Background-dynamic joint decoupling method for dynamic-scene reconstruction in shift-encoded optically multiplexed imaging

Constrained by detector-array size, spaceborne staring remote sensing systems struggle to achieve both wide coverage and high spatial resolution. Optically multiplexed imaging addresses this trade-off by superimposing multiple fields of view onto a shared focal plane. Shift encoding further changes inter-channel positions across frames, providing priors for demultiplexing; however, most existing methods assume that the scene remains nearly static during the coded sequence. When independently moving targets coexist with channel displacement, time-varying aliasing can cause target loss, smearing, and cross-channel leakage. We propose a background-dynamic joint decoupling and moving-target sorting method for shift-encoded dual-FOV remote sensing imaging. Under known coding trajectories, the method jointly models background consistency and target spatiotemporal characteristics, enabling dual-channel background recovery, dynamic-target preservation, and source-channel attribution. Experiments on 15 controlled simulated video sequences show that the proposed method achieves an average PSNR of 40.892 dB and an average SSIM of 0.9867, with a target recovery rate above 97% under nominal noise. A hardware-in-the-loop dual-channel optically multiplexed experiment further evaluates the model under beam combination, FSM-induced displacement modulation, detector acquisition, and laboratory imaging nonidealities.