改进MobileNetV3的脱机手写汉字识别  

Offline handwritten Chinese character recognition based on improved MobileNetV3

作　　者：程若然 周浩军 刘露露 贺炎 CHENG Ruoran;ZHOU Haojun;LIU Lulu;HE Yan(School of Electrical and Electronic Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)

机构地区：[1]上海工程技术大学电子电气工程学院,上海201620

出　　处：《智能计算机与应用》2022年第7期160-164,共5页Intelligent Computer and Applications

摘　　要：针对目前手写识别网络训练时间长、高资源消耗等问题,本文提出了一种脱机手写汉字识别网络模型。使用轻量级网络MobileNetV3作为主干网络,以减少网络参数量;针对汉字识别分类数庞大的特点,使用多尺度卷积核,提取更丰富的特征信息;针对形近字易产生识别错误的问题,使用注意力机制进行局部、全局特征提取并融合。实验结果表明,所提模型能在保持较少参数量的情况下,使其识别准确率有所提升。To solve the problem of long training time and high resource consumption of handwritten Chinese character recognition network, an offline handwritten Chinese character recognition network model is proposed in this paper. The lightweight network MobileNetV3 is used as the backbone network to reduce the number of network parameters. In view of the large number of Chinese character recognition classification, multi-scale convolution kernel is used to extract richer feature information. In view of the problem that Chinese characters with similar shapes are easy to be recognized incorrectly, an attention mechanism is adopted to extract local features and global features and fuse them. Experimental results show that the proposed model can improve the accuracy while keeping fewer parameters.

关 键 词：脱机手写汉字识别 深度学习 MobileNetV3 特征融合 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]