Precise detection of Eimeria oocysts in sheep: a deep learning model based on microscopic images.

BACKGROUND

Parasitic infections remain a major cause of productivity loss in global livestock production. Traditional microscopic diagnostic methods are labor-intensive and require specialized veterinary expertise. Recent automated detection systems are hindered by limited annotated microscopy datasets and the difficulty of extracting discriminative features from small, overlapping targets.

METHODS

We propose YOLO-GA, an enhanced object detection framework, for accurate identification of Eimeria oocysts in ovine microscopy images. Built upon the YOLOv5's architecture, the model incorporates two lightweight attention modules: (1) Contextual Transformer (CoT) blocks for local-global contextual enhancement and (2) Normalized Attention Mechanisms (NAM) for adaptive feature recalibration. The proposed model is optimized for both accuracy and computational efficiency.

RESULTS

Experiments on a curated dataset of 2000 microscopy images (200× magnification) demonstrated that YOLO-GA achieves a mean (± standard deviation) average precision (mAP@0.5) of 98.9% ± 0.1, with 95.2% ± 0.3 precision and real-time inference speed. Comparative evaluations against recent detectors, including YOLOv8, YOLOv10 and DETR variants, confirmed the superior performance of YOLO-GA across multiple runs.

CONCLUSIONS

YOLO-GA offers a high-accuracy solution with balanced computational efficiency for automated detection of Eimeria oocysts under complex microscopy conditions. This work lays a foundation for intelligent diagnostics of ovine Eimeria coccidiosis and provides a reference for scalable health monitoring of sheep flocks, with potential extension to other small ruminant coccidiosis (e.g. goat Eimeria) pending further validation.