A Physics-Informed Two-Stage Learning-Based Jitter Correction for Spaceborne Time Delay and Integration Imagery

Platform jitter is a common phenomenon of high-resolution optical satellites, which severely degrades the imaging quality both in geometric and radiometric aspects, particularly for time delay and integration (TDI) imagery with submeter ground sample distance. Traditional physics-based jitter correction methods need to establish the jitter model, considering the motion equation and imaging parameters depending on the degradation mechanism, which are complex for practical application. Learning-based methods correct images end-to-end but rely on training datasets, which are commonly simulated by setting supposed jitter and imaging parameters, due to the nonexistence of real-world datasets. However, jitter degradation diverse from different satellites and imaging payloads, which leads to the instability in correction accuracy of the learning-based method. To address the limited practicality of complex physical models and the insufficient adaptability of deep learning methods, we propose a physics-informed learning framework for restoring jitter-induced degradation in satellite TDI images. Different from conventional learning-based approaches that rely on empirically predefined degradation settings, the proposed framework incorporates satellite-specific jitter characteristics estimated from real degraded imagery into the simulation process, enabling more realistic and targeted training data construction. Within this framework, a two-stage network for geometric correction and radiometric restoration is constructed, in which tailored simulation datasets guided by estimated physical parameters are employed to enable stable and effective learning. Experimental results show that the proposed method outperforms existing approaches in both simulated and real scenarios. In addition, global correction of large-scale remote sensing images was achieved, which confirmed the robustness and practicality of the proposed framework for high-resolution space-borne TDI image restoration.