Improved convolutional neural network algorithm based on weight freezing method

As neural network became one of the indispensable technologies in many fields, a large number of data processing methods emerged one after another. Convolutional neural network (CNN) soon became a useful tool to classify images. However, the drawback of CNN is quite obviously, which is the training time would be tremendous long if the network running on CPU. To tackle this issue, this paper improves the BP algorithm based on the weight freezing method and entropy weight data processing method, and uses the experiment to verify the effectiveness of this approach. With MNIST handwritten database, CIFAR-10 database and PUBFIG83 and LFW database as background, the original LeNet-5 model has been optimized and the back propagation (BP) algorithm used in the original network training is improved. The results show that the improved algorithm can effectively reduce the complexity of the algorithm, and the training time reduces to 8%.