A lightweight iterative error reconstruction network for infrared image super-resolution in smart grid

被引:14
作者
Chen, Lihui [1 ]
Tang, Rui [1 ]
Anisetti, Marco [2 ]
Yang, Xiaomin [1 ]
机构
[1] Sichuan Univ, Coll Elect & Informat Engn, Chengdu 610064, Sichuan, Peoples R China
[2] Univ Milan, Dipartimento Informat DI, Via Celoria 18, I-20133 Milan, MI, Italy
基金
中国国家自然科学基金;
关键词
Convolutional neural networks; Deep learning; Smart grid; Infrared image super-resolution; ENHANCEMENT; DICTIONARY;
D O I
10.1016/j.scs.2020.102520
中图分类号
TU [建筑科学];
学科分类号
0813 ;
摘要
Thermal infrared (IR) images are widely used in smart grids for numerous applications. These applications prefer high-resolution (HR) IR images since HR IR images benefit the performance. However, HR IR imaging devices are extremely expensive. To save the cost of upgrading imaging devices, an iterative error reconstruction network (IERN) is proposed to improve the resolution of IR images. We first achieve efficient dense connections based on linearly compressive skip links. Slightly sacrificing the performance, the efficient dense connections can mark-edly reduce the parameters and computations of the vanilla dense connections. Then, an iterative error recon-struction mechanism is proposed to boost the performance, which enables IERN to restore many more textures and edges. Specifically, an initial SR image, high-level features, and up-sampled features are obtained firstly. Secondly, a SR error image is acquired by reconstructing the errors between the initial high-level features and the back-projected features from the up-sampled features. Thirdly, a new SR image is obtained by adding the SR error image to the initial SR image. Iterating the above process, the final SR image is achieved when the number of iterations reaches to the iteration threshold. Experimental results reveal the superiority of the proposed method over state-of-the-art methods.
引用
收藏
页数:13
相关论文
共 63 条
[21]   Lightweight Image Super-Resolution with Information Multi-distillation Network [J].
Hui, Zheng ;
Gao, Xinbo ;
Yang, Yunchu ;
Wang, Xiumei .
PROCEEDINGS OF THE 27TH ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA (MM'19), 2019, :2024-2032
[22]   Fast and Accurate Single Image Super-Resolution via Information Distillation Network [J].
Hui, Zheng ;
Wang, Xiumei ;
Gao, Xinbo .
2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2018, :723-731
[23]   IMPROVING RESOLUTION BY IMAGE REGISTRATION [J].
IRANI, M ;
PELEG, S .
CVGIP-GRAPHICAL MODELS AND IMAGE PROCESSING, 1991, 53 (03) :231-239
[24]   Medical image super-resolution for remote medical diagnosis in smart city: A case study based on the new healthcare reform of China [J].
Jiang, Lihua ;
Ye, Shuang ;
Zhao, Li ;
Ma, Xiao ;
Yang, Xiaomin .
SUSTAINABLE CITIES AND SOCIETY, 2019, 48
[25]   Perceptual Losses for Real-Time Style Transfer and Super-Resolution [J].
Johnson, Justin ;
Alahi, Alexandre ;
Li Fei-Fei .
COMPUTER VISION - ECCV 2016, PT II, 2016, 9906 :694-711
[26]   IDNet-A : Variant of DenseNet with Inception-family [J].
Kim, Cheol-jin ;
Ha, Young-guk .
2020 IEEE INTERNATIONAL CONFERENCE ON BIG DATA AND SMART COMPUTING (BIGCOMP 2020), 2020, :109-112
[27]  
Kim J, 2016, PROC CVPR IEEE, P1637, DOI [10.1109/CVPR.2016.182, 10.1109/CVPR.2016.181]
[28]  
Kim J, 2016, IEEE CONF COMPUT
[29]  
Kingma D.P., 2015, INT C LEARN REPR REP
[30]   ImageNet Classification with Deep Convolutional Neural Networks [J].
Krizhevsky, Alex ;
Sutskever, Ilya ;
Hinton, Geoffrey E. .
COMMUNICATIONS OF THE ACM, 2017, 60 (06) :84-90