论文部分内容阅读
随着深度神经网络图像处理领域的研究与应用,图像分类精度得到了大幅提升.然而深度神经网络的随机处理过程导致同一网络对相同的训练图像,重复训练会提取到有差异性的特征.为了利用这种差异性,本文提出了一种对称神经网络模型,将两个特征维度相同的深度神经网络作为对称模型的左右子网络,通过前向传播得到有差异的图像特征,并在联合层对差异特征进行融合.为了优化网络,采用差异度量函数度量左右子网络的差异,用差异优化网络的损失函数,进而通过反向传播微调模型参数.基于上述思想,本文将扩展了左右子网络为深度置信网的对称深度置信网及左右子网络为卷积神经网络的对称卷积神经网络,在数据集MNIST和CIFAR-10上的实验测试表明,相较于深度置信网及卷积神经网络,该方式集成的对称深度模型能取得较好的分类性能.
With the research and application of deep neural network image processing, the accuracy of image classification has been greatly improved.However, the random processing of deep neural networks leads to the same network to the same training images, repetitive training will extract the characteristics of different Taking advantage of this difference, a symmetric neural network model is proposed in this paper. Two deep neural networks with the same feature dimension are used as the left and right sub-networks of the symmetric model to obtain differentiated image features by forward propagation. In order to optimize the network, the difference measure function is used to measure the difference between the left and right sub-networks, the difference is used to optimize the loss function of the network, and the fine tuning of the model parameters is made through back propagation.Based on the above idea, The symmetric depth of belief network and the left and right sub-networks are symmetric convolutional neural networks of convolutional neural networks. Experimental results on MNIST and CIFAR-10 data sets show that compared with deep belief networks and convolutional neural networks, Symmetric depth model integration can achieve better classification performance.