ABSTRACT Environmental monitoring and public health require proper bacterial detection in water samples. This study proposes an efficient implementation of the faster R‐CNN architecture for detecting bacteria in microscopic imagery within a multi‐target setting. The data used in this study were obtained from a private laboratory and are subject to confidentiality restrictions. The dataset includes 800 training images with 200 test images of four bacterial classes, which are stained Escherichia coli , unstained E . coli , Pseudomonas aeruginosa , and Staphylococcus aureus . They added data augmentation, pixel intensity normalization, and a multi‐scale feature pyramid network (FPN) to enhance model robustness. The experimental results indicate that the highest performance was demonstrated by stained E. coli with the following values of 87.07% precision, 100% recall, 92.25% F1‐score, and 100% accuracy. Unstained E. coli also had a high performance, whereas P. aeruginosa had a high recall and a low precision. Conversely, the most unstable findings were determined in S. aureus because of its spherical morphology and unstable staining. These results show that staining is an effective method to enhance detection accuracy, especially for E. coli , and some bacterial classes might need further optimization. In general, the optimized faster R‐CNN model offers an efficient way of automating bacterial detection in water samples, and it has the possibility of real‐time monitoring of water quality. The next step in future research will be to improve stability in the detection of difficult types of bacteria and increase the dataset to enable better generalization.
Mursyidah et al. (Wed,) studied this question.