An improvement of Gram-negative bacteria identification using convolutional neural network with fine tuning

Abstract

This paper proposes an image processing approach to identify Gram-negative bacteria. Gram-negative bacteria are one of the bacteria that cause lung lobe damage-bacterial samples obtained through smears of the patient's sputum. The first step bacterium should pass the pathogen test process. After that, it bred using Mc Conkey's media. The problem faced is that the process of identifying bacterial objects is still done manually under a fluorescence microscope. The contributions offered from this research are focused on observing bacterial morphology for the operation of selecting shape features. The proposed method is a convolutional neural network with fine-tuning. In the stages of the process, a convolutional neural network of the VGG-16 architecture used dropout, data augmentation, and fine-tuning stages. The main goal of the current research was to determine the method selection is to get a high degree of accuracy. This research uses a total sample of 2520 images from 2 different classes. The amount of data used at each stage of training, testing, and validation is 840 images with dimensions of 256x256 pixels, a resolution of 96 points per inch, and a depth of 24 bits. The accuracy of the results obtained at the training stage is 99.20%.