基于非监督预训练的结构优化卷积神经网络

针对典型卷积神经网络卷积核由经验设置且网络结构固定不变难以后期再学习的问题，基于稀疏自编码器（Sparse Autoencoder, SAE）和卷积神经网络（Convoutional Neural Network, CNN），提出新的CNN模型。该模型通过SAE预训练CNN网络的卷积核，提取有效特征；并在典型CNN结构基础上增加一条网络支路，使得后续再学习时只更新支路权值，记忆已有特征并增加新特征。文中模型在MNIST数据集上迭代更新10次网络权值可以使测试识别率达到97.65%；在手写汉字数据集HCL2000中的简单字，中等字，复杂字及相似字上的测试正确率能达93%以上；50个训练样本，250个测试样本时，相似字识别率可达80.36%；比典型CNN及传统手写汉字识别方法更具泛化性。实验表明所提出方法可有效应用于手写字等图像识别应用中。

Structure Optimized Convolutional Neural Network Based on Unsupervised Pre-training

For the problem that convolutional kernels of typical convolutional neural network set by experience and being difficult for fixed network architecture to realize subsequent re-learning, a new Convoutional neural network (CNN) model was proposed based on Sparse Autoencoder(SAE) and typical CNN. By SAE pre-training convolution kernels and increasing a branch based on typical CNN, only branch weights of the model were updated while doing subsequent re-learning, and new features were added while memorizing existing characteristics. On dataset MNIST, recognition rate of 97.65% was achieved by updating weights for 10 times. Recognition rates of more than 93% were achieved on dataset HCL2000. The recognition rate of similar Chinese characters reached 80.36% using 50 samples for training, 250 for testing. Compared with typical CNN and traditional methods, the proposed method was more generalized. Experiments show that the proposed method can be effectively applied to image recognition applications such as handwritten characters.