A Deep Learning-Based Approach for High-Throughput Hypocotyl Phenotyping

Abstract

Hypocotyl length determination is a widely used method to phenotype young seedlings. The measurement itself has advanced from using rulers and millimeter papers to assessing digitized images but remains a labor-intensive, monotonous, and time-consuming procedure. To make high-throughput plant phenotyping possible, we developed a deep-learning-based approach to simplify and accelerate this method. Our pipeline does not require a specialized imaging system but works well with low-quality images produced with a simple flatbed scanner or a smartphone camera. Moreover, it is easily adaptable for a diverse range of datasets not restricted to Arabidopsis ( Arabidopsis thaliana ). Furthermore, we show that the accuracy of the method reaches human performance. We not only provide the full code at https://github.com/biomag-lab/hypocotyl-UNet, but also give detailed instructions on how the algorithm can be trained with custom data, tailoring it for the requirements and imaging setup of the user.

Plant phenotyping relevance

幼苗胚軸長を画像から高速・高スループットに推定する深層学習手法の開発であり、植物形質取得が研究の中心です。

Code and data availability

The paper's authors publicly released their full analysis code (U-Net-based hypocotyl segmentation/measurement pipeline) on GitHub and the training images used for phenotyping on Kaggle. Trained models are only available upon request and are therefore not listed as public assets.