A MEMORY-EFFICIENT DEEP CNN APPROACH FOR RETINAL DISEASE DETECTION

Abstract

The project focuses on the development of a convolutional neural network (CNN) model for the detection and classification of retinal diseases using minimal memory consumption. The proposed model aims to address the high memory and CPU consumption issues associated with the U-Net Segmentation technique, which is commonly used for retinal disease classification. The model is evaluated on a standard benchmark dataset called EyeNet, which consists of 32 classes of retinal diseases. Experimental evaluation shows that the proposed model achieves better performance in terms of memory management and accuracy compared to existing techniques. The evaluation is based on precision, recall, and accuracy metrics, considering different numbers of epochs and time consumption
for each step. The proposed technique achieves an high accuracy of on the EyeNet dataset, demonstrating its effectiveness in multi-class retinal disease classification. And also we incorporated mobilenet, densenet and hybrid approach for enhancing the accuracy in which MobileNet achieved 97% accuracy, Xception achieved 100%, and A hybrid. A Flask-based front-end with authentication is developed for streamlined and secure user interactions.