Segmentation of Retinal Vessels Based on DenseNet-Attention-Unet Model Network

Abstract

The feature information of retinal vascular image is complex. There are some problems in the existing algorithms, such as poor segmentation effect of microvascular segmentation and pathological vascular misclassification. Therefore, a vascular segmentation model based on DenseNet-Attention-unet (DA-Unet) is proposed. Firstly, retinal blood vessels were enhanced by adaptive histogram equalization and gamma correction, and then the DA-Unet model was constructed by DenseNet, Convolutional Block Attention Module and U-Net for retinal blood vessel segmentation.The experimental results show that the average accuracy of retinal vascular segmentation is 97.01%, and the ROC is 98.65%, when it was tested on the DRIVE data set.