一种基于双重网络模型的单幅图像超分辨率方法  

A Single Image Super-Resolution Method Based on the Dual Network Model

作　　者：倪翠[1] 王朋[1] 张广渊[1] 李克峰[1] NI Cui;WANG Peng;ZHANG Guangyuan;LI Kefeng(School of Information Science and Electric Engineering,Shandong JiaoTong University,Jinan 250357,Shandong,China)

机构地区：[1]山东交通学院信息科学与电气工程学院,山东济南250357

出　　处：《应用科学学报》2021年第2期321-329,共9页Journal of Applied Sciences

基　　金：国家自然科学基金青年基金(No.61502277)资助。

摘　　要：本文对深度学习领域中的高效亚像素卷积神经网络(efficient sub-pixel convo-lutional neural network,ESPCN)算法进行了改进,通过加入残差网络知识,调整原有的ESPCN构造结构,提出了一种双重网络模型下单幅图像超分辨率重建方法。通过实验证明:该算法能够有效地提高单幅图像超分辨率重建的精度,丰富重建后的细节信息。This article mainly improves the efficient sub-pixel convolutional neural network(ESPCN)algorithm in the field of deep learning.By adding residual network knowledge and adjusting original ESPCN structure,a dual network model is proposed for single frame image super-resolution reconstruction method.Experimental results show that this algo-rithm can effectively improve the accuracy of single-image super-resolution reconstruction and enrich the detailed information after reconstruction.

关 键 词：残差网络 亚像素卷积 带组归一化 隐藏层 

分 类 号：P315.69[天文地球—地震学]