[1]许鸿奎,*,邵星,等.基于堆栈自编码的刻划字符检测研究[J].山东建筑大学学报,2018,(05):24-30.[doi:10.12077/sdjz.2018.05.004]
 Xu Hongkui,*,Shao Xing,et al.Study on recognition of carving characters based on stacked auto-encoder[J].Journal of Shandong jianzhu university,2018,(05):24-30.[doi:10.12077/sdjz.2018.05.004]
点击复制

基于堆栈自编码的刻划字符检测研究()
分享到:

《山东建筑大学学报》[ISSN:1673-7644/CN:37-1449/TU]

卷:
期数:
2018年05期
页码:
24-30
栏目:
研究论文
出版日期:
2018-10-09

文章信息/Info

Title:
Study on recognition of carving characters based on stacked auto-encoder
文章编号:
1673-7644(2018)05-0024-07
作者:
许鸿奎12*邵星1韩晓1宫淑兰12王兆斌1
(1.山东建筑大学 信息与电气工程学院,山东 济南 250101;2.山东省智能建筑技术重点实验室,山东 济南 250101)
Author(s):
Xu Hongkui12* Shao Xing1 Han Xiao1 et al.
(1.School of Information and Electrical Engineering, Shandong Jianzhu University, Jinan 250101, China; 2.Shandong Provincial Key Laboratory of Intelligent Buildings Technology, Jinan 250101, China)
关键词:
刻划字符堆栈自编码工业检测特征提取
Keywords:
carving characters stacked auto-encoder industrial inspection feature extraction
分类号:
TU996
DOI:
10.12077/sdjz.2018.05.004
文献标志码:
A
摘要:
刻划字符在工业生产中具有广泛的应用,对于工业中大量生产的产品,由于其无色差性和立体性使得人工检查费时费力,研究刻划字符检测是工业生产自动化检测提高的重要基础。文章针对前向高角度环形光照明方案下采集的刻划字符图像,建立了一种基于深层堆栈自编码的刻划字符检测模型,并对模型中所需参数进行了配置,对比分析了堆栈欠完备自编码、变分自编码、HOG算子的特征提取效果,将提取的特征在K近邻、支持向量机和BP神经网络等3种分类器下的检测效果。结果表明:堆栈欠完备自编码器提取的特征有利于字符缺陷检测;堆栈变分自编码器提取的特征有利于字符识别检测;在堆栈自编码器提取的特征下,K近邻分类器的效果最好。
Abstract:
Characterizations are widely used in industrial production. For a large number of products in industry, the study of character detection is an important basis for automation detection in industrial production because of its colorless difference and stereotyping. In this paper, a characterization model based on the deep stack selfencoding is established for the character image collected under the forward high angle ring light illumination scheme. The parameters needed in the model are configured. And the under complete self-coding, variational self-coding and HOG operator are compared and analyzed. The feature extraction effect and the extracted features are detected under 3 classifiers, namely K nearest neighbor, support vector machine and BP neural network. The results show that the features extracted from the stack under complete selfencoder are beneficial to the character defect detection, and the feature extraction from the stack selfencoder is beneficial to the character recognition and detection. With the feature extracted from the stack, the K nearest neighbor classifier has the best effect.

参考文献/References:

[1]方承志,周品,付世清.基于小波核LS-SVM的车牌字符识别算法研究\[J\].计算机技术与发展,2015(3):86-90.[2]刘明珠,郑云非,樊金斐,等.基于深度学习法的视频文本区域定位与识别\[J\].哈尔滨理工大学学报,2016,21(6):61-66.[3]宋春静,丁香乾,郭乙运.基于深度信念网络的集装箱字符识别方法\[J\].计算机工程与设计,2016,37(3):742-745.[4]李国平,路长厚,李健美,等.阴影莫尔法中光栅传感器的研究\[J\].传感技术学报,2007,20(3):550-553.[5]李学勇,路长厚,李建美.融合轮廓矩和Fourier描述子特征的压印字符识别\[J\].光电子·激光,2007,18(10):1244-1247,1259.[6]李建美,路长厚,李国平.基于Gabor变换的凹凸字符图像特征抽取新方法\[J\].系统仿真学报,2008,20(8):2133-2136.[7]王琼.低质量压印字符的分割与识别技术研究\[D\].济南:山东大学,2015.[8]周成伟.基于卷积神经网络的自然场景中数字识别\[J\].计算机技术与发展,2017,27(11):101-105.[9]董峻妃,郑伯川,杨泽静.基于卷积神经网络的车牌字符识别\[J\].计算机应用,2017,37(7):2014-2018.[10]赵朋成,冯玉田,涂云轩.基于高倍特征深度残差网络的手写数字识别\[J\].电子测量技术,2018,6(41):86-89.[11]Erhan D, Bengio Y, Courville A,et al. Why does unsupervised pretraining help deep learning\[J\]. Journal of Machine Learning Research,2010,11:625-660.[12]Sch-lkopf B, Platt J, Hofmann T. Greedy layerwise training of deep networks\[J\]. Advance in Neural Information Processing Systems,2007,19:153-160.[13]Hosseini-Asl E, Zurada J M, Nasraoui O. Deep learning of partbased representation of data using sparse autoencoders with nonnegativity constraints\[J\]. IEEE Transactions on Neural Networks & Learning Systems,2016,27(12):2486-2498.[14]Olshausen B A,Field D J.Sparse coding with an overcomplete basis set: A strategy employed by V1\[J\]. Vision Research,1997,37(23):3311-3325.[15]Vu D, Aitkin M. Variational algorithms for biclustering models\[J\]. Computational Statistics & Data Analysis,2015,89:12-24.[16]李俊.机器视觉照明光源关键技术研究\[D\].天津:天津理工大学,2007.

备注/Memo

备注/Memo:
收稿日期:2018-09-08基金项目:山东省科技发展计划(2014GGX101050);山东省自然科学基金项目(ZR2014FM016)作者简介:许鸿奎(1966-),男,副教授,博士,主要从事信号与信息处理相关理论等方面的研究.E-mail:xhkui2009@163.com\[*通讯作者\]
更新日期/Last Update: 2018-08-08