Frequency-Domain Modeling and Removal of Platform Jitter Stripes in GF‑7 DSMs

Platform jitter of stereo mapping satellites induces strip‑like periodic elevation artifacts in digital surface models (DSMs), which severely degrades the geometric quality of stereo mapping products. Existing jitter detection methods mostly rely on disparity fields or high‑frequency attitude data. When disparity maps are unavailable or the jitter frequency is higher than the sampling rate of the attitude sensors, these methods become difficult to apply. Focusing on jitter in GF‑7 DSMs, this paper proposes a frequency‑domain jitter detection and stripe removal framework based solely on DSM data. The framework consists of two main components. (1) In the 2D spectral narrow‑band notch destriping method, the 2D Fourier power spectrum of a high‑pass DSM (after removing the low‑frequency terrain trend) is computed. Within a fixed annular range of periods, the prominence of peaks in the directional energy spectrum and radial period spectrum is analyzed to robustly estimate the dominant stripe orientation and period. A smooth Gaussian narrow‑band stop filter is then constructed in the full‑image frequency domain to concentrate suppression on the narrow‑band energy around the jitter fundamental frequency. (2) In the profile‑template destriping method, 1D median profiles in the direction perpendicular to the stripes are used as observations. Under the prior constraints of stripe direction and period obtained from the 2D spectral method, an infinite impulse response (IIR) notch filter is designed to extract candidate stripe components. A 2D stripe model is then constructed by combining a global 1D stripe template with a slowly varying 2D amplitude field, and the stripe component is explicitly modeled and removed in the spatial domain. Experiments on three GF‑7 DSM scenes over the plains near Shangqiu, Henan Province, show that the 2D spectral narrow‑band notch method robustly estimates a stripe orientation of northeast–southwest, which is highly consistent with the subsatellite ground‑track direction. The dominant period is about 90 m. The profile‑template method yields a spatial period of about 90 m as well, and the consistency of the dominant period between the two methods exceeds 99.7%. Using the resulting narrow‑band stop filter and stripe template, the energy suppression ratio η_PSD of the stripe narrow‑band in the power spectrum exceeds 92% for all DSMs, while the overall spectral shape of the DSM remains nearly unchanged, and the striping artifacts are significantly reduced or eliminated. The results demonstrate that the proposed DSM frequency‑domain jitter detection and stripe modeling approach achieves high parameter stability and destriping performance even in the absence of attitude and disparity information.