A NOVEL DECISION-LEVEL FUSION METHOD FOR SAR AND OPTICAL IMAGES FOR TARGET DETECTION TASKS IN CLOUD-OCCLUDED SCENES

Cloud cover leads to information loss in certain regions of optical remote sensing images (RSI), resulting in missed detection in target detection tasks. Existing research on RSI in cloud occlusion generally focuses on cloud removal tasks, which are inadequate to address the needs of remote sensing object detection tasks. Considering the cloud-penetrating capability of Synthetic Aperture Radar (SAR), this paper proposes a general decision-level fusion detection method leveraging the complementarity of optical and SAR images. An image segmentation method is used to extract dense cloud-covered regions, followed by a decision-level fusion approach based on improved evidence theory to integrate the separate detection results from optical and SAR images, enabling target fusion detection in cloud-occluded scenes. Experiments conducted on a cloud-covered optical-SAR image dataset derived from SpaceNet6 demonstrate that the proposed method achieves outstanding performance in addressing missed detection caused by cloud cover.