作者：白雪杰1, 廉飞宇1,2, 付麦霞1,2
作者单位：1. 河南工业大学信息科学与工程学院，河南 郑州 450001;
2. 河南工业大学 粮食信息处理与控制教育部重点实验室，河南 郑州 450001
关键词：贝壳;太赫兹时域光谱系统;主成分分析;支持向量机;多维特征融合
摘要：
传统的贝壳检测方法对新旧贝壳的分类精度低，样品破坏程度大，稳定性差，而太赫兹光谱又缺乏可直接人工分辨的特征。为此，提出一种基于太赫兹时域光谱和机器学习的新旧贝壳识别方法。首先采用太赫兹时域光谱系统(THz－TDS)技术，研究新旧贝壳太赫兹时域光谱、频域光谱、折射率谱和吸收谱特性。然后使用主成分分析法(PCA)在满足所有主成分的累计贡献率达到80%以上的原则的前提下，提取光谱的特征数据。4种光谱分别提取4、4、5和4个主成分，最后使用Adaboost对主成分进行多维特征融合，将融合后的主成分作为支持向量机(SVM)模型的输入用于识别新旧贝壳的种类，其中通过3种核函数(Linear，Polynomial，Radial Basis Function)的对比分析，选出最佳核函数为Radial Basis Function。结果表明：在使用Radial Basis Function核函数，参数C为2.1、 $ \sigma $为4.4的情况下，PCA—Adaboost—SVM模型对新旧贝壳识别准确率可达到98%。通过与BP神经网络、偏最小二乘回归法(PLS)和主成分回归分析(PCR)方法的比较，PCA—Adaboost—SVM方法具有更高的准确性和更稳定的性能，同时也说明采用太赫兹时域光谱系统技术结合机器学习方法可以精准鉴别新旧贝壳种类。
Recognition of old and new shells based on terahertz time-domain spectroscopy and machine learning
BAI Xuejie1, LIAN Feiyu1,2, FU Maixia1,2
1. College of Information Science and Engineering, Henan University of Technology, Zhengzhou 450001, China;
2. Ministry of Education Key Laboratory of Grain Information Technology & Control, Henan University of Technology, Zhengzhou 450001, China
Abstract: The classification of old and new shells by traditional methods is inaccurate and unstable, which will cause damage to the sample. In addition, the terahertz spectrum lacks features that can be directly resolved manually. Therefore, a new shell recognition method based on terahertz time-domain spectroscopy and machine learning is proposed. Firstly, the terahertz time-domain spectroscopy system (THz-TDS) is adopted to study the terahertz time domain spectrum, frequency domain spectrum, refractive index spectrum and absorption spectrum characteristics of new and old shells. Secondly, spectral characteristic data were obtained by using principal component analysis (PCA) and this process accord with the requirement that all principal components contain more than 80% of the original data. Four kinds of spectrum can be used to extract four kinds of principal components, whose number is respectively 4, 4, 5 and 4. Finally, Adaboost was used to fuse the principal components and support vector machine (SVM) model was used to identify the new and old shell species. Through the comparative analysis of three kinds of kernel functions (Linear, Polynomial and Radial Basis Function), the Radial Basis Function has the best results and can be selected as the optimal kernel function. The results show that the PCA-Adaboost-SVM model has a 98% correct recognition rate of old and new shells when the Radial Basis Function kernel is used and the parameter C is 2.1 and σ is 4.4. PCA-Adaboost-SVM has higher accuracy and more stable performance than BP neural network, partial least squares regression (PLS) and principal component regression (PCR).In addition, terahertz time-domain spectroscopy combined with machine learning is feasible to identify new and old shell species.
