Transfer Learning Approach on Bacteria Classification from Microscopic Images

The ability to identify and categorize bacteria is crucial in modern medicine for disease diagnosis, infection treatment, and epidemic investigation. However, manually identification and categorization of bacteria requires a lot of time and effort from humans. As technology has progressed, computer systems-based techniques are now doing the duty of identifying images captured by digital electron microscopes. On top of that, modern Deep Learning (DL) methods have shown remarkable improvement in the area of image classification. In this research, we explore a method for using a DL model to automate the identification and categorization of bacteria. To develop the DL model, we used a dataset consisting of more than 600 images of 33 distinct bacteria taken with a microscope and the ‘transfer learning’ technique. GoogLeNet and AlexNet are two examples of transfer learning models used in this research. The DL classification accuracy was evaluated using 20% randomly selected and isolated images from the dataset. Experimental findings of prediction obtained an accuracy of roughly 98.67% by GoogLeNet, and both transfer learning models recognized and classified all 33 bacterial species with better success rates.