Application of multi-path network combined with attention mechanism in the diagnosis of diabetes retinopathy

Diabetes retina is one of the most serious causes of blindness. In order to improve the diagnostic accuracy of the severity of diabetes retinopathy, a multipath model was proposed to classify retinopathy. This model designs a multipath network to fully extract features, fuse contextual information, then design a feature fusion module, and combine attention mechanisms to fuse the extracted results to form the final classification features. The classification accuracy of the proposed model reaches 97.51%, effectively classifying different degrees of retinal lesions and preventing further deterioration of the lesions.