Research on Super Resolution Image Reconstruction Method Based on Generative Adversarial Networks
Based on the ESRGAN network,this paper introduces the deep multiscale convolution(DMCONV)module into the generative network encoder,and integrates the innovation channel attention(ICA)and capsule network(CapsNet)multidimensional neurons into the generative network decoder.A kind of AC-ESRGAN image super-resolution network based on multi-scale convolution,attention mechanism and vector neurons was constructed,and the training,testing and ablation experiments were conducted on BSD100,Manga109,Set14 and other data sets.The experiment shows that the network can extract the deeper features of the image better,realize the local cross-channel interaction of the network,and enhance the ability of expression and understanding of the image.
NETL
NSTL
万方数据
