基于条件生成对抗网络的空气预热器内红外补光监测视频图像清晰化方法  

Method for sharpening infrared compensation image for monitoring video inside air preheater based on cGAN network

作　　者：刘君 邓毅[2] 杨延西[2] 魏永贵 薛燕辉 史雯雯 LIU Jun;DEND Yi;YANG Yanxi;WEI Yonggui;XUE Yanhui;SHI Wenwen(Dongfang Electric Corporation Dongfang Boiler Group Co.,Ltd.,Chengdu 611731,China;School of Automation and Information Engineering,Xi’an University of Technology,Xi’an 710048,China)

机构地区：[1]东方电气集团东方锅炉股份有限公司,四川成都611731 [2]西安理工大学自动化与信息工程学院,陕西西安710048

出　　处：《热力发电》2021年第10期130-134,共5页Thermal Power Generation

基　　金：国家重点研发计划项目(2018YFB1703000);陕西省现代装备绿色制造协同创新中心项目(304-210891702)。

摘　　要：火电厂空气预热器(空预器)内部的灰尘、烟雾、光照变化等因素导致监控视频画面不清晰,影响监控效果。鉴于此,本文提出一种基于条件生成对抗网络(cGAN)的恶劣工业环境下红外补光监控视频图像清晰化方法。针对获取的红外补光图像样本数据进行预处理,包括高斯滤波去噪以及图像拼接操作,得到低清晰度图像和高清晰度图像的合成图像,低清晰图像作为待重建图像,高清晰图像作为重建图像的理想参考图像,采用建立的cGAN模型对低清晰图像进行重建,调节优化参数生成高清晰图像。试验采用空预器现场监控视频作为训练集对网络模型进行离线训练,实现了空预器红外补光监控图像清晰化处理。本文方法cGAN模型小、训练过程简单、计算效率高、图像清晰化处理效果好,适于相似复杂工业环境下对监控视频图像的恢复和清晰化处理。Changes of factors such as dust,smoke and light inside air preheater cause unclear monitoring video images,which affects the monitoring effect.To solve this problem,this paper proposes a method for sharpening infrared compensation image of monitoring video inside the air preheater based on conditional generative adversarial network(cGAN).In this method,the acquired infrared compensation images are preprocessed,including Gaussian filter de-noising and image splicing,then the composite graphics of low resolution image and high resolution image are obtained.The low resolution image serves as the one to be reconstructed and the high resolution image as the ideal reference one for reconstructing.The cGAN model is established to reconstruct the low-resolution images,and the optimized parameters are adjusted to generate the corresponding high-resolution images.Moreover,two sets of on-site monitoring videos inside the air preheater are adopted in the experiments as the training sets for offline training,and clear processing of infrared compensation images is realized.The method in this paper has small model,simple training process,high computational efficiency and good image clarity processing effect,which is suitable for restoration and clarification of the monitoring image in similar complex engineering scenarios.

关 键 词：空气预热器 监控视频 图像清晰化 红外补光 条件生成对抗网络 图像重建 

分 类 号：TK311[动力工程及工程热物理—热能工程]