A contemporary method on feature selection and classification using multi-model deep learning technique for identifying diabetic retinopathy

Abstract

Diabetic retinopathy is one of the leading reason for preventable blindness in the world. 10-18 % of diabetic people having diabetic retinopathy. The feature selection and classification is a vital task to find the seriousness of the diabetic retinopathy. The different researchers introduced different techniques to extract the features and classification of diabetic retinopathy images. The deep learning is one of the essential methods to extract the features. Most of the previous techniques are extracted information's with the help of texture and extracted the whole image feature data. Some feature missed and thereby the accuracy is significantly less. Hence a proposed new technique called FRCNN (Fast Region-based Convolution Network) and Nearest Neighbour (NN) algorithm used to extract the features and classifications. The proposed method yields better accuracy (96%), sensitivity (98%) and specificity (97%) compared to the previous methods. The implementations Messidor Dataset is used for training and testing