面向遥感影像解译的观测小目标超分辨率重建方法  

A super-resolution reconstruction method for observing small targets in remote sensing image interpretation

作　　者：冯晓亮[1] 陈欢[1] 李厚芝[1] FENG Xiaoliang;CHEN Huan;LI Houzhi(Institute of Exploration Technology,Chinese Academy of Geological Sciences,Chengdu 611734,China)

机构地区：[1]中国地质科学院探矿工艺研究所,成都611734

出　　处：《测绘工程》2025年第2期32-37,共6页Engineering of Surveying and Mapping

摘　　要：为解决遥感图像受环境复杂、光照变换等原因造成的小目标遥感影像解译难的问题,提出面向遥感影像解译的观测小目标超分辨率重建方法。使用由特征融合网络加特征细化网络组合成的边缘感知网络,采集并提取观测小目标遥感图像,确定其位置和范围,作为生成对抗网络的输入值;生成器持续生成真实的小目标遥感影像分布,并发送给判别器进行鉴别,经生成和鉴别过程的不断迭代训练,直至判别器无法判断遥感影像真伪后,则输出小目标遥感影像超分辨率重建结果。实验结果表明,该方法能够准确地获取遥感影像中的小目标位置和范围,识别出人眼无法确定的小目标;能够十分清晰的将遥感影像中的小目标进行重建,超分辨率重建后的影像亮度、分辨率有了明显提升;且选取5处地点进行遥感影像的超分辨率重建后,得到的均方根误差、峰值信噪比和通用质量评价指数的平均值分别为29.43,51.01和0.92,可有效助力遥感影像解译。To solve the problem of difficult interpretation of small targets in remote sensing images due to complex environments and lighting transformations,a super-resolution reconstruction method for observing small targets in remote sensing image interpretation is proposed.Using an edge perception network composed of a feature fusion network and a feature refinement network,the collected observed small target remote sensing images are processed,small targets are extracted from the remote sensing images,their positions and ranges are determined,and used as input for generating adversarial networks.The small target remote sensing images are continuously generated by the generator and sent to the discriminator for identification.Through continuous iterative training of the generation and identification processes,Until the discriminator is unable to determine the authenticity of the remote sensing image,the super-resolution reconstruction result of the small target remote sensing image is output.The experimental results show that the proposed method can accurately obtain the position and range of small targets in remote sensing images,and identify small targets that can not be determined by human eyes.The small targets in the remote sensing images can be reconstructed clearly,and the brightness and resolution of the super-resolution reconstructed images have been significantly improved.After super-resolution reconstruction of remote sensing images in five locations,the average values of root mean square error,peak signal-to-noise ratio and general quality evaluation index are 29.43,51.01 and 0.92,respectively,which can effectively assist in remote sensing image interpretation.

关 键 词：遥感图像 小目标 超分辨率重建 边缘感知网络 生成对抗网络 图像评价 

分 类 号：P237[天文地球—摄影测量与遥感]