收稿日期: 2017-09-29
修回日期: 2017-11-13
网络出版日期: 2018-05-31
基金资助
黑龙江省自然科学基金（No.ZD201403）；国家林业局林业行业公益专项基金（No.201504307）；哈尔滨市应用技术研究和开发项目基金（No.201504307）；黑龙江工程学院教育科学研究规划项目基金（No.JG1410）；黑龙江省大学生创业实践项目基金（No.201611802059）资助
Restricted Boltzmann Machine Algorithm for Image Classifcation and Its Parallel Implementation Based on (2D)2 PCA
Received date: 2017-09-29
Revised date: 2017-11-13
Online published: 2018-05-31
为解决受限玻尔兹曼机（restricted Boltzman machine，RBM）理论对高分辨率图像分类的时间复杂度高的问题，提出了一种基于双向二维主成分分析（two-way 2-dimension principal component analysis，（2D）2PCA）的RBM图像分类算法.该算法首先应用（2D）2PCA对待处理图像在X和Z两个方向上进行降维处理，从而提取出图像的主成分，将主成分作为RBM网络可见层的输入数据，应用对比散度算法训练构建玻尔兹曼机网络，达到对图像进行分类的目的.该算法有效解决了RBM处理高分辨率图像时网络训练速度慢，甚至整个网络训练状态无法收敛的问题.通过在Hadoop并行数据处理平台的实验表明：该算法不仅能有效提高处理高分辨率图像的速度，而且具备良好的并行性，在具有4台处理机的并行集群下，其加速比达到了3.13.
宋海峰, 陈广胜, 景维鹏, 杨巍巍 . 基于(2D)2PCA的受限玻尔兹曼机图像分类算法及其并行化实现[J]. 应用科学学报, 2018 , 36(3) : 495 -503 . DOI: 10.3969/j.issn.0255-8297.2018.03.009
In this paper, in order to solve the problem of high time complexity when using restricted Boltzmann machine (RBM) to classify the high resolution image, a RBM algorithm for image classifcation based on two-way 2-dimension principal component analysis ((2D)2PCA) is put forward. The algorithm frstly reduces the dimension in X and Z direction on the image by using (2D)2PCA, secondly extracts the principle components as the input data of the visible layer of RBM network, fnally, builds the RBM network with contrastive divergence algorithm and realizes the image classifcation. The proposed algorithm can solve the drawbacks of the long training time of RBM network, which might lead to the convergence failure of the entire network training state as processing the high resolution image. The parallel experimental results show that the algorithm can achieve both high speed and good parallelism as processing high resolution images. The ratio of acceleration reaches 3.13 as employing a cluster of four parallel machines.
[1] Flusser J, Suk T. Pattern recognition by afne moment invariants[J]. Pattern Recognition, 1993, 26(1):167-174.
[2] Ren J, Li X, Haupt J. Robust PCA via tensor outlier pursuit[C]//Conference on Signals, Systems & Computers, 2017:1744-1749.
[3] Lu M, Huang J Z, Qian X. Sparse exponential family principal component analysis[J]. Pattern Recognition, 2016, 60:681-691.
[4] Cheng Z D, Zhang Y J, Fan X. Criteria for 2DPCA superior to PCA in image feature extraction[J]. Chinese Journal of Engineering Mathematics, 2009, 26(6):951-961.
[5] Shi Z G. A human face recognition method of 2DPCA based on modular residual image[J]. Journal of Inner Mongolia Normal University, 2015, 44(3):380-384.
[6] Wang H X. 2DPCA with L1-norm for simultaneously robust and sparse modelling[J]. Neural Networks, 2013, 46(10):190-198.
[7] Li W. The image feature extraction algorithm based on the DWT and the improved 2DPCA[J]. Applied Mechanics and Materials, 2014, 556-562:5042-5045.
[8] Yang J, Zhang D, Frangi A F, Yang J Y. Two-dimensional PCA:a new approach to appearance-based face representation and recognition[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2004, 26(1):131-137.
[9] Yang J, Yang J Y. From image vector to matrix:a straightforward image projection techniqueIMPCA vs. PCA[J]. Pattern Recognition, 2002, 35(9):1997-1999
[10] Liu S, Feng L, Qiao H. Scatter balance:an angle-based supervised dimensionality reduction[J]. IEEE Transactions on Neural Networks & Learning Systems, 2015, 26(2):277-289.
[11] Mahanta M S, Plataniotis K N. Ranking 2DLDA features based on fsher discriminance[J]. IEEE International Conference on Acoustics, 2014:8307-8311.
[12] Mashhoori A, Zolghadri M J. Block-wise two-directional 2DPCA with ensemble learning for face recognition[J]. Neurocomputing, 2013, 108(5):111-117.
[13] Längkvist M, Karlsson L, Loutfi A. A review of unsupervised feature learning and deep learning for time-series modeling[J]. Pattern Recognition Letters, 2014, 42(1):11-24.
[14] Krizhevsky A, Sutskever I, Hinton G E. Imagenet classifcation with deep convolutional neural networks[J]. International Conference on Neural Information Processing Systems, 2012, 60(2):1097-1105
[15] Zorzi M, Testolin A, Stoianov I P. Modeling language and cognition with deep unsupervised learning:a tutorial overview[J]. Frontiers in Psychology, 2013, 4(5):515-528.
[16] Hinton G E, Osindero S, Teh Y W. A fast learning algorithm for deep belief nets[J]. Neural Computer, 2006, 18(7):1527-1554.
[17] Ranzato M, Krizhevsky A, Hinton G E. Factored 3-way restricted Boltzmann machines for modeling natural images[J]. Journal of Machine Learning Research, 2010(9):621-628.
[18] Hinton G E. Training products of experts by minimizing contrastive divergence[J]. Neural Computer, 2002(148):1771-1800.
[19] Shvachko K, Kuang H, Radia S. The Hadoop distributed fle system[C]//IEEE Symposium on MASS Storage Systems and Technologies, 2010:1-10.
[20] Vemula S, Crick C. Hadoop image processing framework[C]//IEEE International Congress on Big Data, 2015:506-513.
[21] Kachris C, Sirakoulis G C, Soudris D. A map reduce scratchpad memory for multi-core cloud computing applications[J]. Microprocessors & Microsystems, 2015, 39(8):599-608.
/
| 〈 |
|
〉 |
