Pooling Method On PCNN in Convolutional Neural Network
Li-sha Yao, Guo-ming Xu, Feng Zhao · Journal of Physics Conference Series · 2020
Abstract The pooling method aggregates the points in the neighborhood in Convolutional Neural Network(CNN). It can not only reduce the dimension, but also improve the results, so that the results are not easy to over-fit. However, the common pooling methods have the problems of single feature and lack of self-adaptability. In order to solve this problem, the Pulse Coupled Neural Network (PCNN) is introduced and a pooling method based on PCNN is proposed. The algorithm learns the weights of each eigenvalue from the convoluted neighborhood sub-region by PCNN and fuses them to get the final pooling result. The experimental results on image recognition datasets MNIST, CIFAR-100 show that the proposed PCNN-based pooling method has better recognition effect and improves the performance of CNN compared with the existing pooling methods.