Study of convolutional neural network based on fusion residual mechanism

Abstract

In recent years, due to the rapid development of computer technology, the calculate ability has been continuously improved. The depth of the deep convolutional neural network is getting larger and larger, and the performance of the model can also be improved. However, beyond a certain range, the training of the network model is prone to gradient explosion or vanishing. This article starts from the design of the residual mechanism, integrates with the traditional Vgg-16 network model, transfers image information from multiple aspects, obtains richer input features, “reduces” the depth of the network, and controls each network parameter operation effectively, suppress the occurrence of gradient explosion and vanishing, and improve the accuracy of picture classification. Experimental results show that the model proposed in this paper improves the accuracy of the network classification by 5.8% compared with that before the improvement.