Classification of Diabetic Retinopathy Grade Based on G-ENet Convolutional Neural Network Model Classification of Diabetic Retinopathy Grade Convolutional Neural Networks are Used to Solve the Problem of Diabetic Retinopathy Grade Classification

Diabetic retinopathy (DR) is a diabetic complication with a high incidence and blindness. According to the existing deep learning model, the identification rate of similar lesion points in DR is low, resulting in low classification accuracy. This paper first designs a new attention mechanism called Global Channel Attention (GCA) mechanism, and then combines the GCA attention mechanism with the classical deep learning model EfficientNetB0, and proposes the convolutional neural network model GCA-EfficientNet (G-ENet) used for DR grade classification. This paper adopts a transfer learning strategy to train the Kaggle-DR dataset. Finally, the accuracy, precision, recall, specificity, and F1-score of the G-ENet model on the verification set reached 96.0%, 96.1%, 95.9%, 99.0%, and 96.0%, respectively. The classification performance was significantly better than that of classical models such as ResNet-50, GoogLeNet, DenseNet-121 and VGG16. The experimental results show that the algorithm proposed in this paper has good results in the classification task of diabetic retinopathy grade, which can provide help for early screening and diagnosis of diabetes.