학술논문
基于深度学习的红外图像超分辨率重建方法 / Super-resolution reconstruction method of infrared image based on deep learning
Document Type
Academic Journal
Author
李岩; 杨得成; 于光华; 高爽; 刘禹彤; 翟茁; 张宝金; LI Yan; YANG Decheng; YU Guanghua; GAO Shuang; LIU Yutong; ZHAI Zhuo; ZHANG Baojin
Source
激光杂志 / Laser Journal. 45(1):142-147
Subject
Language
Chinese
ISSN
0253-2743
Abstract
为了提高红外图像的超分辨率重建效果,提出基于深度学习的红外图像超分辨率重建方法.利用红外图像的反射特性与红外辐射特性建立红外图像的显著性区域检测模型;通过可见光与近红外图像之间样貌差异度水平检测图像的边缘轮廓特征,提取可见光与近红外光融合性特征参数;根据融合层次不同对图像信号级、像素级、特征级、决策级四个维度进行重建,提取图像的边缘、形状、纹理特征;根据特征分布的噪声水平与配准质量,采用深度学习算法实现对红外图像超分辨率重建.仿真测试结果得出,该方法进行红外图像重建的显著性特征检测能力较强,重建后将图像分辨率提升到1 280×960 PPI,模板匹配准确率为49.4％,峰值信噪比PSNR值高于36.34 dB,结构相似度SSIM值高于0.972,重建效果较好,更适合用于特定场景下的红外图像目标特征识别.
In order to improve the effect of infrared image super-resolution reconstruction,an infrared image super-resolution reconstruction method based on the fusion of visible light and near-infrared light based on depth learning is proposed.The salient region detection model of infrared image is established by using the reflective characteristics and infrared radiation characteristics of infrared image;The edge contour features of the image are detected by the appear-ance difference level between the visible light and near-infrared images,and the fusion feature parameters of visible light and near-infrared light are extracted;According to different fusion levels,image signal level,pixel level,feature level and decision level are reconstructed to extract image edge,shape and texture features;According to the noise lev-el of the feature distribution and the registration quality,the infrared image super-resolution reconstruction is realized by using the depth learning algorithm.The simulation test results show that the method has a strong ability to detect the salient features of infrared image reconstruction,and the image resolution is improved to 1 280x960 PPI,the template matching accuracy is 49.4％,the peak signal to noise ratio PSNR value is higher than 36.34 dB,and the structure similarity SSIM value is higher than 0.972.The reconstruction effect is good,and it is more suitable for infrared image target feature recognition in specific scenes.
In order to improve the effect of infrared image super-resolution reconstruction,an infrared image super-resolution reconstruction method based on the fusion of visible light and near-infrared light based on depth learning is proposed.The salient region detection model of infrared image is established by using the reflective characteristics and infrared radiation characteristics of infrared image;The edge contour features of the image are detected by the appear-ance difference level between the visible light and near-infrared images,and the fusion feature parameters of visible light and near-infrared light are extracted;According to different fusion levels,image signal level,pixel level,feature level and decision level are reconstructed to extract image edge,shape and texture features;According to the noise lev-el of the feature distribution and the registration quality,the infrared image super-resolution reconstruction is realized by using the depth learning algorithm.The simulation test results show that the method has a strong ability to detect the salient features of infrared image reconstruction,and the image resolution is improved to 1 280x960 PPI,the template matching accuracy is 49.4％,the peak signal to noise ratio PSNR value is higher than 36.34 dB,and the structure similarity SSIM value is higher than 0.972.The reconstruction effect is good,and it is more suitable for infrared image target feature recognition in specific scenes.