Automatic Detection and Categorization of Skin Lesions for Early Diagnosis of Skin Cancer Using YOLO-v3 - DCNN Architecture

Abstract

Malignant melanoma is a type of benign skin cancer that is the most lethal due to its rapid development and affects a large number of people worldwide. Also, it is one of the deadliest diseases in the world. Moreover, existing research has stated that risk factors may be significantly decreased by making it nearly curable if diagnosed early on. This prompt identification and categorization necessitate using an automated system, even though the existing method is rather difficult. Hence our research employs the YOLO v3 - DCNN architecture to discover and categorize the deadliest kinds of skin cancer. Initially, YOLO v3 generates the feature map; simultaneously, colour features are extracted using colour moments with QuadHistogram, whereas Grey Level Co-occurrence Matrix (GLCM) with Redundant Contourlet Transform(RCT) generated texture features, and both (colour and texture) features get fused. Then, fused features are fed into the Deep Convolutional Neural Network (DCNN), which classifies the different types of skin cancer. Finally, our proposed approach is compared with the current works. Consequently, our proposed YOLO-v3 –DCNN has greater accuracy when contrasted with the baseline techniques.