Classification of Burn Images into 1st, 2nd, and 3rd Degree Using State-of-the-Art Deep Learning Techniques

Abstract

To choose the right treatment for healing a burn, it is necessary to classify the burn into a broad category that can then be given the right treatment. Presently the most popular method of classifying burn images is by manual inspection. An automatic system that can classify the burn wound into a broad category is proposed. The system aims to classify burned skin images into 4 broad categories, such as 1st, 2nd, 3rd, and not-a-burn images. 514 images were collected, which consist of 128 images in no burn category and 127 images for each of 1st, 2nd, 3rd degree burns. A comparison of several state-of-the-art classifiers that use deep learning architectures like VGG16, VGG19, ResNet, DenseNet, InceptionNet, and EfficientNet was performed. In addition to that, various optimizers like Adam, AdaGrad, and RMSProp have been compared to get the best results. It was found that ResNet 101 gives the best accuracy of 95% with the AdaGrad optimizer.