视网膜图像的血管分割方法OA

Color retinal image segmentation is a computer-assisted medical method.This paper proposes an improved Unet image segmentation method to address the issue of low segmentation accuracy in retinal vascular images.Replace the output layer of the network from the full convolutional layer to the PSP pyramid structure layer,replace the activation function with ELU,and add a dropout mechanism.Before the image is transmitted to the network,data enhancement processing is randomly performed on the image with a certain probability.The test results show that compared with Unet,the improved algorithm has a 1.56%improvement in mAcc performance indicators and a 2.33%improvement in mdice performance indicators.