Abstract
Accurate identification of plant-parasitic nematodes is fundamental to effective crop protection and the maintenance of soil ecosystem integrity. This study integrates morphological characterization with deep learning-based object detection to enhance diagnostic accuracy for economically important nematode genera, Xiphinema and Mesocriconema, associated with vineyards. Three advanced YOLO architectures [YOLO-NAS, YOLOv11, and Roboflow 3.0 (YOLOv8 architecture)] were trained and evaluated on a high-resolution annotated microscopic image dataset consisting of 961 images and 1.034 bounding-box annotations. Although the target nematode genera display considerable morphological variability and genetic divergence among populations, the present investigation focused on genus-level detection of M. xenoplax and X. pachtaicum. These two major ectoparasitic nematodes cause significant damage to grapevine root systems. Among the models tested, YOLOv11 achieved the highest detection accuracy, with a precision of 95.7 % and an mAP@50 of 93.2 %. YOLO-NAS exhibited comparable performance (mAP@50 = 92.7 %, precision = 93.1 %, recall = 84.9 %), while Roboflow 3.0 (YOLOv8 architecture) yielded satisfactory results (mAP@50 = 89.4 %), indicating its applicability for real-time diagnostic workflows. This integration of taxonomic expertise with deep learning represents a new methodological framework for nematode identification. All models exhibited rapid convergence and stable learning dynamics during training. The findings underscore the potential of YOLO-based frameworks as efficient, scalable, and reproducible tools that complement classical morphological and molecular identification, contributing to precision agriculture and sustainable nematode management strategies.