To solve the problem that existing algorithms cannot remove the rain pattern thoroughly and may cause background information loss，an image deraining algorithm based on Dual Attention Dense residual Contraction (DADC) network is proposed. In the network，various scale information is collected through a mixed feature compensation module. At the encoding stage，the DADC module is taken as the basic encoding module of the encoder，and as for the collected feature information，the useless information is zeroed out by using the soft threshold network，and dual-attention of spatial and channel is added to annotate the location information of the rain pattern. At the decoding stage，the feature information at different stages is aggregated，the spatial and channel excitation is conducted through the scSE attention mechanism，and the feature information is compressed and then passed into the decoder for decoding and outputting the rain removal image. The experiments are conducted on the publicly-available datasets of Rain100H，Rain100L，Rain800 and Rain12，and the Peak Signal-to-Noise Ratio (PSNR) is improved by 1.07～7.45 dB，and the structural similarity is improved by 0.021～0.139 on Rain100H compared with those of other algorithms.