<![CDATA[This study introduces the implementation of YOLOv8 instance segmentation model, for automating the quantification and identification of multispecies bacterial colonies. In microbiological research, accurate counting and rapid identification of various species within colonies are essential. YOLOv8's real-time capabilities and high accuracy make it particularly well-suited for this task. We demonstrate its efficacy in accurately detecting and segmenting individual bacterial colonies, even when they comprise multispecies like Bacillus subtilis, Escherichia coli, Pseudomonas aeruginosa, and Staphylococcus aureus. This innovation streamlines the labor-intensive processes of colony counting and species identification. To improve precision, we incorporate post-processing techniques to handle overlapping colonies, significantly enhancing accuracy compared to manual counting methods. Our proposed method exhibits superior performance and provides a valuable tool for microbiologists and researchers, expediting bacterial colony analysis. In this study, we fine-tuned hyperparameters to achieve the best mean average precision (mAP) for masks and bounding boxes. We used a primary and a secondary dataset for training. The hyperparameter modifications, including using an Adam optimizer with a learning rate of 0.01 and an epoch value of 50, resulted in mAP values of 90%. These findings underscore the importance of optimizing the optimizer type, learning rate, and the number of epochs, revealing their impact on the model's performance in automating bacterial colony quantification and identification.]]>
Copyrights © 2026
