YOLO-Based Counting of Small and Overlapping Bacterial Colonies: Performance Analysis and Real-Time Mobile Deployment

Accurate counting of bacterial colonies is essential for ensuring microbiological safety and supporting diagnostic, research, and quality control practices in critical domains such as food production, pharmaceuticals, and environmental monitoring. This study presents a comparative evaluation of different YOLO architectures for the automatic counting of small and overlapping bacterial colonies. Using the AGAR dataset, YOLOv3-Tiny, YOLOv5s, YOLOv7-Tiny, and YOLOv8s models were tested. Among them, YOLOv5s achieved the best overall performance, with a mAP@0.5 of 96.4%, recall of 92.5%, and an inference speed of approximately 320 FPS. The model also showed strong agreement with expert annotations, with an average counting error of 1.7 colonies per class. The selected architecture achieved an optimal balance between accuracy and speed, offering a practical and deployable solution for mobile laboratory applications.