and efficiency. The accuracy of plant part segmentation reaches 90 %, and the extracted parameters (e.g., leaf area, stem height, branch angle, and internode distance) are highly correlated with the manual measurements (e.g., R 2 = 0.875 for the leaf area). This study provides a low-cost and scalable solution for 3D plant analysis, with direct benefits for automated monitoring of seedling quality in nursery production. Moreover, the proposed framework can be extended to other crops with complex structures, thus supporting wider applications in smart agriculture.

Plant phenotyping relevance

トマト苗の3D再構成、植物部位分割、形態形質抽出を統合した低コスト画像ベース手法を開発・検証しており、植物フェノタイピングが研究の中心である。

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.