Keywords: shell;terahertz time-domain spectroscopy;PCA;SVM;multidimensional feature fusion
2022, 48(12):172-180 收稿日期: 2022-06-15;收到修改稿日期: 2022-07-15
基金项目: 粮食信息处理与控制教育部重点实验室开放基金（KFJJ-2021-103）；河南省高等学校重点科研项目（22A510014）
作者简介: 白雪杰(1999-),男,河北张家口市人,专业方向为电子信息工程
参考文献
[1] 陈稚, 陈波, 揭新明, 等. 邻苯二甲醛柱前衍生反相高效液相色谱法检测珍珠粉中的氨基酸含量[J]. 时珍国医国药, 2007(7): 1680-1681
[2] 夏静芬, 钱国英, 陈亮, 等. 傅里叶变换红外光谱法对珍珠粉和贝壳粉的研究[J]. 光谱实验室, 2010, 27(2): 524-528
[3] 曾跃武, 吕光烈, 廖杰, 等. 珍珠粉与蚌壳粉的鉴别[J]. 材料科学与工程学报, 2012, 30(6): 937-939
[4] 蒲月华, 何锦锋, 高振声, 等. 珍珠粉与珍珠层粉微量元素的对比研究[J]. 食品研究与开发, 2016, 37(16): 125-128
[5] 乔艺涵, 索亚然, 孟雪丹, 等. 中药珍珠层粉X射线衍射指纹图谱研究[J]. 药物分析杂志, 2019, 39(5): 911-918
[6] DUVILLARET L, GARET F, COUTAZ J L. A reliable method for extraction of material parameters in terahertz time-domain spectroscopy[J]. IEEE Journal of Selected Topics in Quantum Electronics, 2002, 2(3): 739-746
[7] HAN P Y, TANI M, USAMI M, et al. A direct comparison between terahertz time-domain spectroscopy and far-infrared Fourier transform spectroscopy[J]. Journal of Applied Physics, 2001, 89(4): 2357-2359
[8] WALTHER M, PLOCHOCKA P, FISCHER B M, et al. Collective vibrational modes in biological molecules investigated by terahertz time-domain spectroscopy[J]. Biopolymers, 2010, 67(4-5): 310-313
[9] ABDI H, WILLIAMS L J. Principal component analysis[J]. Wiley Interdisciplinary Reviews Computational Statistics, 2010, 2(4): 433-459
[10] LIU R X, KUANG J, GONG Q, et al. Principal component regression analysis with SPSS[J]. Computer Methods and Programs in Biomedicine, 2003, 71(2): 141-147
[11] ZHANG L, ZHOU W, JIAO L. Wavelet support vector machine[J]. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 2004, 34(1): 34-39
[12] MERLER S, CAPRILE B, FURLANELLO C. Parallelizing AdaBoost by weights dynamics[J]. Computational Statistics & Data Analysis, 2007, 51(5): 2487-2498
[13] 段国腾, 李林, 崔海林, 等. 全光纤耦合异步光学采样THz-TDS系统[J]. 北京理工大学学报, 2016, 36(2): 170-174
[14] 董家蒙, 彭晓昱, 马晓辉, 等. 超宽带太赫兹时域光谱探测技术研究进展[J]. 光谱学与光谱分析, 2016, 36(5): 1277-1283
[15] 郑甲宏, 赵敬超. 一种基于PCA-BP的直升机起落架着舰载荷评估方法[J]. 中国测试, 2021, 47(5): 156-161
[16] 文大鹏, 梁西银, 苏茂根, 等. 激光诱导击穿光谱技术结合PCA-PSO-SVM对矿石分类识别[J]. 激光与光电子学进展, 2021, 58(23): 191-199
[17] 耿俊成, 郭志民, 李晓蕾, 等. 基于LOF和SVM的配电网线变关系数据校验方法[J]. 中国测试, 2021, 47(4): 49-54
[18] 徐华晟, 李超, 方广有. 基于多维特征融合的太赫兹隐蔽目标检测方法研究[J]. 电子测量技术, 2021, 44(2): 82-86
[19] 王宽, 李盘文, 祁晓鹏. 特征信息加权融合的进气道斜板故障诊断[J]. 中国测试, 2020, 46(11): 43-47+58
[20] HAO Z, WANG Z, BAI D, et al. Towards the steel plate defect detection: multidimensional feature information extraction and fusion[J]. Concurrency and Computation: Practice and Experience, 2021, 33(21): 6384
[21] LI X, WANG L, SUNG E. AdaBoost with SVM-based component classifiers[J]. Engineering Applications of Artificial Intelligence, 2008, 21(5): 785-795
[22] HU L, CUI J. Digital image recognition based on Fractional-order-PCA-SVM coupling algorithm[J]. Measurement, 2019, 145: 150-159
[23] 谈爱玲, 楚振原, 王晓斯, 等. 基于拉曼光谱和DCGAN数据增强的珍珠粉掺伪检测研究[J]. 光谱学与光谱分析, 2022, 42(3): 769-775
