Abstract
Diabetic Retinopathy (DR) occurs as a result of Diabetes mellitus over an extended period, and it is a microvascular disorder. People with diabetes are more likely to develop diabetes-related complications. According to the World Health Organization, there were 285 million people with diabetes in 2010, and this number will rise to 439 million by 2030. The number of people with DR with vision-threatening disease approximates one-fourth of the total number. Early detection and classification of DR are essential to maintaining the patient's vision. Thus, in this work, we have proposed a multilevel fire hawk convolution neural network (MLFHCNN)-based technique for DR detection. Initially, the retinal fundus images are collected from the dataset and preprocessed using image enhancement and illumination correction. Following preprocessing, the images are segmented using enhanced UNet. After segmentation, the segmented image is classified using the MLFHCNN. The CNN classifier is optimized using Fire Hawk Optimizer (FHO) Optimizer, which allows us to detect both the structure and hyperparameters of CNN simultaneously. For the evaluation of this proposed method, the retinal fundus multi-disease image dataset (IDRiD) is used. Python is used to implement the proposed method.