为了提高深度卷积神经网络(DCNN)的图像并行处理能力,提高其图像识别的准确率和运行效率,研究过程以MapReduce并行计算框架和从图像到矩阵(Image to Column,Im2col)算法,分别进行原始图像特征并行提取和筛选、模型并行训练和参数并行更新,构建了并行DCNN优化算法.在性能检测阶段,将全连接神经网络和基于特征图和并行计算熵的深度卷积神经网络算法作为对照组,对比TOP-1准确率、浮点运算量、损失函数振荡性、运算时长四项指标,结果显示,此次提出的并行DCNN优化算法性能最佳.
Parallel DCNN optimization algorithm study based on MapReduce and Im2col
In order to improve the depth of the convolutional neural network(deep convolutional neural network,DCNN)im-age parallel processing ability,improve the accuracy of image recognition and running efficiency,research process with MapRe-duce parallel computing framework and from the image to the matrix(Image to Column,Im2col)algorithm,the original image fea-ture extraction and screening,model parallel training and parameter parallel update,build the parallel DCNN optimization algo-rithm.In the performance detection stage,the fully connected neural network and the deep convolutional neural network algorithm based on feature graph and parallel computing entropy were used as the control group to compare the four indexes of TOP-1 accu-racy,floating point operation,loss function oscillator and operation time.The results showed that the proposed parallel DCNN opti-mization algorithm had the best performance.
万方数据
