Research On Classification And Recognition Of Skin Cancer Based On ResNet

Human skin is directly exposed to air and sunlight.Because skin is the first line of defense against harmful substances in the human body,it may cause skin canceration due to a variety of reasons(such as the environment,food,and genetic factors).Malignant tumors on the skin are called skin cancers.Different types of skin cancers are defined according to the tumor cells,including epidermis,soft skin tissue,melanocytes,skin lymphoid reticulum and hematopoietic tissue.As far as the current diagnosis is concerned,the death rate of skin cancer is still high,so the development of a ResNet-based skin cancer diagnosis platform can be used for early diagnosis and subsequent treatment at the hospital,which can reduce the death rate of skin cancer.However,the existing technology is still difficult to achieve good results on large-scale data sets.We will use this to discuss,which has great significance in the medical field.The feature detection layer of the convolutional neural network(CNN)learns from the training set samples.Therefore,when using a convolutional neural network(CNN),it is necessary to remove the explicit feature extraction and learn implicitly from the training data.In this article,we outline a single deep convolutional neural network that is applied to skin lesion classification.The network just uses disease labels as input and is trained end-to-end directly in the image.ResNet: Residual Network(hereinafter abbreviated as ResNet).In essence,compared with VGG network,this network uses avg pool instead of full connection,saving a lot of parameters.It learns the residual representation method between input and output using identity mapping.Unlike ordinary CNN networks(such as Alexnet / VGG,etc.),it does not use convolutional layers to directly try to map between input and output.Experiments show that compared with directly learning the mapping between input and output,the former is easier(faster convergence)and effective(you can reachhigher levels by using more layer classification).In order to achieve good results,we adopted the ResNet network architecture for the diagnosis and research of skin cancer degeneration.First,the area of the skin cancer image as the network input was adjusted to a small block suitable for the network.While training the network,the number of channels was adjusted,the number of image layers and the number of training samples are optimal.There may be overfitting problems.The experiments in this article will select and test a variety of data.For data sets that are not a type of experiment,they correspond to the individualized normalization method.Research shows that the diagnostic method in this paper performs better than other methods.