Abstract
The most frequent ailment in diabetics that can cause blindness is diabetic retinopathy. It takes time for the clinician to recognize the stage of diabetic retinopathy and report the findings. As a result, the suggested paper takes little time to analyse and is highly helpful in outlining the stages of the condition look for a way to categorise. This essay compares and contrasts various research papers and methodologies. Convolutional and pooling layers can be utilized to boost accuracy, and ReLU can be employed as an activation function, according to the methods we suggested. In this study, fully connected layers which are also useful for classification are proposed. There are numerous indicators of retinal injury, including the fovea, optic nerve head, effusions, blood vessels, hemorrhages, and ocular micro aneurysms. Classification, pre-processing, feature extraction, segmentation, and detection are among the difficulties. The purpose of this research is to use convolutional neural networks to identify diabetic retinopathy (CNN). We compared various CNN architecture models and found that pre-processing can raise model accuracy by up to 5%.