Development of an intelligent technology for extracting phenotype information from eggplant plants based on deep learning for field work

Abstract

In the field of modern agriculture, plant phenotype analysis is crucial. It directly impacts the accurate assessment of photosynthesis, the refinement of farmland management, and serves as the foundation for crop breeding optimization, effective monitoring of diseases and pests, and reasonable adjustment of planting density. To deepen the application of precision agriculture, this study, combined with deep learning technology, proposed a low-cost hardware and software integration solution for non-contact acquisition of field eggplant plant point cloud data, achieving high-precision phenotypic information extraction.Utilizing depth camera technology, this study designed a multi-view point cloud scanning device capable of capturing point cloud data at different growth stages of eggplant plant in the field, accurately reflecting its natural growth state. In terms of data processing, this study improved the PointMLP model by introducing an attention mechanism and multi-scale feature extraction. These enhancements increased the model's ability to learn complex spatial relationships, making it more suitable for crops with intricate geometric forms. Experimental results indicated that the model achieved excellent performance in eggplant plant stem and leaf segmentation tasks, with Precision, F1-score, Recall, and mIoU reaching 92.35%, 90.4%, 88.7%, and 85.82%, respectively, demonstrating significant competitiveness compared to existing advanced networks.To further improve the accuracy of leaf point cloud extraction, the concepts of "erosion" and "dilation" from image processing were integrated into the DBSCAN algorithm, achieving high-precision extraction of single eggplant leaf point clouds in complex scenes. The Delaunay algorithm was utilized to calculate the area of each eggplant leaf, with an R² coefficient as high as 0.87, verifying the accuracy and reliability of the approach.The innovative solution proposed in this study not only enhances the efficiency and accuracy of plant phenotypic analysis but also reduces labor costs and crop damage, providing robust support for plant phenotypic analysis in large-scale field operations.

Plant phenotyping relevance

深度カメラによる圃場ナスの点群取得、セグメンテーション、葉面積推定を開発・検証しており、植物表現型取得手法が研究の中心である。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.