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

Restricted Boltzmann Machine Algorithm for Image Classification and Its Parallel Implementation Based on (2D)~2 PCA

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.

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 classification based on two-way 2-dimension principal component analysis （（2D）^2PCA） is put forward. The algorithm firstly 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, finally, builds the RBM network with contrastive divergence algorithm and realizes the image classification. 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.