AI-Based Detection of Coliform Colonies Using CNN Transfer Learning for Application to Cultured Plate Analysis in Water Quality Research
Mallela, S.; Gates, A.; Bhattacharya, S.; Okeke, B.; Kursun, O.
Show abstract
Pathogenic bacterial contamination of water poses a severe public health risk, particularly in settings with limited laboratory resources. We propose a two-stage artificial intelligence (AI) pipeline for automated detection and classification of coliform colonies on agar plates. In the first stage, a YOLOv8-based detector localizes colonies on full-plate images, eliminating the need for manual annotation. In the second stage, detected colony patches are classified using a convolutional neural network (CNN) trained via transfer learning, where models are first pretrained on a diverse public bacterial colony dataset and subsequently fine-tuned on coliform-specific classification tasks. Across both in-house and public datasets, transfer learning consistently improves classification performance relative to training from scratch. The complete pipeline processes each plate in under five seconds and outperforms classical feature-based baselines, including Histogram of Oriented Gradients, Local Binary Patterns, and Haralick descriptors with conventional classifiers. These results demonstrate the potential of a modular, low-cost AI framework for scalable and accessible microbiological analysis, with future work targeting color-aware models and on-device inference for field deployment.
Matching journals
The top 6 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Deep Learning Classification of Lipid Droplets in Quantitative Phase Images 96%
- Complementary Performances of Convolutional and Capsule Neural Networks on Classifying Microfluidic Images of Dividing Yeast Cells 94%
- microbeSEG: A deep learning software tool with OMERO data management for efficient and accurate cell segmentation 94%
Similar papers in this journal
Similar papers in this journal
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.