基于衰减因子的双通道神经网络图像分类算法

发布时间：2018-05-18 16:04

本文选题：图像分类 + 深度学习　；参考：《系统工程与电子技术》2017年06期

【摘要】：为解决深度卷积神经网络(convolutional neural networks,CNN)难以训练的问题,提出一种快速、高效的双通道神经网络(dual-channel neural networks,DCNN),该神经网络由直通通道和卷积通道两种通道构成,直通通道负责保障深度网络的畅通性,卷积通道负责深度网络的学习。考虑到深层网络在训练时容易出现性能不稳定的问题,在卷积通道上引入卷积衰减因子,对其响应数据进行约束。设计一种"双池化层"对同一特征图进行降采样,不仅可以防止训练过拟合,还能保证各通道的维度一致性。在3个图像数据集CIFAR-10、CIFAR-100和MNIST上的实验结果表明,无论是神经网络的可训练深度、稳定性和分类精度,DCNN都明显优于现有的深度卷积神经网络。
[Abstract]:In order to solve the problem that convolutional neural networks are difficult to train, a fast and efficient dual-channel neural network is proposed. The neural network consists of two channels: through channel and convolutional channel. The pass-through channel is responsible for ensuring the smooth flow of the deep network, and the convolution channel is responsible for the learning of the deep network. Considering that the deep network is prone to unstable performance during training, the convolution attenuation factor is introduced into the convolution channel to constrain the response data. A kind of "double cell layer" is designed to decimate the same feature map, which can not only prevent the training from over-fitting, but also ensure the consistency of each channel dimension. The experimental results on three image data sets CIFAR-10 CIFAR-100 and MNIST show that DCNNs are superior to the existing deep convolution neural networks in terms of training depth, stability and classification accuracy.
【作者单位】：中国民航大学天津市智能信号与图像处理重点实验室;
【基金】：国家自然科学青年基金(11402294) 天津市智能信号与图像处理重点实验室开放基金项目(2015AFS03) 中国民航大学第六期波音基金(20160159209)资助课题
【分类号】：TP183;TP391.41

【相似文献】