摘要 针对可见光红外遥感图像转换过程中由于复杂背景、光线干扰等因素引起的纹理和结构信息差的问题，提出了一种基于全局多尺度反射Transformer和局部边缘感知Transformer生成对抗网络(LGGAN-Transformer)的遥感图像可见光红外转换算法。首先设计物理反射系数提取模块提取短波红外图像特征。然后提出具有多层次共享策略的多尺度反射Transformer和边缘感知Transformer分别提取图像的全局特征和局部特征。最后将边缘引导模块嵌入到判别器中,提高生成器的边缘识别能力。在哨兵二号数据集和自建数据集上进行实验，MS-SSIM分别提高4.27%和5.71%。实验结果表明,本文方法在短波红外图像生成任务中有着较高的生成图片质量和较清晰的局部纹理。

Abstract Aiming at the problem of poor texture and structure information caused by complex background and light interference in the process of visible infrared remote sensing image conversion, a visible infrared conversion algorithm of remote sensing image based on global multi-scale reflection Transformer and local edge-aware Transformer generative adversarial network ( LGGAN-Transformer ) is proposed. Firstly, the physical reflection coefficient extraction module is designed to extract the features of shortwave infrared images. Then, a multi-scale reflection Transformer and an edge-aware Transformer with a multi-level sharing strategy are proposed to extract the global and local features of the image respectively. Finally, the edge guidance module is embedded into the discriminator to improve the edge recognition ability of the generator. Experiments on Sentinel-2 data set and self-built data set show that MS-SSIM increases by 4.27 % and 5.71 % respectively. The experimental results show that the proposed method has higher image quality and clearer local texture in the shortwave infrared image generation task.