An automatic non-invasive classification for plant phenotyping by MRI images: An application for quality control on cauliflower at primary meristem stage

Abstract

During the past few years, milder autumn and winter seasons have caused severe problems to cauliflower harvest of Brittany region in France, mainly due to curd deformation. Consequently, cauliflower breeders are working on breeding new varieties that are more robust to climate change to stabilize the quality of cauliflower production. The aim of this study was to identify at which stage of the curd formation, significant difference can be detected between healthy and stressed cauliflower. A non-invasive classification based on Magnetic Resonance Imaging (MRI) images for cauliflower phenotyping was proposed. Plants exposed to vernalization stress were sampled at different times around primary meristem stage, then both MRI imaged and apex dissected. A work flow was developped to extract features from MRI images. A classification on phenotype was learned by LDA, QDA, PLSDA and CNN binary classification between two groups: healthy and stressed cauliflower. Promising F1 score and MCC up to 95% were achieved. Curd deformation is the main cause for cauliflower’s later physiological disorders when reaching maturity. Therefore, the cauliflowers with deformation could be removed at the earliest, e.g., screening for plant breeding. At the same time, the healthy cauliflowers are not destroyed and continue their life cycle.

Plant phenotyping relevance

MRI画像からカリフラワーの健全・ストレス状態を分類する非侵襲的表現型解析ワークフローを開発し、複数の分類器で性能評価しており、表現型取得・抽出法が中心である。

Code and data availability

The paper's MRI phenotyping analysis code and a sample dataset are not publicly deposited; the authors state they are available upon request to the corresponding author. No public URL for these assets is provided.

No evidence-backed public reproduction asset is currently recorded.