Combining Deep Convolutional Neural Network with Support Vector Machine to Classify Microscopic Bacteria Images

Microorganisms such as Bacteria are responsible for the contamination of numerous infectious diseases such as Cholera, Botulism, Gonorrhoea, Lyme disease, Strep throat, Tuberculosis and so on. Therefore, proper identification and classification of bacteria is essential to prevent the outbreak of such life-threatening diseases. But manual identification and classification of bacteria from microscopic image samples requires professional individuals and reasonable amount of time. However, the process could be automated with the implementation of artificial intelligence (AI) and computer-vision technologies. An effectively trained AI could efficiently classify bacteria and save a large amount of time as well as human-effort. In this paper, a unique approach has been investigated to classify bacteria from microscopic image samples. An AI system has been developed by combining a Deep Convolutional Neural Network (DCNN) with Support Vector Machine (SVM) to perform this operation. Using the transfer-learning method, the Inception V3 DCNN architecture has been modified and retrained with more than 800 image samples of seven separate bacteria species, which are 80% of the image-dataset. The features extracted by the retrained DCNN were then used to train a Support Vector Machine (SVM) classifier. The hybrid network was then tested on the rest 20% of images and the network efficiently classified the images of seven individual kinds of bacteria samples with accuracy-level of around 96%.