基于实例分割的黄瓜霜霉病菌侵染结构定量表征

Instance Segmentation‑based Quantitative Characterization of Infection Structures of Cucumber Downy Mildew Pathogen

摘要:

黄瓜霜霉病害严重影响其产量和品质，在显微图像中定量表征黄瓜霜霉病菌侵染结构对判断病菌侵染程度、分析侵染行为具有重要意义。针对各类侵染结构尺寸不一、相互交叉等问题，本文基于实例分割网络实现黄瓜霜霉病菌侵染结构定量表征。构建了黄瓜霜霉病菌原位染色显微图像数据集和黄瓜霜霉病菌显微图像实例分割模型DS?YOLO v8s，提升了侵染结构检测和分割能力。基于形态学分析方法构建了黄瓜霜霉病菌侵染结构定量表征体系。基于侵染结构定量表征体系设计并实现了黄瓜霜霉病菌侵染结构定量表征系统。试验结果表明，DS?YOLO v8s模型mAP_box@.5为89.5%，mAP_mask@.5为80.1%。形态分布上孢子、孢子囊、分叉结构和菌丝周长分别为40~90像素、200~350像素、0~1 500像素、0~800像素，面积分别为100~400像素×像素、3 000~6 000像素×像素、0~20 000像素×像素和0~6 000像素×像素，孢子与孢子囊圆形度分别为0.68~0.78和0.70~0.85，菌丝长度为0~300像素。黄瓜霜霉病菌侵染结构定量表征系统实现了用户登录注册、图像分割和定量表征等功能。研究结果为黄瓜霜霉病早期精准监测与智能防控提供了可复用的技术路径，并为抗病育种高通量表型筛选与病害表型组学标准化评价体系建设奠定了方法学基础。

Abstract:

Cucumber downy mildew significantly impacts yield and quality. Quantitative characterization of the infection structures of the cucumber downy mildew pathogen in microscopic images is crucial for assessing the degree of pathogen infection and analyzing infection behavior. To address challenges such as varying sizes and overlapping of different infection structures, the quantitative characterization of cucumber downy mildew pathogen infection structures was implemented based on an instance segmentation network. Firstly, an in?situ stained microscopic image dataset of the cucumber downy mildew pathogen was constructed. Secondly, an instance segmentation model for microscopic images of the cucumber downy mildew pathogen, DS?YOLO v8s, was developed, significantly improving the detection and segmentation capabilities of infection structures. Thirdly, a quantitative characterization system for the infection structures of the cucumber downy mildew pathogen was established by using morphological analysis methods. Finally, a quantitative characterization system for the infection structures of the cucumber downy mildew pathogen was designed and implemented. Experimental results demonstrated that the proposed DS?YOLO v8s model achieved mAP_box@.5 of 89.5% and mAP_mask@.5 of 80.1%. In terms of morphological distribution, the perimeters of spores, sporangia, branching structures, and hyphae were concentrated in 40~90 pixels, 200~350 pixels, 0~1 500 pixels, and 0~800 pixels, respectively;their areas were concentrated in 100~400 square pixels, 3 000~6 000 square pixels, 0~20 000 square pixels, and 0~6 000 square pixels, respectively. The circularity of spores and sporangia was concentrated in 0.68~0.78 and 0.70~0.85, respectively, and the hyphal length was concentrated in 0~300 pixels. The quantitative characterization system for cucumber downy mildew pathogen infection structures implemented functions such as user registration and login, image segmentation, and quantitative characterization. The research result can provide a reusable technical pathway for the early precise monitoring and intelligent control of cucumber downy mildew, and lay a methodological foundation for high?throughput phenotypic screening in disease?resistant breeding and the establishment of a standardized evaluation system for disease phenomics.