全卷积多并联残差神经网络

随着人工智能的火热发展,深度学习已经在很多领域占有了一席之地.作为深度学习中一个典型网络--残差神经网络模型自提出之日起就成为了众多研究者的关注点.然而,残差神经网络还有很大的改进空间.为了更好地解决反向传播中梯度减小的问题,本文提出了一种改进的残差神经网络,称为全卷积多并联残差神经网络.在该网络中,每一层的特征信息不仅传输到下一层还输出到最后的平均池化层.为了测试该网络的性能,分别在三个数据集(MNIST,CIFAR-10和CIFAR-100)上对比图像分类的结果.实验结果表明,改进后的全卷积多并联残差神经网络与残差网络相比具有更高的分类准确率和更好的泛化能力.

Fully Convolutional Multi-parallel Residual Neural Network

With the rapid development of artificial intelligence,deep learning has occupied a place in many fields.As a typical network in deep learning,the residual neural network has become a hot spot for many researchers since its inception.However,there is still much room to improvement in the residual neural network.In order to better solve the problem of gradient reduction in backpropagation,this paper proposes an improved residual neural network called fully convolutional multi-parallel residual neural network.In this network,the feature information of each layer is transmitted not only to the next layer but also to the last average pooling layer.To test the performance of the network,the results of the image classification were compared on three data sets(MNIST,CIFAR-10 and CIFAR-100).The experimental results showthat the improved fully convolutional multi-parallel residual neural network has higher classification accuracy and better generalization than the residual network.