基于ResNet的象棋文字识别研究  

Research on Chess Text Recognition Based on ResNet

作　　者：翟乃强[1] ZHAI Naiqiang(Vocational Education School,Qingdao Ocean Shipping Mariners College,Qingdao 266427,China)

机构地区：[1]青岛远洋船员职业学院职教分院,山东青岛266427

出　　处：《青岛远洋船员职业学院学报》2024年第4期26-29,共4页Journal of Qingdao Ocean Shipping Mariners College

基　　金：2024年度青岛远洋船员职业学院课题:“基于现有监控系统的火情预警系统研发”。

摘　　要：在象棋文字识别中存在的主要难点是字体差异及文字具有旋转角度。其中具有旋转角度的文字最是识别的难点。本文基于ResNet原理构造了深度学习网络,为提高文字识别的精度,将图片进行灰度化处理,并计算训练数据集的均值和方差用于规范化图片数据。研究结果表明,相较于彩色图片形式,灰度图及规范化图片数据的形式在象棋文字的识别方面表现出更好的效果,在识别精度方面提高了6.7个百分点,验证了自定义ResNet深度学习网络在象棋文字识别方面的有效性。The main difficulties in recognizing chess characters are font differences and the rotation angle of the characters,among which the characters with rotation angles are the difficulties in recognition.This article constructs a deep learning network based on the ResNet principle.In order to improve the accuracy of text recognition,images are grayscale processed and the mean and variance of the training dataset are calculated to normalize image data.The research results show that compared to the form of color images,the form of grayscale images and standardized image data shows better performance in the recognition of chess text,improving recognition accuracy by 6.7 percentage points,verifying the effectiveness of the custom ResNet deep learning network in chess text recognition.

关 键 词：残差网络 灰度图 规范化 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]