Frequency domain decomposition network for optical remote sensing image destriping
Yu Shi, Feiyan Wu, Yaozong Zhang, Lianying Zou · Applied Optics · 2025
Due to imaging limitations in optical systems, optical remote sensing images are affected by periodic or non-periodic stripe noise, which interferes with subsequent target detection and recognition. Features are more intuitively expressed in the frequency domain than in the spatial domain. To fully utilize the spatial and frequency domain features, we propose an optical remote sensing image destriping network based on frequency domain decomposition. First, wavelet decomposition is applied to help the network distinguish stripe and background features, and singular value decomposition is used to extract stripe information from high-frequency components. Second, a spatial-frequency coupling block is constructed to exploit the interaction of the spatial and frequency domains. Additionally, a multi-scale adaptive fusion block is designed to enhance the feature information transmission. Simulated and real experiments demonstrate that the proposed method outperforms state-of-the-art methods in stripe noise removal and detail preservation.