A Network with Composite Loss and Parameter-free Chunking Fusion Block for Super-Resolution MR Image

Figure 1

The proposed architecture of super-resolution. The architecture mainly includes the parameter-free chunking fusion block (PCFB) model, residual block, and reconstruction block. The residual block consists of convolutions, activation functions, and skip connections. PCFB divides the feature map into branches through a split channel, and each branch is fused after passing SimAM to obtain parameter-free attention feature enhancement. The reconstruction block consists of convolution, convolution, PReLU, and the PixelShuffle layer.