To improve the performance of skin lesion segmentation,a version of U-Net network based on convolutional neural network was proposed combined with attention and context information.Using Resnet-34 network as the encoder,coordinate at-tention mechanism was added in the process of skip connections,and more accurate target area was located by capturing accurate position information.The context information module was designed to strengthen the learning ability of the foreground features and the high-efficiency channel attention module was added to recalibrate the weight and obtain a higher quality segmentation map.The designed model was verified on the public dataset ISIC 2017.Experimental results show that the recall rate and F1-score reach 85.29％and 87.03％,respectively.The proposed method outperforms the existing methods in terms of accuracy,recall rate(Recall),intersection over union(IOU),F1-score,and yields competitive results.
维普
万方数据