无人机遥感系统可实现高分辨率成像且成本低、风险小，但成像系统视场小、获取的图像数量多，应采用图像拼接技术进行处理。遥感图像拼接主要存在的问题有系统的实际成像方式为倾斜成像并且图像本身的复杂性对图像配准提出了很大的考验。本文提出一种基于位姿数据校正的遥感红外图像相似性先验拼接。首先，设计了一种拼接成像系统，并利用同步位姿数据校正倾斜成像获取的图像，然后结合网格变形的优化方法，提出了基于相似性先验的图像拼接。实验结果表明，该方法既保证了遥感图像拼接中重叠区域的高精度对齐，又确保了非重叠区域在变换前后的形状相似性，拼接效果出色。

The remote sensing system of an unmanned aerial vehicle(UAV) can achieve high resolution imaging with low cost and risk, but the imaging system has a small field-of-view and receives a large number of images; hence, image stitching should be used for image processing. The main problem with remote sensing image stitching is that the actual imaging method is tilt imaging, and the complexity of the image itself poses a great challenge to image registration. In this work, a similarity prior method of image stitching of remote sensing infrared images is proposed based on the correction of pose data. First, a mosaic imaging system is designed, and the tilt image is corrected by synchronous pose data. Then, an image stitching method based on similarity prior is proposed by combining the optimization method of mesh deformation. The experimental results show that this method ensures not only a high-precision alignment of overlapping regions in remote sensing image mosaic, but also shape similarity of the non-overlapping regions, before and after the transformation, with an excellent mosaic effect.