计算机应用

• 人工智能与仿真 •    下一篇

基于高斯函数的池化算法

王宇航,周永霞,吴良武   

  1. 中国计量大学信息工程学院
  • 收稿日期:2021-07-13 修回日期:2021-09-21 发布日期:2021-10-18 出版日期:2021-10-18
  • 通讯作者: 王宇航

Pooling algorithm based on Gaussian function

  • Received:2021-07-13 Revised:2021-09-21 Online:2021-10-18 Published:2021-10-18

摘要: 针对卷积神经网络中的传统池化算法不能很好地考虑到池化域内每个元素与该池化域所含特征之间关联性的问题,提出一种基于高斯函数的池化算法。该算法首先根据池化域内各元素的值和所有元素的最大值计算高斯函数的三个参数值,然后运用高斯函数计算池化域内所有元素的权重,最后根据该权重对池化域内所有元素值计算加权平均值,并以此作为池化结果。选择LeNet5、VGG16、ResNet18和MobileNet v3作为实验模型,在公开数据集CIFAR-10、Fer2013和德国交通标志识别基准(GTSRB)上进行实验,并与最大池化、平均池化、随机池化、混合池化、模糊池化、融合随机池化和soft池化七种池化算法进行对比。实验结果表明,该算法在三个数据集上相较其它算法在精度方面均有0.5到6个百分点的提升,且该算法在运行时间方面优于上述除最大池化和平均池化两种池化算法外的其它池化算法,从而验证该算法有效且具适合应用于对运算时间要求不高但对精度要求较高的情况。

Abstract: Abstract: Aiming at the problem that the traditional pooling algorithm in Convolutional Neural Networks cannot well consider the correlation between each element in the pooling domain and the features contained in the pooling domain, a pooling algorithm based on Gaussian function was proposed. Firstly, according to the value of each element in the pooling domain and the maximum value of all elements, the three parameter values of the Gaussian function was calculated by the algorithm. Then the Gaussian function was used to calculate the weight of all elements in the pooling domain. Finally, the weighted average value of all elements in the pooling domain was calculated according to the weight, which was used as the pooled result. LeNet5, VGG(Visual Geometry Group)16, ResNet(Residual Network)18 and MobileNet v3 were selected as the experimental models. and experiments were carried out on public data sets CIFAR-10, Fer2013 and the German Traffic Sign Recognition Benchmark (GTSRB). Max pooling, average pooling, random pooling, mix pooling, fuzzy pooling, fusion random pooling and soft pooling were compared in the experiments. The results show that the accuracy by 0.5 to 6 percentage points is improved by the algorithm compared with other algorithms on the three data sets, and the running time of the algorithm is better than the other pooling algorithms except the max pooling algorithm and average pooling algorithm, so as to verify that the algorithm is effective and suitable for the situation where the operation time is not high but the accuracy is high.

中图分类号: