查询结果:   贺瑜飞,高宏伟.基于多层连接卷积神经网络的单帧图像超分辨重建[J].计算机应用与软件,2019,36(5):220 - 224,326.
中文标题
基于多层连接卷积神经网络的单帧图像超分辨重建
发表栏目
图像处理与应用
摘要点击数
599
英文标题
SUPER-RESOLUTION RECONSTRUCTION USING MULTILAYER CONNECTED CONVOLUTIONAL NEURAL NETWORK FOR SINGLE-FRAME IMAGE
作 者
贺瑜飞 高宏伟 He Yufei Gao Hongwei
作者单位
榆林学院数学与统计学院 陕西 榆林 719000 榆林学院现代设计与先进制造技术研究中心 陕西 榆林 719000    
英文单位
School of Mathematics and Statistics, Yulin University, Yulin 719000, Shaanxi, China Research Center for Contemporary Design and Advanced Manufacturing Technology, Yulin University, Yulin 719000,Shaanxi, China    
关键词
超分辨重建 红外图像 深度学习 特征级联 损失函数
Keywords
Super-resolution reconstruction Infrared image Deep learning Feature cascade Loss function
基金项目
榆林学院高层次人才科研启动基金项目(12GK43)
作者资料
贺瑜飞,高级讲师,主研领域:模式识别,光电设计,深度学习,红外图像预处理。高宏伟,教授。 。
文章摘要
现有光电观瞄装备采取的电子变倍功能大都是采用线性插值进行放大重建,存在细节不明显, 边缘模糊的现象。针对目前超分辨重建算法存在的问题,提出一个多连接卷积网络。该网络构建出多连接网络结构,通过一个较长的跳跃式策略进行恒等映射,实现低层次特征和高级特征的级联,能够同时表征各种复杂的重构场景。采用双参数损失函数来优化训练深度网络,提高网路模型的泛化能力。仿真实验结果表明,该方法能够生成具有丰富细节而且清晰的高分辨红外图像,同时也在客观定量评价上都有很大提高。
Abstract
Most of the electronic magnification functions adopted by existing photoelectric sight equipment are magnified and reconstructed by linear interpolation, and the details are not obvious and the edges are blurred. In this paper, we proposed a multi-connection convolutional network to solve the problems of the super-resolution reconstruction (SR) algorithm. The network constructed a multi-connection network structure, where a long skip strategy was used to obtain the identity mapping. It could concatenate the low-level features and high-level features, and simultaneously represented various complex reconstruction scenes. A two parameter loss function was applied to optimize and train the deep network and improve the generalization ability of the network model. The simulation results show that the proposed SR algorithm can generate high resolution infrared images with rich details and clearness, and the objective quantitative evaluation is also greatly improved.
下载PDF全文