Cloud Removal in Remote Sensing Using Sequential-Based Diffusion Models

The majority of the optical observations collected via spaceborne satellites are corrupted by clouds or haze, restraining further applications Earth observation; thus, exploring an ideal method for cloud removal is great concern. In this paper, we propose a novel probabilistic generative model named sequential-based diffusion models (SeqDMs) cloud-removal task in remote sensing domain. proposed consists multi-modal (MmDMs) and training inference strategy (SeqTIS). particular, MmDMs that reconstructs reverse process denosing (DDPMs) to integrate additional information from auxiliary modalities (e.g., synthetic aperture radar robust corruption clouds) help distribution learning main modality (i.e., satellite imagery). order consider across time, SeqTIS designed temporal arbitrary length both input sequences without retraining again. With SeqTIS, SeqDMs have flexibility handle sequences, producing significant improvements only with one two samples greatly reducing time cost retraining. We evaluate our on public real-world dataset SEN12MS-CR-TS multi-temporal task. Our extensive experiments ablation studies demonstrate superiority quality reconstructed over multiple state-of-the-art approaches.