A improved pooling method for convolutional neural networks
Authors:
Top Cited Papers
Open Access
- 18 January 2024
- journal article
- research article
- Published by Springer Nature in Scientific Reports
- Vol. 14 (1) , 1-22
Abstract
The pooling layer in convolutional neural networks plays a crucial role in reducing spatial dimensions, and improving computational efficiency. However, standard pooling operations such as max pooling or average pooling are not suitable for all applications and data types. Therefore, developing custom pooling layers that can adaptively learn and extract relevant features from specific datasets is of great significance. In this paper, we propose a novel approach to design and implement customizable pooling layers to enhance feature extraction capabilities in CNNs. The proposed T-Max-Avg pooling layer incorporates a threshold parameter T, which selects the K highest interacting pixels as specified, allowing it to control whether the output features of the input data are based on the maximum values or weighted averages. By learning the optimal pooling strategy during training, our custom pooling layer can effectively capture and represent discriminative information in the input data, thereby improving classification performance. Experimental results show that the proposed T-Max-Avg pooling layer achieves good performance on three different datasets. When compared to LeNet-5 model with average pooling, max pooling, and Avg-TopK methods, the T-Max-Avg pooling method achieves the highest accuracy on CIFAR-10, CIFAR-100, and MNIST datasets.Keywords
Funding Information
- National Natural Science Foundation of China (61662043)
- Science and Technology Program of Gansu Province (21JR7RA288, 21JR7RE174)
This publication has 32 references indexed in Scilit:
- CNN architectures for large-scale audio classification2017
- Convolutional neural networks for hyperspectral image classificationNeurocomputing, 2017
- Deep Residual Learning for Image Recognition2016
- Rethinking the Inception Architecture for Computer Vision2016
- Bilinear CNN Models for Fine-Grained Visual Recognition2015
- Machine learning: Trends, perspectives, and prospectsScience, 2015
- Deep learningNature, 2015
- Mixed Pooling for Convolutional Neural Networks2014
- Understanding the meaning of accuracy, trueness and precisionAccreditation and Quality Assurance, 2006
- Gradient-based learning applied to document recognitionProceedings of the IEEE, 1998