To improve the diagnostic efficiency of diabetic retinopathy(DR),we proposed a DR classification model based on the VOLO network.First,the depthwise convolution module was added to Outlook attention to improve the expression ability of attention.Then the normalization-based attention module was embedded in the backend of the attention mechanism to highlight the salient fea-tures,thereby completing the DR classification of lesion severity.The experimental results showed that the classification accuracy of the model reached 89.10％.This research model has good feasibility and can better assist doctors in clinical precision treatment.
维普
万方数据