| 要旨トップ | 目次 | 日本生態学会第72回全国大会 (2025年3月、札幌) 講演要旨
ESJ72 Abstract


一般講演(口頭発表) C02-02  (Oral presentation)

Automatic visualization of fine root branching using the deep learning model ARATA【EPA】

*Xingyi ZHAO, Takuto YAMAGATA, Mizue OHASHI(University of Hyogo)

Tree fine roots ≤2 mm in diameter are important organs that absorb water and nutrients from the soil . The abilities of absorbance and transport of nutrients are different between the roots with different orders. Lower-order roots, which are closer to the tips, have a faster turnover rate. However, it is hard to observe them in soil. Also, their complex morphology makes manual analysis slow and difficult. There is a need for faster and more accurate method of analyzing the characteristics of roots with various orders. Recent study utilized ARATA, an automatic root tracing and analysis software, for detecting fine roots in flatbed optical scanner images, enabling long-term dynamic analysis across various study sites. By using ARATA, we aimed to visualize how cedar fine roots make branches for enhancing our understanding of fine root dynamics in this study. We surveyed soil profile in three survey sites in a cedar plantation in Hyogo Prefecture, Japan, using a scanner method over three years. Scanner images were then manually analyzed using GIMP and ImageJ. Then, from each of the three survey sites, nine images with visible roots were selected. These images and manually extracted root data from them were used to retrain ARATA to make a model optimized for the survey sites. Then, automated extraction was conducted on other images and the results were compared with those of the manual analysis. We found ARATA succeeded to visualize fine root branching in high-contrast images. However, distinguishing roots from soil was difficult in low-contrast images. In the future, improvements in image processing and model optimization are needed for more precise automatic analysis of fine root branching.


日本生態学会