An improved LeNet-5 model for Image Recognition
He Yanmei, Bo Wang, Zhu Zhaomin · Proceedings of the 2020 4th International Conference on Electronic Information Technology and Computer Engineering · 2020
This paper proposes an improved convolutional neural network structure which greatly reduces the scale of network training parameters. It uses the global average pooling algorithm instead of the full connection algorithm to improve the LeNet-5 network. The number of convolution kernels is increased, while the number of subsampling layers is reduced to an optimal value. After verification with MINST handwritten Arabic numeral data set, the results show that the improved network training parameters are only 34.8% of the original, and the recognition accuracy can achieve 99.3%.