Classification of bacterial images obtained optically using some pre-trained models

Rapid and accurate detection of water-borne bacteria is critical for safeguarding public health and preventing the spread of infections. Conventional bacterial identification techniques are often time-consuming, resource-intensive, and require specialized personnel, limiting their suitability for automated water-quality monitoring. This study presents an optical detection and classification framework that uses some pre-trained architectures to automatically classify optically acquired images of water-borne bacterial pathogens. The system distinguishes among four classes: Escherichia coli (E. coli), Fecal streptococci (Strept), co-occurrence of both bacteria (Both), and safe water (None). Experimental evaluation demonstrates strong classification performance, with ResNet-50, ResNet-152, and EfficientNet-B7 achieving accuracies of 94.15%, 94.45%, and 95.54%, respectively. Except DenseNet-201, which yields the worst results (accuracy of 77.49%), the corresponding precision, recall, and F1-scores of the other pre-trained models exceed 94%. Furthermore, an analysis of the ROC (Receiver Operating Characteristic) curve reveals that the Area Under the Curve (AUC) exceeds 98% for each bacterial class, thereby demonstrating strong discriminative performance. The results highlight the potential of transfer learning-based convolutional neural networks for accurate, rapid, and cost-effective bacterial detection, emphasizing their promise for scalable and automated water-quality monitoring systems. This approach represents a step toward improved monitoring tools aligned with sustainable water, sanitation, and hygiene (WASH) objectives.