Research On Rosacea Classification Method Based On Computer Vision

The accurate classification of rosacea is the premise of effective treatment.It is an important research direction to use computer vision technology to classify rosacea.In the diagnosis of rosacea,traditional methods rely heavily on dermatologists’ clinical experience and subjective judgment,which is easy to cause misdiagnosis.Therefore,different classification algorithms in the field of computer vision are considered in this paper.Based on the collected facial images of rosacea patients,the classification of rosacea is studied as follows.First of all,aiming at the problems of blurred images,repeated images and indistinct features in rosacea data set,a series of preprocessing such as data cleaning and data specification were carried out on the dataset.In order to solve the problem of small amount of data,data augmentation technology is used to increase the training set data by two times.Using the powerful feature extraction function of convolutional neural network,residual network models with different depths are designed for experiments.The experimental results show that the 18-layer residual networks(Resnet18)model has better classification performance in the current rosacea dataset.Secondly,a semantic segmentation model Resnet18-Unet,which combines Resnet18 model and Unet model,is designed according to the characteristics of relatively fixed location and relatively concentrated area of rosacea.According to the task requirements,a series of preprocessing of rosacea dataset is carried out,and semantic tags were manually labeled for the dataset.Aiming at the problem of slow training speed and difficult convergence of semantic segmentation model,the idea of transfer learning is adopted,and the pre-trained Resnet18 model is used as the initial weight to accelerate the convergence of the model.Experimental results show that Resnet18-Unet model has faster convergence speed and better segmentation performance than Unet model.Finally,according to the fuzzy edge contour and the unclear semantic features of rosacea,a object detection algorithm is proposed to classify and locate rosacea.According to the characteristics of the task,the rosacea dataset was preprocessed,and the rosacea object detection dataset was manually labeled.Experiments based on YOLOv3 object detection algorithm show that the model can detect rosacea quickly and accurately.