Accurate diabetic retinopathy segmentation and classification model using gated recurrent unit with residual attention network

•To construct Diabetic Retinopathy detection by segmentation and classification.•To segment the essential regions by Adaptive R2Unet++ (AR2Unet++) mechanism.•To implement RAN-GRU to offer accurate DR-classified outcomes.•To construct ESSA for tuning the parameters to increase accuracy of detection.•To compute the efficiency by considering various performance measures. Diabetic Retinopathy (DR) is a preventable purblindness caused between aged diabetic-affected individuals. Basic discovery and grading of the DR is an important process to provide accurate treatment for saving the individual’s life from vision loss. Yet, manual validations executed in the fundus images consume more time, and also the DR screening procedures are complicated. Presently, more automated techniques are designed to execute the DR grading with joint detection of various signs. But, several problems are attained due to the scalability of data which act an efficient part in improving the efficiency of the network. Thus, resolving these kinds of issues is important to increase the accessibility and precision of the DR detection schemes. So, in this work, an innovative DR detection through deep learning is suggested to deal with the issues involved in the conventional models. At first, the segmentation is carried out using the Adaptive R2Unet++ (AR2Unet++), where several parameters are optimized using the Enhanced Sparrow Search Algorithm (ESSA). Then, the segmented images are fed to the classification is performed by the Residual Attention Network with Gated Recurrent Unit (RAN-GRU). Finally, the developed RAN-GRU-based classification framework provided the classified outcome. At last various validation processes are performed in the suggested mechanism to compute its efficiency.