双树复小波特征融合的板材压缩感知协同检测与分选

提出一种对板材表面缺陷和纹理进行协同快速准确检测的算法. 根据双树复小波所特有的方向性和时移不变性,研究了板材表面图像的双树复小波特征提取及融合算法,对板材表面图像进行3级双树复小波分解得到40个特征向量,并通过遗传算法优选出23个关键特征,优选后的特征能够较为完整地表达板材图像的复杂信息并减小数据冗余. 最后采用压缩感知理论,将优选后的特征向量作为样本矩阵列,构造出训练样本数据字典,通过最小残差完成对板材表面信息的分类识别. 实验对木材表面存在的弦切纹、径切纹、活结和死结等4类柞木样本进行了检测,正确率分别为91. 8%、100%、96. 4%和91. 8%,该算法能够以95%的平均识别率完...

Full description

Saved in:
Bibliographic Details
Published in电机与控制学报 Vol. 19; no. 8; pp. 81 - 87
Main Author 李超 张怡卓 于慧伶 曹军
Format Journal Article
LanguageChinese
Published 东北林业大学 机械工程博士后流动站 黑龙江 哈尔滨150040%东北林业大学 机电工程学院 黑龙江 哈尔滨150040%东北林业大学 信息与计算机工程学院 黑龙江 哈尔滨150040 2015
东北林业大学 机电工程学院 黑龙江 哈尔滨150040
Subjects
Online AccessGet full text
ISSN1007-449X
DOI10.15938/j.emc.2015.08.012

Cover

More Information
Summary:提出一种对板材表面缺陷和纹理进行协同快速准确检测的算法. 根据双树复小波所特有的方向性和时移不变性,研究了板材表面图像的双树复小波特征提取及融合算法,对板材表面图像进行3级双树复小波分解得到40个特征向量,并通过遗传算法优选出23个关键特征,优选后的特征能够较为完整地表达板材图像的复杂信息并减小数据冗余. 最后采用压缩感知理论,将优选后的特征向量作为样本矩阵列,构造出训练样本数据字典,通过最小残差完成对板材表面信息的分类识别. 实验对木材表面存在的弦切纹、径切纹、活结和死结等4类柞木样本进行了检测,正确率分别为91. 8%、100%、96. 4%和91. 8%,该算法能够以95%的平均识别率完成板材表面缺陷、纹理的协同检测.
Bibliography:LI Chao, ZHANG Yi-zhuo, YU Hui-ling, CAO Jun(1. College of Electromechanical Engineering, Northeast Forestry University, Harbin 150040, China; 2. Post-Doctoral Mobile Station of Mechanics, Northeast Forestry University, Harbin 150040, China ;3. College of Information and Computer Engineering, Northeast Forestry University, Harbin 150040, China)
23-1408/TM
A quick and accurate collaborative classification method for wood defects and texture was pro-posed. As dual-tree complex wavelet has the advantages of approximate shift invariance and good direc-tional selectivity, dual-tree complex wavelet feature was extracted from wood board image and the fusion method was discussed. Three-level dual-tree complex wavelet decomposition was carried out to the surface image and 40 features were got, then genetic algorithm ( GA) was used for feature selection and 23 fea-tures were chosen. Feature fusion can better express the surface information and meanwhile heavily re-duce the data redundancy. Finally, wood surface classific
ISSN:1007-449X
DOI:10.15938/j.emc.2015.08.012