Application of Deep Learning Models in Automatic Diagnostics of Burn Lesions

In this article, the problem of determining the classes of burn types based on the use of deep learning models for the diagnosis of burn types during the treatment of burn injuries is studied. The dataset is a collection of different photographs of burn injuries for computer vision-based assessment of real medicine tasks. On the basis of the research, the tasks of identifying, segmenting and classifying burn injuries due to the use of deep learning models are considered. In particular, two models of convolutional neural networks, the most popular classic U-Net and fast Yolov5, were used in the study. Implementation of methods was carried out on the basis of photo images in jpeg format. The study directly showed that the Unet method can reduce the number of parameters and accelerate the convergence of the model with little loss of accuracy, while Yolov5 showed good performance and confirmed high learning ability. U-Net reduces the sample size by a difference of 0.8 MB and mAP (0.5) drops by 0.7%. The Yolov5 model was effective in increasing the processing speed, and the mAP (0.5) index increased by 10%, that is, it proved that the performance is high. As a result of the study, we propose an effective method for burn damage classification by comparing U-net and YOLOv5 deep learning models to identify, segment, and classify images with burn damage into 1st degree, 2nd degree, and 3rd degree classes.