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

Area of Science:

• Microbiology
• Computer Science
• Environmental Science

Background:

• Accurate detection of water-borne bacteria is crucial for public health.
• Traditional methods are slow, costly, and require expertise, hindering automated monitoring.
• There is a need for rapid, cost-effective bacterial identification systems.

Purpose of the Study:

• To develop and evaluate an optical detection and classification framework for water-borne bacterial pathogens.
• To utilize pre-trained convolutional neural network (CNN) architectures for automated image classification.
• To distinguish between Escherichia coli (E. coli), Fecal streptococci (Strept), co-occurrence, and safe water.

Main Methods:

• An optical detection system captured bacterial images.
• Pre-trained CNN models (ResNet-50, ResNet-152, EfficientNet-B7, DenseNet-201) were employed for classification.
• Performance was evaluated using accuracy, precision, recall, F1-score, and Receiver Operating Characteristic (ROC) analysis.

Main Results:

• EfficientNet-B7 achieved the highest accuracy (95.54%), followed by ResNet-152 (94.45%) and ResNet-50 (94.15%).
• Most models demonstrated precision, recall, and F1-scores exceeding 94%, with Area Under the Curve (AUC) over 98% for all bacterial classes.
• DenseNet-201 showed lower performance (77.49% accuracy).

Conclusions:

• Transfer learning-based CNNs offer accurate, rapid, and cost-effective bacterial detection.
• The developed framework shows promise for scalable and automated water-quality monitoring.
• This approach supports sustainable water, sanitation, and hygiene (WASH) initiatives.