Automatic Detection and Classification of Ladybird Beetles in Wildlife Images

This work aims to improve the early detection of ladybird beetles (Coccinellidae), which are important natural predators of agricultural pests, but can also become invasive, by evaluating lightweight YOLO-based object detection models on wildlife images. We hypothesize that small YOLO architectures can achieve high detection accuracy with an efficiency suitable for real-world monitoring. As a contribution, we tested three compact models (YOLOv10, YOLOv11, YOLOv12) trained and validated on 2,899 images from the iNaturalist database, collected in Ecuador, Colombia, Chile, Peru, and Bolivia. YOLOv11 achieved the best performance at a 0.6 confidence threshold, with mAP@50 of 0.876 and 0.868 in training and test sets, respectively, demonstrating comparable results to state-of-the-art methods and robust generalization to realworld conditions.