Open AccessEpoch interrogation for skin cancer detection using convolutional neural network models
Author(s) -
A. Poornima,
M. Shyamala Devi,
M. G. Sumithra,
Mr. A. Aswin Bharath,
| Swathi,
S. Sathishkumar,
K M Yogesh,
Shubham Upadhyay,
Nitesh Kumar Sah
Publication year - 2021
Publication title -
iop conference series. materials science and engineering
Language(s) - English
Resource type - Journals
eISSN - 1757-899X
pISSN - 1757-8981
DOI - 10.1088/1757-899x/1074/1/012025
Subject(s) - computer science , convolutional neural network , artificial intelligence , rgb color model , feature (linguistics) , pattern recognition (psychology) , kernel (algebra) , image (mathematics) , feature vector , filter (signal processing) , computer vision , mathematics , philosophy , linguistics , combinatorics
Skin cancer is very important notable disease and it is probable to everyone nowadays, it flourishes on the area of body where it exposed to ultraviolet rays. It leads anomalous gain in skin cells. It initiate on various parts of body like face, hand and bottoms of the feet as cautious hole or spot. The initial investigation of anomalous gain is essence to cure the disease at early stage, and it still remains a feasible challenge in the scientific improvements. From the analysis, this paper endeavour to inspect the category of disease with the following improvements. Initially, the skin dataset from ISIC machine archive is utilized for image processing. Secondly, the values of dataset images are normalized by dividing all the RGB values by 255. Thirdly, keras sequential API is used to add one layer at a time, initiating from the input. The CNN can extract the features that are useful for classifying the image, by using the kernel filter matrix. MaxPool reduce the computational cost by down-sampling the image, and the relu activation function is implemented to provide non linearity to the network. The flatten layer is utilized to remodel the final feature maps into 1D vector. CNN model provides accuracy of 94.83% with 3297 images and ResNet 50 model has attained accuracy of 90.78% due to less number of images used for classification. AlexNet model has attained accuracy of 81.8% with 1300 images and GoogleNet V3 inception has attained accuracy of 96% with 3374 images. Finally Vgg16 model has attained accuracy of 97.3% with 5636 samples.