遥感图像双模态融合去云方法

打开文本图片集

关键词：遥感图像；合成孔径雷达图像；光学图像；多模态；数据融合；云去除中图分类号：TP751 文献标识码：Adoi：10.37188/OPE.20253318.2996 CSTR：32169.14.OPE.20253318.2996

Abstract：In order to solve the problem thatthe diference between cloud and background could not be accurately handled in the process of cloud removal from Synthetic Aperture Radar （SAR） images combined with optical image data，this paper proposed a cloud removal network based on the combination of SAR images and optical remote sensing images with a double-branched structure. In the feature extraction stage，a multi-scale atention mechanism was introduced to effectively capture the global and local information in the image，and provide a more representative representation of feature information for subsequent fusion and cloud removal operations.The paralel structure of the local fusion branch and the diferential branch was redesigned，and the contribution of the two was dynamically balanced through the gating mechanism，so as to fully explore the complementarity of the two，and refine the cloud edge in the optical cloudy image，so as to restore a more fine cloudless optical image. The location-aware enhanced Swin Transformer was used to densely connect local features，so that the network model had beter robustness in the face of complex environments. The Peak Signal-to-Noise Ratio （PSNR） and Structural Similarity Index（SSIM）of the proposed model areO.8331dB and O.O24 6 higher than those of the curent optimal algorithm，respectively. It is proved that the proposed algorithm has bettr performance than other methods in the image removal task.

Key words： remotely sensed imagery； synthetic aperture radar image； optical images；multimodal; data fusion；cloud removal

1引言

卫星遥感技术已成为获取地理信息数据的重要手段[1]，但云层的存在会导致数据失真，增加了对遥感图像分析解译的难度，进而限制了光学遥感图像的使用价值与发展潜力[2-4]，所以遥感图像去云是地球观测领域的核心挑战。（剩余17274字）