基于各向异性滤波的无人机红外图像超分辨率重建方法  

Super-resolution Reconstruction of Thermal Images by UAV Based on Anisotropic Filtering

作　　者：高阳 杨磊 刘枫琪 李宋玉 GAO Yang;YANG Lei;LIU Fengqi;LI Songyu(Henan Jiuyu Enpai Electric Power Technology Co.LTD.,Zhengzhou 450000,China)

机构地区：[1]河南九域恩湃电力技术有限公司,河南郑州450000

出　　处：《红外技术》2024年第8期933-939,共7页Infrared Technology

摘　　要：为提高无人机红外图像分辨率,研究基于各向异性滤波的无人机红外图像超分辨率重建方法。基于人眼视觉特性自适应设定图像内边缘区域与平滑区域梯度阈值,在此设定下通过各向异性扩散滤波方法去除初始红外图像内平滑区域噪声,并保留与增强图像内边缘区域细节信息,获得去噪后红外图像,向生成对抗网络内输入此图像,经其中生成模型与判别模型的博弈学习,获得最终超分辨率重建红外图像输出。结果显示,该方法具有较高的图像边缘细节保护能力与背景噪声去除能力,综合消噪效果较好,可获得分辨率高、视觉呈现清晰的超分辨率重建红外图像,保障无人机巡检中红外图像的辨识精度与效率。The super-resolution reconstruction method of UAV thermal infrared images based on anisotropic filtering was studied to improve UAV thermal infrared images resolution and provide efficient and accurate identification of UAV inspection images.The adaptive setting was based on the image edge area and smooth area gradient threshold,under the set by the anisotropic diffusion filter method.We aimed to remove the initial thermal infrared image noise smoothing area,maintain details,enhance the image edges,and obtain a thermal infrared image after denoising.To generate an image against the network input,the model and discriminant model were generated by game learning for super-resolution reconstruction of the thermal infrared image output.This method has high image edge detail protection,the ability to remove background noise,and an improved comprehensive denoising effect.Furthermore,it can obtain high-resolution and visual presentation of super-resolution reconstruction of thermal infrared images,as well as ensure the identification accuracy and efficiency of thermal infrared images in UAV inspection.

关 键 词：各向异性滤波 无人机 红外图像 超分辨率重建 梯度阈值 生成对抗网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]