Detection and Classification of Skin Disease Using Modified Mobilenet Architecture

Abstract

Abstract  

Skin is outer covering of body that separates human body from outer environment. Skin acts as protection from the UV rays of sun. Nowadays, skin disease cases are increasing but it can be curated if disease is detected at an early stage [1]. Therefore, there is a need of a system that can detect skin disease at an early stage. Deep Learning has attained success for its outstanding performance in the medical field [2-4]. Deep learning is the most efficient, supervised, time- and cost-effective method. These algorithms are fast, inspite of this, they are also used in different sectors and shows their success and adaptability. In this paper, a transfer learning based Mobilenet architecture has been modified by removing the last five layers and adding one average pooling layer, one dropout layer and one dense layer. The modified Mobilenet architecture has been simulated on HAM10000 skin disease dermoscopy dataset. Different transformation techniques have been applied for data augmentation to solve the problem of data imbalance. The model has attained an accuracy of 90% followed by precision of 86.14%.