Image Semantic Segmentation Algorithm Based on U-Net

The U-Net algorithm, with its unique network structure and excellent performance, has become a classic algorithm in the field of image semantic segmentation. However, there are still some deficiencies at present. Based on the classic U-NET algorithm, this paper improves in two aspects. First, GAN is introduced. Through the adversarial training of the generator and the discriminator, the segmentation results generated by U -Net are closer to the real labels, enhancing the visual effect and accuracy of the segmentation results. Secondly, the introduction of deep residual learning and the use of a simple residual learning module to train deep networks can solve the problems of low learning rate and inability to improve accuracy caused by the deepening of network depth. On the PASCAL VOC 2012 dataset, comparative experiments were conducted between the improved original U-Net and the improved U-Net, and comparative experiments were also carried out between the improved U-Net and the FCN and SegNet baseline models, demonstrating the advantages of the improved algorithm.