一种卷积神经网络的图像矩正则化策略  

Convolutional neural network's image moment regularizing strategy

作　　者：殷瑞[1,2] 苏松志[1,2] 李绍滋[1,2] 

机构地区：[1]厦门大学信息科学与技术学院,福建厦门361005 [2]厦门大学福建省仿脑智能系统重点实验室,福建厦门361005

出　　处：《智能系统学报》2016年第1期43-48,共6页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(61202143;61572409);福建省自然科学基金资助项目(2013J05100)

摘　　要：卷积神经网络的池化策略包含极大池化和平均池化,极大池化选择池化区域中的最大值,极易出现过抑合现象;平均池化对池化区域中所有元素赋予相同权重,降低了高频分量的权重。本文提出将矩池化作为卷积神经网络的正则化策略,矩池化将几何矩概念引入到卷积神经网络的池化过程中,首先计算池化区域的中心矩,然后根据类插值法依概率随机地从中心矩的4个邻域中选择响应值。在数据集MNIST、CIFAR10、CIFAR100上的实验结果表明随着训练迭代次数的增加,矩池化的训练误差和测试误差最低,矩池化的高差别性和强鲁棒性使其获得了比极大池化和平均池化更好的泛化能力。There are two kinds of pooling strategies for convolutional neural network（ CNN） as follows： max pooling and average pooling. Max pooling simply chooses the maximum element,which makes this strategy extremely prone to overfitting. Average pooling endows all elements with the same weight,which lowers the weight of the high-frequency components. In this study,we propose moment pooling as a regularization strategy for CNN. First,we introduce the geometric moment to CNN pooling and calculate the central moment of the pooling region. Then,we randomly select the response values based on the probability-like interpolation method from the four neighbors of the moment as per their probability. Experiments on the MNIST,CIFAR10,and CIFAR100 datasets show that moment pooling obtains the fewest training and test errors with training iteration increments. This strategy＇s robustness and strong discrimination capability yield better generalization results than those from the max and average pooling methods.

关 键 词：中心矩 随机选择 池化 卷积神经网络 过抑合 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]