AI-Based Detection of Coliform Colonies Using CNN Transfer Learning for Application to Cultured Plate Analysis in Water Quality Research

Pathogenic bacterial contamination of water poses a severe public health risk where laboratory resources are scarce. We propose a two-stage Artificial Intelligence (AI) pipeline for automated detection and classification of coliform colonies on agar plates. In Stage 1, a YOLOv8 detector localizes every colony (replacing manual ImageJ annotation) and achieves a mean average precision at 0.5 intersection-over-union (mAP@50) of 87.6% on 105 held-out public images. In Stage 2, each detected patch is classified by a convolutional neural network (CNN) that is first “warmed up” via pretraining on ten classes drawn from a 24-class public bacterial-colony dataset (∼5,000 patches) and then fine-tuned on two separate four-class tasks: our in-house-collected coliform dataset (80/20 train/test split), where accuracy rose from 73% (no pretraining) to 86%, and an independent four-class subset from the same public dataset, where accuracy reached 91%. The full pipeline processes each plate in under five seconds. Comparative baselines using Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP), and Haralick features with Support Vector Machine (SVM) classifiers underscore the deep-learning approach’s superiority. Future work will integrate full-color media cues and contextual metadata, and optimize on-device inference for truly portable field deployment.