Diabetic Retinopathy is a disease caused by the damage of type 2 diabetes on the retina. In this study, two different completely Graphic Processing Unit (GPU) based methods are proposed in which Diabetic Retinopathy lesions were detected independently and automatically from the datasets and the detected lesions were classified. In the first step of the proposed first method, a data pool was created by collecting Diabetic Retinopathy data from different data sets. Then, lesions were detected with Faster Region Based Convolutional Neural Network (Faster RCNN) and the region of interest is marked. In the second stage, the images obtained were classified using the transfer learning and attention mechanism. In order to increase the results obtained in the networks in the first method, a second method has been developed. In the first step of the second method, the unused Region of Interest (ROI) in the image was removed with Compute Unified Device Architecture (CUDA). Then, the lesion areas were marked with Mask Region Based Convolutional Neural Network (Mask RCNN) to cover them completely. With these improvements, the attention layer was removed in the model in the second stage, the network was simplified and training was carried out with direct transfer learning. The methods tested in the Kaggle and MESSIDOR datasets reached 100% ACC (Accuracy) and 99,9% and 100% AUC (Area Under Curve) values with VGG model, respectively. When the results obtained were compared with other results in the literature, it was seen that more successful results were obtained. |