Improved YOLOv8 Model for Phenotype Detection of Horticultural Seedling Growth Based on Digital Cousin

Abstract

Crop phenotype detection is a precision way to understand and predict the growth of horticul-tural Seedling in smart agriculture era, to make the agricultural production more costly and en-ergy efficiency. And it bridges the plant statues and the agricultural devices, like robots and au-tonomous vehicles in smart greenhouse ecosystem, to know each other well. However, the im-aging data set collection is a neckless of deep learning of phenotype detection, as the dynamic coverings among leaves and time-spatial limits of camara sampling. To address this issue, digital cousin is boosting digital twins and virtual entities of plants, and considered to create dynamical 3D structures, attributes and RGB image data sets in a simulation environment, with the princi-ples of varies and interactions in physical world. Thus, this work presents a two-phase method to obtain the phenotype of horticultural seedling growth. In the first phase, 3D Gaussian Splatting is selected to reconstruct and store the 3D model of the plant, enabling to capture RGB images and detect the phenotypes of seedlings transcending temporal and spatial limitations. In the second phase, an improved the YOLOv8 model is created to segment and measure the seedlings, and it is modified by adding modules of the LADH, SPPELAN and Focaler-ECIOU to the original YOLOv8 model. Moreover, a case study of watermelon seeding is explored, and the results show that 3D Gaussian Splatting has good performance in 3D reconstruction of seedlings, and the peak sig-nal-to-noise ratio (PSNR) of the trained models is generally above 24. As for semantic segmenta-tion, compared with the original YOLOv8, the computation of our model decreased by 7.50%, the convergence speed increased by 31.35%.

Plant phenotyping relevance

幼 horticultural seedlings の3D再構成、画像取得、セグメンテーション、表現型測定を統合した手法開発が中心であり、植物表現型の抽出方法を直接扱っている。

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.