基于形变卷积神经网络的手写体数字识别研究  

Handwritten Digital Recognition Based on Deformable Convolutional Neural Network

作　　者：茹晓青 华国光 李丽宏[1] 李莉 RU Xiao-qing;HUA Guo-guang;LI Li-hong;LI Li(School of information and electrical engineering. Hebei University of Engineering, Handan 056000, China)

机构地区：[1]河北工程大学信息与电气工程学院,河北邯郸056000

出　　处：《微电子学与计算机》2019年第4期47-51,共5页Microelectronics & Computer

基　　金：河北省自然科学基金(sF2015402150);河北省教育厅资助项目(ZD2015087);邯郸市科学技术研究与发展计划项目(1721203049-1)

摘　　要：本文引入形变卷积模块来增强网络对数字几何变换的建模能力,提出了一种基于改进的形变卷积神经网络手写体数字识别框架,在提高识别精度的同时,还有效的减少了训练的参数量,提高识别速度.本文在手写体数据集及变换后的数据集中进行验证.实验结果的分析以及与相应算法的比较,证明了本算法是有效的.In this paper, the deformable convolution module is introduced to enhance the modeling ability of the network to digital geometric transformation, and an improved handwritten digital recognition framework based on deformable CNN is proposed. In addition to improving the recognition accuracy, the framework can effectively reduce the training parameters and improve the recognition speed. This paper demonstrate state-of-the-art performance competing methods on the handwritten dataset and the transformed dataset.

关 键 词：手写体数字识别 卷积神经网络 形变卷积 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]