Research on image classification based on improved DenseNet
Hao Wang, Song Zhili · 2021
In order to improve the accuracy of image classification, many researchers will improve the network structure, enhance the way of data processing or design a new activation function. This paper improves the initial convolution block and activation function of DenseNet's network structure, and proposes an improved DenseNet network. The improved initial convolution module can better fit the initial channel of the data image, so that different data sets can get better feature extraction results. The improved neural network activation function has a better convergence effect on model training, leading the performance of multiple activation functions in ablation experiments. Experiments have proved that the improved DenseNet model has reached a high classification accuracy rate on public data sets, leading more than a dozen classic convolutional neural network structures.