| 127 | 1 | 38 |
| 下载次数 | 被引频次 | 阅读次数 |
经典的自编码网络存在计算复杂度大、全连接层丢失特征图位置信息、图像重构质量不佳等不足。本文提出了一种基于子像素全卷积的自编码网络结构优化,用卷积层优化全连接层,用子像素网络优化解码层,不仅提升了自编码网络的效率,而且图像重构质量也有很大提升,聚类性能也有提升。实验结果表明,本文优化自编码网络结构方法与经典自编码网络相比,在图像重构PSNR上平均提升3%,运算时间平均节省48%。
Abstract：The classical self-encoding network has some disadvantages such as large computational complexity, loss of feature map location information of the all-connected layer, and poor image reconstruction quality. In this paper, a self-encoding network structure optimization based on sub-pixel full convolution is proposed. The convolution layer is used to optimize the fully connected layer, and the sub-pixel network is used to optimize the decoding layer, which not only improves the efficiency of the self-encoding network, but also improves the image reconstruction quality. Moreover, the clustering performance is also improved. The experimental results show that compared with the classical self-encoding network, the method of optimizing the self-encoding network structure improves the average PSNR of image reconstruction by 3%, and the operation time saves by 48%.
[1]Gillum T L,George R H,Leitmeyer J E.An autoencoder for clinical and regulatory data processing[J].Drug information journal,1995,29(1):107-113.
[2]Park S C,Park M K,Kang M G.Super-resolution image reconstruction:a technical overview[J].IEEE signal processing magazine,2003,20(3):21-36.
[3]Pu Y,Gan Z,Henao R,et al.Variational autoencoder for deep learning of images,labels and captions[C]//2016Advances in neural information processing systems,Barcelona,Spain:NIPS,2016:2352-2360.
[4]Arganda-Carreras I,Turaga S C,Berger D R,et al.Crowdsourcing the creation of image segmentation algorithms for connectomics[J].Frontiers in neuroanatomy,2015,9:142.
[5]Krizhevsky A,Sutskever I,Hinton G E.Imagenet classification with deep convolutional neural networks[C]//Advances in neural information processing systems.2012:1097-1105.
[6]Zhou H,Li Z.Deep networks with non-static activation function[J].Multimedia Tools and Applications,2018:1-15.
[7]Long J,Shelhamer E,Darrell T.Fully convolutional networks for semantic segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.2015:3431-3440.
[8]Shi W,Caballero J,Huszár F,et al.Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2016:1874-1883.
[9]Nene S A,Nayar S K,Murase H.Columbia University Image Library(COIL-20)[J].1996-02)[2010-05-23].
[10]肖宇.聚类分析及其在图像处理中的应用[D].北京交通大学,2012.Xiao Yu.Cluster Analysis and its Application on Image Processing[D].Beijing Jiaotong University,2012.(in Chinese)
[11]He K,Zhang X,Ren S,et al.Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.2016:770-778.
[12]Huang G,Liu Z,Weinberger K Q,et al.Densely connected convolutional networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.2017,1(2):3.
基本信息:
中图分类号:TP391.41
引用信息:
[1]杨火祥,柳伟,孟凡阳,等.基于子像素全卷积的自编码网络结构优化[J].深圳信息职业技术学院学报,2018,16(02):78-84.
基金信息:
深圳市科技计划项目(KJYY20170724152625446)
2018-06-15
2018-06-15