Quantitative morphological phenotyping of infection structures in cucumber downy mildew and powdery mildew.

Introduction: Cucumber diseases severely affect yield and quality. Deep learning-based analysis of microscopic pathogen images enables high-throughput identification and counting of pathogens, thereby facilitating early disease detection. However, most existing pathogen-recognition methods focus mainly on qualitative identification and cannot quantitatively characterize pathogen morphology, which limits their ability to reveal the developmental characteristics and functional differentiation of different infection structures from the perspective of pathogen morphology-function adaptability. Methods: To address this issue, this study focused on cucumber powdery mildew and downy mildew and achieved precise extraction and characterization of pathogen morphological features based on microscopic image instance segmentation. First, an in situ stained microscopic image dataset of cucumber pathogens was constructed. Second, an instance segmentation model, SWS-YOLO11n, was developed for cucumber pathogen infection structures to accurately identify and segment different infection structures in microscopic images. Finally, morphological analysis methods were used to quantitatively extract and characterize pathogen infection-structure features. Results: Experimental results showed that SWS-YOLO11n achieved a detection mAP@0.5 of 97.4% and a segmentation mAP@0.5 of 96.2%, with a model size of only 5.6 MB. The extraction accuracy of perimeter, area, and curvature reached R Discussion: This study provides an effective tool for high-throughput phenotyping of cucumber pathogen infection structures. The proposed method offers methodological support for disease diagnosis, pathogen morphological phenotyping, and precision disease management in horticultural production.