DIAGNOSIS AND SEPARATION OF SKIN DISEASE IMAGES USING CONVOLUTIONAL NEURAL NETWORK AND VOTING METHOD
Keywords:
Convolutional Neural Network (CNN), skin images, deep learning, ResNet architecture, VotingAbstract
Misrecognition of skin images is a common occurrence worldwide. Misdiagnosis of skin diseases that are very similar causes many problems. Misdiagnosis has problems for doctors such as their similarities and complications. Diagnosis methods using convolutional neural network based on deep learning are getting much attention nowadays.
These methods have been able to show their ability to recognize and distinguish images well. Therefore, in this article, it has been tried to distinguish images with acceptable accuracy and high accuracy by using convolutional neural network based on deep learning.
In this article, first the deep features of the images were extracted using the convolutional neural network without manual intervention, and finally skin images with various architectures of convolutional networks including AlexNet, ResNet, VggNet, MobileNet, DarkNet and GoogleNet were checked, and the accuracy of the ResNet network was compared to other Architectures were distinguished with a higher accuracy of 99.1% and a sensitivity of 98.9%. In the final part, using the Voting method, the accuracy of the total result reached 99.93%. This study shows that the proposed method differentiates skin disease with acceptable accuracy.
