Dynamic analysis of the infection process of cucumber powdery mildew based on instance segmentation

Abstract

Powdery mildew represents a significant threat to cucumber yield, with its infection process encompassing stages such as “attachment, colonisation, and dispersal.” With the advancement of deep learning, computer vision techniques are increasingly applied to study powdery mildew infection patterns. However, existing biological methods are low-throughput, subjective, and struggle to capture the dynamic characteristics of pathogen infection throughout the entire process. Current microscopic image analysis methods also fail to simultaneously recognise and segment various infection structures across different stages of infection, making it difficult to reveal the evolving infection patterns over time. To overcome these limitations, this paper proposes an integrated SR-QC-TA framework for modelling the infection behaviour of cucumber powdery mildew. First, a time-series dataset of microscopic images covering all stages of infection was constructed, systematically documenting the evolution of key infection structures from attachment to dispersal. Second, an instance segmentation algorithm, FS-YOLOv8s, was developed to achieve high-precision, multi-class recognition of pathogen structures in complex backgrounds. Additionally, a multi-dimensional quantitative characterisation method for pathogen infection features was designed, describing infection characteristics in terms of quantity, morphology, and location. Finally, based on these recognition and characterisation results, a temporal analysis framework was established to quantify dynamic changes in infection and reveal the stages of infection behaviour. Experimental results demonstrate that FS-YOLOv8s achieved mAPᵇᵒˣ@0.5 and mAPᵐᵃˢᵏ@0.5 scores of 91.8 % and 92.3 %, respectively, enabling high-precision segmentation across all infection stages. This research advances intelligent monitoring and control of cucumber powdery mildew and drives disease monitoring in horticultural crops toward bioengineering systems.

Plant phenotyping relevance

キュウリうどんこ病の感染状態を顕微鏡画像からインスタンスセグメンテーションで抽出し、感染構造の数量・形態・位置と時間変化を定量化する手法が研究の中心であるため。

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.