A multi-channel multiplexing compressive sensing imaging approach based on compressive sensing is proposed for physical realizable remote sensing systems. First, multi-masks coded with random binary Bernoulli matrix are explored for different optical channels, and the undersampled data of an image are collected in an exposure time. Next, non-local similarity of spatial remote sensing images is presented as the regularization term for reconstruction to remove the reconstructed interference caused by local prominent features in remote sensing scene. The experimental results demonstrate the feasibility of this compressive remote sensing imaging. The proposed algorithm can preserve image details and achieve an effective image reconstruction compared with traditional algorithms.