A Comprehensive Study of DCNN Algorithms Based Transfer Learning for Human Eye Cataract Detection

Abstract

Cataract, a common eye disorder characterized by clouding of the lens, is a leading cause of vision loss. Each year numerous numbers of people are falling sufferer to visionary loss globally. The majority of the time, this problem arises as people age. This problem also can occur in young age people because of injury or certain clinical situations. It's far known as cataract while a dense and cloudy layer create on the eye lens and consequences the clear vision which can cause a problem like blurry eyesight, diminished vision and prescient. Additionally, they face difficulty seeing in robust light and gradually it could be the reason for full blindness. An excellent way to manipulate the hazard and avoid blindness is to stumble on cataracts well-timed and correctly before it become more complicated. In this study, we propose a cataract detection system using deep learning and image processing techniques. Our system aims to automatically analyze ocular images and predict the presence of cataracts with high accuracy. We are trying to pick out an efficient and accurate manner of detecting cataracts primarily based on a Deep Convolutional Neural network (DCNN) with the publicly accessed dataset. We used the transfer learning methods with DCNN models which are VGG19, NASnet, Resnet50 and MobileNetV2 achieving the highest accuracy across 2000 image sets. Also, MobilNnetV2 achieved accuracy rates of 97.75% on the test images. Compared to other models, the final result indicates that MobileNetV2 takes the least time to recognize images and classify them.