A Pooling Algorithm of Convolutional Neural Network Based on Frequency Domain Transformation
Zhichen Wang, Qian He · 2023
Pooling is an important component in neural networks. The main methods of pooling are mean pooling, maximum pooling, and random pooling. At the same time, there are also related frequency domain pooling methods. However, these methods all have corresponding drawbacks. In this paper, a pooling operation method of convolutional neural network based on frequency domain change is proposed. This algorithm has the characteristic of retaining the most original information in the image and better transmitting the main information in the neural network. The neural network will therefore have better accuracy performance and complete pooling operations while convolution is in progress. Therefore, the computation time of the entire neural network will also be reduced.