Lung Segmentation Network Based on Lightweight and Attention Fusion with U-Net

In order to solve the problem that the traditional U-Net network is bloated, the amount of calculation and the amount of parameters are too large. We fuse the improved depthwise separable convolution of the lightweight network. In order to suppress the influence of background pixels in the image on the segmentation effect, attention gates are introduced to highlight the target area containing key information. We propose a MAU-Net network structure. It is verified by using the LUNA competition lung image dataset. When the calculation amount and parameter amount of the network are reduced by about 78% and 87% respectively compared with the original traditional U-Net. The LOSS of the experimental results on the test set is 0.012, the Accuracy reaches 0.992, and the IOU is 0.961, which is 1.5% higher than the traditional U-Net’s Accuracy, and the evaluation index IOU is improved by 1.9%. It is proved that the improved MAU-Net network has better effect on lung parenchyma segmentation.