DcCNN network structure for first-layer LH component. The encoding module of DnCNN consists of 17 layers. The first layer begins with 64 3 × 3 convolutional kernels, followed by ReLU activation, and outputs a 525 × 525 × 64 feature map. From the second to the 16th layer, each layer starts with 64 3 × 3 convolutional kernels, followed by batch normalization and then ReLU activation. Specifically, at the fifth and 12th layers, a channel-attention mechanism is introduced after these operations to refine the weighting of the feature maps. Each of these layers outputs a 525 × 525 × 64 feature map. The 17th layer generates a disturbance-estimation map of 525 × 525 × 1 using a single 3 × 3 convolutional kernel. Finally, this disturbance estimation map is subtracted from the original data to obtain clean data with the RFI mitigated, resulting in an output size of 525 × 525 × 1.