Deep Convolutional Network Based on Interleaved Fusion Group

摘要

It is known that the classification accuracy of the deep convolutional network can be remarkably improved by increasing its depth and width. However, as the network size increases, the number of network parameters will increase significantly, which results in network redundancy and performance degradation. In order to reduce network redundancy and improve classification accuracy by means of reducing the number of parameters, a highly modularized and lightweight deep interleaved fusion group convolutional network is proposed. The main idea is to design a more efficient "network computing manner," so as to simplify the network model, reduce redundancy, and improve computational efficiency without the loss of classification accuracy. First, a template block is constructed by using the stacking and split-transform-merge strategies, which simplifies the network model by reducing the number of parameters. Then, the introduced group convolution and structured sparse convolution further simplify the network model and improve the computational efficiency. Experiments on the standard image recognition tasks have shown that the proposed convolutional network achieves better classification performance and has superior generalization ability compared with state-of-the-art deep networks.

关键词