文献类型：期刊文献

中文题名：基于主成分分析和支持向量机的木材近红外光谱树种识别研究

英文题名：Research on Near Infrared Spectrum with Principal Component Analysis and Support Vector Machine for Timber Identification

作者：谭念[1] 孙一丹[1] 王学顺[1] 黄安民[2] 谢冰峰[1]

第一作者：谭念

通信作者：Wang, XS[1]

机构：[1]北京林业大学数学系;[2]中国林业科学研究院木材工业研究所

年份：2017

卷号：37

期号：11

起止页码：3370-3374

中文期刊名：光谱学与光谱分析

外文期刊名：Spectroscopy and Spectral Analysis

收录：CSTPCD;;EI(收录号：20181004881451);Scopus(收录号：2-s2.0-85042879909);WOS:【SCI-EXPANDED(收录号:WOS:000416393900010)】；北大核心:【北大核心2014】；CSCD:【CSCD2017_2018】；PubMed;

基金：国家自然科学基金项目(31670564)资助

语种：中文

中文关键词：树种识别;主成分分析;支持向量机;遗传算法;粒子群算法

外文关键词：Timber identification; Principal component analysis; Support vector machine; Genetic algorithm; Particle swarm optimization

分类号：O657.3

摘要：为了探究一种新型高效的树种鉴别方法,以桉木、杉木、落叶松、马尾松和樟子松近红外光谱数据为研究对象,分别建立了基于主成分分析和支持向量机的木材树种定性识别模型。在主成分识别模型中,样本光谱数据经过预处理后绘制了其二维和三维主成分得分图,可以看出:主成分分析得分图能有效区分五种木材树种,且三维得分图比二维得分图更能直观、清晰展示树种之间的差异,表明主成分分析在可视化层面上可对小样本树种进行有效判别。在支持向量机识别模型中,分别建立了以遗传算法和粒子群算法为代表的智能算法优化支持向量机树种识别模型,结果显示,遗传算法-支持向量机模型的交叉验证最佳判别准确率为95.71%,测试集预测准确率为94.29%,算法用时134.08s;粒子群算法-支持向量机模型的交叉验证最佳判别准确率为94.29%,测试集预测准确率为100.00%,算法用时19.98s,表明基于智能算法支持向量机树种识别模型能够实现对木材树种的有效鉴别。该研究对近红外光谱分析技术在木材科学领域的应用进行了有益探索,为木材树种的快速识别提供了新方法。
In order to explore an efficient method of timber species identification, the near-infrared spectral data of the eucalyptus, the Chinese fir, the larch, the Pinus massoniana and the Pinus sylvestris were selected as the research object. The qualitative identification model of timber species based on principal component analysis and support vector machine were established respectively. In the principal component analysis identification model, the 2D and 3D principal component analysis scores were drawn after preprocessing the sample spectral data. It is found that five kinds of timber species can be distinguished effectively in the principal component analysis score scatter plots, and the 3D principal component analysis score scatter plot shows the difference between the timber species more intuitively and clearly than the 2D principal component analysis score scatter plot. It is shown that the principal component analysis can distinguish the small sample timber species at the visual level. In the support vector machine identification model, the methods of genetic algorithm and particle swarm optimization were selected respectively for parameter optimization. Results showed that, the best discrimination accuracy of cross-validation was 95.71%, and the prediction accuracy rate of test set was 94. 29% in the genetic algorithm-support vector machine model, which cost 134.08 s. While in the particle swarm optimization-support vector machine model, the best discrimination accuracy of cross-validation was 94. 29%, and the prediction accuracy rate of test set was 100.00%, which cost 19.98 s. It indicates that the model based on intelligent algorithm and support vector machine can effectively identify the timber species. This study has made a useful exploration of the application of near infrared spectroscopy in the wood science, and provided a new method for rapid identification of timber species.