基于生成对抗网络的湍流退化图像校正方法研究  

Research on Turbulence Degradation Image Correction Method Based on Generative Adversarial Network

作　　者：刘嘉萌 LIU Jiameng(Suzhou College of Information Technology,Suzhou,Jiangsu 215200,China)

机构地区：[1]苏州信息职业技术学院,江苏苏州215200

出　　处：《移动信息》2024年第8期270-272,共3页MOBILE INFORMATION

摘　　要：利用无线光通信技术传输的图像信息易受湍流影响出现退化失真现象。针对传统算法校正湍流图像效果不佳等问题,文中根据湍流效应特性创建了湍流退化图像模型,同时提出了一种基于生成对抗策略的UP-GAN网络模型。该模型融入了目标检测中的特征图金字塔结构,并增加了特征融合模块,进一步提升了模型校正的能力。实验结果表明,该方法对多种湍流强度下造成的退化图像有着更高的校正精度。The image information transmitted by wireless optical communication technology is prone to degradation and distortion under the influence of turbulence.Aiming at the poor effect of traditional algorithms in correcting turbulent images,a turbulent degradation image model is created according to the characteristics of turbulent effects,and a UP-GAN network model based on generative adversarial strategy is proposed.The model integrates the pyramid structure of feature map in object detection and adds a feature fusion module to further improve the ability of model correction.Experimental results show that the method has higher correction accuracy for degraded images caused by various turbulent intensities.

关 键 词：生成对抗 湍流退化 图像校正 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]