Detection of Different Degrees of Skin Burn using YOLOv3

Abstract

A skin burn is an injury, cellular tissue damage and protein denaturation caused by heat, electricity, chemicals or sunlight. These skin burns are defined by the area they cover and how deep they are. The burn degree is determined by the number of skin layers affected. A large injury may have areas with different depths. In this paper we use convolutional neural network, YOLOv3 to detect different degrees of skin burn. Experimental result shows that our state-of-the-art model, YOLOv3 yielded an incremental accuracy of 86% which is trained on approximately 1400 images compared to SVM and MDS based models and the tested with 24 test images. Also, in webcam live capturing we obtained an average frame per second of 16.7 with NVIDIA 1050.

Keywords:Skin Burn; Degree of Burn; Convolutional Neural Network; YOLOv3