摘要：

针对传统手写数字识别方法识别率较低的问题,提出一种融合卷积神经网络(F-CNN)模型。通过结合暹罗网络(SN)模型和二进制卷积神经网络(B-CNN)模型的高级特征,扩展网络高级层的尺寸,增强F-CNN模型的特征表达能力。在网络训练过程中,设计周期性数据打乱策略,提高F-CNN模型的收敛速度,更好地实现手写数字识别。在MNIST数据集上的实验结果表明,融合模型对于手写数字的识别准确率达到99.10%,识别性能优于SN模型和B-CNN模型。

关键词: 手写数字, 融合模型, 卷积神经网络, 数据打乱策略, 收敛速度

Abstract:

Aiming at the problem that the recognition rate of traditional handwritten digits recognition method is low,this paper proposes a Fused Convolutional Neural Network(F-CNN) model.By combining the high-level features of the Siamese Network(SN) model and Binary Convolutional Neural Network(B-CNN) model,the F-CNN model expands the size of the high-level layers and enhances the features-expression ability of deep CNN network model.In the process of network training,a kind of periodic data shuffle strategy is designed to improve the convergence rate of the F-CNN model to realize better handwritten digits recognition.Experiments results on the public MNIST dataset show that the proposed F-CNN model has 99.10% recognition rate for handwritten digits,which outperforms the SN model and the B-CNN model.

Key words: handwritten digits, fused model, Convolutional Neural Network(CNN), data scrambling strategy, convergence rate

中图分类号: 

• TP391.41