Deep Convolutional Neural Network for Microscopic Bacteria Image Classification

Abstract

Bacterial classification is crucial in medical science in order to diagnosis specific causative agent of numerous fatal diseases. Moreover, treatment of bacterial infection has also been possible through understanding of its nature and classification. Besides that, manual identification process of bacteria is a methodical, time consuming and laborious process that needs accuracy and precision. However, a new window of opportunity has opened with the development of machine learning method to identify bacteria automatically from digital electron microscopy. Hence, this paper has presented an automatic system to recognize at the same time classify bacteria from microscopic image using deep convolutional neural network (CNN) namely, `Xception architecture' based on transfer learning. Here, we have chosen 7 varieties of bacteria for recognition which might proof lethal for human and prepare a dataset of 1150 bacteria images where each variety contains atleast 160 images. We trained the proposed system on 920 bacteria images of 7 varieties from train dataset. Then, finally the performance has been evaluated on 230 bacteria images of 7 varieties from test dataset which shows promising performance with approximately 97.5% prediction accuracy in bacteria image classification.