基于改进的VGG-Net的手写蒙古文字元识别

Mongolian Character Recognition Based on VGG-NET

现有蒙古文文字存在元样本尺寸较小、使用标准VGG-Net进行字元识别不能达到较好效果的问题,提出基于改进VGG-Net的蒙古文字元识别方法。通过改进的VGG-Net模型对手写蒙古文字元图像数据增强后的23类手写蒙古文字元样本进行识别。结果表明,改进的VGG-Net对字元图像识别的准确率达到96.83%,相比传统VGG-Net模型识别速度更快,准确率更高,占用储存空间更少。

With regard to the problem of the smaller metadata sample size existing in Mongol character and dissatisfactory results by using standard VGG-Net for character recognition,a Mongolian character recognition method based on improved VGG-Net was proposed in the paper.Using the improved VGG-Net model to recognize 23 types of handwritten Mongol text meta samples after data enhancement showed that the improved VGG-Net was accurate(96.83%)for character image recognition;Compared with the traditional VGG-Net model,its recognition speed was faster,less memory space occupation and the accuracy rate was increased by 5.5%.