Abstract

Cloud occlusion significantly hinders remote sensing applications byobstructing surface information and complicating analysis. To address this, wepropose DC4CR (Diffusion Control for Cloud Removal), a novel multimodaldiffusion-based framework for cloud removal in remote sensing imagery. Ourmethod introduces prompt-driven control, allowing selective removal of thin andthick clouds without relying on pre-generated cloud masks, thereby enhancingpreprocessing efficiency and model adaptability. Additionally, we integratelow-rank adaptation for computational efficiency, subject-driven generation forimproved generalization, and grouped learning to enhance performance on smalldatasets. Designed as a plug-and-play module, DC4CR seamlessly integrates intoexisting cloud removal models, providing a scalable and robust solution.Extensive experiments on the RICE and CUHK-CR datasets demonstratestate-of-the-art performance, achieving superior cloud removal across diverseconditions. This work presents a practical and efficient approach for remotesensing image processing with broad real-world applications.