Abstract: The construction process of traditional convolutional neural network (CNN) model has many shortcomings, such as over reliance on experience knowledge, a lot of parameters and difficult to training. At the same time, in view of the important value of constructing strategy of CNN model in complex multi-class problems, a new construction method of adaptive deep CNN model was proposed. First, the convolution layer and pooling layer of the initial CNN model are set only to include one feature map; and then, the convergence rate of CNN is used as evaluation index, the global expansion of network is carried out. After global expansion, the CNN is controlled to local expansion according to the recognition rate of cross validation samples. The local network learning is stopped until the recognition rate reaches the expected value. Finally, the training process for new samples, the adaptive incremental learning of network structure is realized by expanding some new branches. The superiority of proposed algorithm in network training time and recognition effect is verified through some image recognition experiments.

Key words: deep convolutional neural network, adaptive model constructing, deep learning, image recognition