基于核主成分分析的GSA-SVM木材单板缺陷识别研究  

Study on GSA-SVM Wood Veneer Defect Identification Based on Kernel Principal Component Analysis

作　　者：贺春光[1] 李璐芳 高峰[2] 袁云梅 高凡 丁安宁 多化琼[1] HE Chunguang;LI Lufang;GAO Feng;YUAN Yunmei;GAO Fan;DING Anning;DUO Huaqiong(College of Materials Science and Art Design,Inner Mongolia Agricultural University,Hohhot 010018,China;Vocational and Technical College,Inner Mongolia Agricultural University,Baotou 014109,China;School of Information Engineering,Shanxi College of Applied Science and Technology,Taiyuan 030000,China)

机构地区：[1]内蒙古农业大学材料科学与艺术设计学院,呼和浩特010018 [2]内蒙古农业大学职业技术学院,内蒙古包头014109 [3]山西应用科技学院信息工程学院,太原030000

出　　处：《森林工程》2023年第2期91-99,共9页Forest Engineering

基　　金：内蒙古自治区重点研发和成果转化计划项目(2022YFDZ0031)。

摘　　要：为支持向量机能够准确识别木材单板表面缺陷,以提高木材单板质量,提出高效准确的单板缺陷核主成分分析(KPCA)的引力搜索算法(GSA)-支持向量机(SVM)识别模型。考虑到图像特征数据间的冗余影响,采用KPCA方法对原始特征数据降维,并通过GSA优化SVM的惩罚因子C和核参数g,建立KPCA-GSA-SVM木材单板缺陷识别模型。基于颜色、纹理、形状3方面的特征以活节、死节、裂纹为研究对象的样本原始数据集,选取8个主要特征(1个颜色特征、1个纹理特征和6个形状特征)作为木材单板识别依据,对木材单板识别模型进行学习训练及预测分析,并与传统粒子群参数优化算法(PSO)构成的KPCA-PSO-SVM识别模型进行识别效果对比。结果表明,基于KPCA-GSA-SVM木材单板识别模型对于活节、死节、裂纹的识别率达到100%、96.78%、100%,较KPCA-PSO-SVM识别模型分别高出21.62%、0.63%、7.41%,且整体耗费时间缩短7.26 s,由此看出预测识别率、识别速度、稳定性高于前者。研究结论从新的角度对单板缺陷进行识别,有助于木材单板缺陷的识别发展。In order for the support vector machine to accurately identify the surface defects of wood veneer and improve the quality of wood veneer,an efficient and accurate recognition model of kernel principal component analysis(KPCA)gravity search algorithm-support vector machine(GSA-SVM)for veneer defects was proposed.Considering the redundant effect between image feature data,KPCA method was used to reduce the dimension of original feature data,and GSA optimized the penalty factor C and kernel parameter g of support vector machine(SVM)to establish KPCA-GSA-SVM wood veneer defect recognition model.Based on the three features of color,texture and shape,the raw data set of samples with live knots,dead knots and cracks as the research objects,8 main features(1 color feature,1 texture feature and 6 shape features)were selected as the basis for wood veneer recognition.The wood veneer identification model was learned,trained,predicted and analyzed,and the identification effect was compared with the KPCA-PSO-SVM identification model composed of the traditional particle swarm parameter optimization algorithm(PSO).The results showed that the recognition rate of live knots,dead knots and cracks of the wood veneer recognition model based on KPCA-GSA-SVM was 100%,96.78%and 100%,which were 21.62%,0.63%and 7.41%higher than that of KPCA-PSO-SVM,and the overall time was shortened by 7.26 s,it can be seen that the prediction recognition rate,recognition speed and stability were higher than the former.The research conclusions identify the veneer defects from a new perspective,which is helpful for the identification and development of wood veneer defects.

关 键 词：单板缺陷 引力搜索算法 支持向量机 核主成分分析 识别模型 

分 类 号：S784[农业科学—林学]