UNet-based architecture used for the temporal reconstruction of solar images, where n represents the input length of the solar image sequence. Each box corresponds to a multichannel feature map. The gray boxes are copied maps. The number of channels is shown at the top of the box. The resolution in pixels is indicated at the side of the box. The arrows represent operations. A similar architecture is also used in spatial super-resolution, but with only three pairs of double convolutional blocks, and a dropout rate of 0.25 is introduced at each convolutional layer.