Abstract: Convolutional neural network (CNN) is a deep learning model with strong expression and classification ability, and is currently widely used, but there are a variety of specific algorithms. In this paper, the realization of deep learning algorithm based on convolutional neural network CNN, including the function of convolution kernel, the role of pooling, the selection of activation function and the training process are discussed. And one example is explained, which facilitates the mastery of CNN. Finally, the future research direction of convolution neural network is summarized and forecasted.

Key words: convolutional neural network, back propagation, deep learning, convolutional layer, pooling layer