基于轻量化卷积神经网络的改进模型与验证

卷积神经网络随着深度和结果复杂度的不断增加，其参数量和计算量大大制约了它的应用场景，本文在SueezeNet网络结构基础上引用分组卷积并采用Channel-shuffel来解决分组卷积后的信息不流通问题。以减少原有网络结构的的参数量提高网络运行效率。在ORL数据集的验证表现也表明，在网络参数减少的情况下分类精度和收敛效率并不会有降低甚至略有提高。可以体现分组卷积在结构轻量化上的有效性。

A new module based on light-weight convolutional neural network

In order to reduce the parameters of the original network structure and improve the efficiency of network operation. With the depth and complexity of the Convolutional Neural Network continuously increasing, its application scenarios are greatly restricted by its parameters and calculations. In this paper, the problem of information circulation with packet convolution is solved based on the SueezeNet network structure, packet convolution and Channel-shuffel. According to verification performance on the ORL data set, The results show that classification accuracy and convergence efficiency will not be reduced or even slightly improved with the network parameters reduced.It is concluded that the effectiveness of packet convolution in lightening the structure.