基于波峰波谷特征提取技术的檀香紫檀光谱识别  

Spectral Recognition of Sandalwood Based on Peak and Valley Feature Extraction Technique

作　　者：庄鹏燕 牛佳顺 程俊 卢静宜 孙建平 何拓 ZHUANG Peng-yan;NIU Jia-shun;CHENG Jun;LU Jing-yi;SUN Jian-ping;HE Tuo(School of Resources,Environment and Materials,Guangxi University,Nanning 530004,China;Wildlife Conservation Monitoring Center,National Forestry and Grassland Administration,Beijing 100714,China;Guangxi Minzu Normal University,Art College(Southwest Guangxi High-end Home Design Industry College),Nanning 532200,China)

机构地区：[1]广西大学资源环境与材料学院,广西南宁530004 [2]国家林业和草原局野生动物保护监测中心,北京100714 [3]广西民族师范学院艺术学院(桂西南高端家居设计产业学院),广西南宁532200

出　　处：《光谱学与光谱分析》2024年第12期3463-3472,共10页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(32260359);贵科中引专项(贵科中引专项2022-);桂西南高端家居设计产业学院平台项目(2021-304475)资助。

摘　　要：伴随着经济水平的飞速增长,人们对红木产品的需求日益增大,而檀香紫檀作为一种珍贵红木因其在精神层面上符合人们的追捧且价格昂贵,具有巨大的商业价值而被不法分子替代和假冒。为维护木材市场的秩序和消费者的利益、实现对檀香紫檀的快速无损检测以及鉴别,有必要建立一种快速、可靠的檀香紫檀木材智能识别方法。采用近红外光谱(NIR)分析技术提取檀香紫檀及其相似血檀的光谱信息,利用定性分析方法、偏最小二乘判别分析(PLS-DA)和误差反向传播人工神经网络(BPNN)对光谱信息建立校正模型,进而对檀香紫檀及其相似木材血檀进行识别;通过分析比较这三种模型对这两类木材识别的优缺点和识别准确率,验证该方法对檀香紫檀和血檀识别的可行性。研究发现三种判别模型均能够快速识别木材光谱图像,具有对檀香紫檀和血檀进行快速无损分类识别的能力,在选取不同图像处理方法的情况下,三种判别模型显示的结果各不相同,识别结果也存在差异。进一步分析表明,在使用BPNN模型对光谱数据进行建模时,通过预处理将原始光谱数据波长范围866~2533 nm的波峰波谷特征值作为输入,当输入层节点数为24个特征值,隐含层13个神经元时,模型的均方根误差最小,准确率达到96.43%。此外,缩短光谱范围并不会提高模型的识别率,而在三种模型中,全波段范围的BPNN模型具有最高的识别率。实验结果表明,基于人工神经网络模型结合NIR特征提取技术识别檀香紫檀木材有着较高的识别准确率,其效果相较于前人有一定的提升。该研究有利于减少人工识别的主观性,使用计算机更能够缩短识别时间,较高的准确率可以帮助维护木材市场的秩序和消费者权益。同时,这也为实现檀香紫檀智能视觉识别提供了参考,为红木产业的可持续发展提供技术支持。With the rapid growth of the economic level,people's demand for mahogany products is increasing day by day.As precious mahogany,criminals replace sandalwood and counterfeit it because of its spiritual pursuit and high price.To maintain the order of the timber market and the interests of consumers and realize the rapid non-destructive detection and identification of sandalwood,it is necessary to establish a fast and reliable intelligent identification method of sandalwood.This paper used near-infrared spectroscopy to extract the spectral information of Pterocarpus santalinus and its similar blood sandalwood.The qualitative analysis method,partial least squares discriminant analysis(PLS-DA),and error back propagation artificial neural network(BPNN)were used to establish the calibration model of spectral information.Then,Pterocarpus santalinus and its similar wood blood sandalwood were identified.By analyzing and comparing the advantages and disadvantages and recognition accuracy of these three models for these two kinds of wood recognition,the feasibility of this method in the recognition of sandalwood and sandalwood is verified.The experimental results show that the three discriminant models can quickly identify the wood spectral images,and can quickly and non-destructively classify and identify sandalwood and sandalwood.In the case of selecting different image processing methods,the results of the three discriminant models are different,and the recognition results are also different.Further analysis reveals that when the spectral data are modeled using the BPNN model,the peak and trough eigenvalues of the original spectral data within the wavelength range of 866~2533 nm are utilized as input following preprocessing.It is observed that when the number of input layer nodes is Setto 24 eigenvalues and the hidden layer consists of 13 neurons,the model achieves the smallest root mean square error,with an accuracy rate of 96.43%.Additionally,shortening the spectral range does not result in an improvement in the model's

关 键 词：近红外光谱 木材识别 PLS-DA BPNN 定性分析 

分 类 号：S781.23[农业科学—木材科学与技术]