Résumé: Diabetes is a chronic disease that affects people all over the world and raises the glucose level in the blood as a result of a lack of insulin. Diabetic Retinopathy causes retinal eye disease, which impairs vision and eventually results in blindness. The two classifications of Diabetic Retinopathy based on retinal indicators are Non-Proliferative and Proliferative Diabetic Retinopathy. The Diabetic Retinopathy diagnosis is a time-consuming process for professionals. But the use of handcrafted features limits this method's performance. To identify Diabetic Retinopathy at an early stage, we propose an Egret Swarm optimized hybrid Mask Region-Based Convolutional Neural Network-Bidirectional Gated Recurrent Unit approach which identifies the interdependencies between different Diabetic Retinopathy stages. Initially, the input samples are preprocessed using data augmentation and partitioned into training and testing data. The parameter of the Hybrid Mask Region-Based Convolutional Neural Network-Bidirectional Gated Recurrent Unit model is optimized through the Egret Swarm Optimization algorithm to minimize the loss of the classifier. Here, Egret Swarm optimized hybrid Hybrid Mask Region-Based Convolutional Neural Network-Bidirectional Gated Recurrent Unit architecture is utilized to classify various Diabetic Retinopathy stages. This paper uses three large baseline datasets: Idrid, APTOS 2019 blindness detection, and Zenodo dataset. The simulation results demonstrated that the proposed technique achieved improved precision, recall, and F-measure scores which are nearly equal to 99.1%, 98.9%, and 99%.

Langue: Anglais