本文已在中国知网网络首发,可在知网搜索、下载并阅读全文。
基于连续投影算法-遗传算法-BP神经网络的可见/近红外光谱木材识别
doi: 10.11833/j.issn.2095-0756.20210377
Visible/near infrared spectrum wood identification based on SPA-GA-BP neural network
-
摘要:
目的 基于可见/近红外光谱技术,以10种木材为研究对象,探索不同预处理和特征提取方法下BP神经网络识别木材的效果。 方法 利用美国ASD公司生产的LabSpec 5000光谱仪采集10种木材的光谱图,分别进行移动平均法处理、移动平均法+多元散射校正(MSC)、移动平均法+标准正态变量变换(SNV)、Savitzky-Golay卷积平滑算法(S-G滤波器)、S-G滤波器+MSC和S-G滤波器+SNV的预处理,运用主成分分析法(PCA)、连续投影算法(SPA)、SPA和遗传算法(GA)联合分别进行特征提取,将提取的特征结合BP神经网络进行木材识别试验。 结果 以SPA和GA联合提取光谱特征时,移动平均法+SNV的预处理效果最佳,以吸收峰为起始波段(Winitial=1 445 nm)、吸收峰个数为特征个数(Ntot=9)时,识别率较高,特征个数大部分减少为SPA提取特征值个数的一半左右。BP神经网络的平均识别速度提升明显。10种木材的平均识别率为98.0%,其中7种木材的识别率达到了100.0%。 结论 在移动平均法+SNV的预处理下,SPA和GA联合提取光谱图的特征,既可提高BP神经网络识别木材的正确率,又可提升识别速度。图3表6参23 Abstract:Objective The purpose of this study is to explore the effect of BP neural network identification under different pretreatment and feature extraction methods based on visible/near infrared spectroscopy technology, with 10 wood species as objects. Method The LabSpec 5000 spectrometer produced by American ASD company was used to collect the spectrograms of 10 species of wood, which were pretreated by moving average method, moving average method + multiplicative scattering correction(MSC), average method+standard normal variable transformation (SNV), Savitzky-Golay convolution smoothing algorithm (S-G filter), S-G filter+MSC and S-G filter+SNV. Meanwhile, principal component analysis(PCA), successive projections algorithm(SPA), and SPA combined with genetic algorithm(GA) were used for feature extraction respectively. The extracted features were combined with BP neural network for wood identification test. Result When SPA and GA were combined to extract spectral features, the moving average+SNV method had the best preprocessing effect. When absorption peak was used as the initial waveband (Winitial=1 445 nm) and the number of absorption peaks (Ntot=9) as the number of features, the identification rate was high, and the number of features mostly decreased to about 1/2 of the number of feature values extracted by SPA. The average identification speed of BP neural network significantly increased. The average identification rate of the 10 wood species was 98.0%, and the identification rate of 7 of them reached 100.0%. Conclusion Under the pretreatment of moving average method+SNV, the combined use of SPA and GA in spectral feature extraction can improve not only the accuracy of wood identification by BP neural network, but also the identification speed. [Ch, 3 fig. 6 tab. 23 ref.] -
表 1 不同预处理的PCA-BP神经网络识别率
Table 1. PCA-BP neural network recognition with different preprocessing
检测方式 预处理方法 累计贡
献率/%主成分
个数/个平均识
别率/%可见/近红外光谱 对照组 95 12 80.2 移动平均法 95 14 81.4 移动平均法+MSC 95 10 82.1 移动平均法+SNV 95 11 83.5 S-G滤波器 95 12 81.3 S-G滤波器+MSC 95 13 82.9 S-G滤波器+SNV 95 15 84.7 表 2 不同预处理的SPA-BP神经网络平均识别率
Table 2. Average recognition rate of SPA-BP neural network with different pretreatments
预处理方法 平均识
别率/%预处理方法 平均识
别率/%对照组 86.1 S-G滤波器 86.4 移动平均法 87.2 S-G滤波器+MSC 86.8 移动平均法+MSC 86.5 S-G滤波器+SNV 87.3 移动平均法+SNVSNV 88.2 表 3 10种木材吸收峰个数和集中波段
Table 3. Number of absorption peaks and concentrated bands of 10 kinds of wood
木材种类 吸收峰个数/个 集中分布波段/nm 红檀 7 920~970、1 010~1 060、1 210~1 260、1 570~1 620、1 779~1 829、1 921~1 971、2 122~2 172 大果紫檀 7 930~980、1 020~1 070、1 220~1 270、1 580~1 630、1 780~1 830、1 920~1 970、2 120~2 170 檀香紫檀 7 932~982、1 023~1 073、1 221~1 271、1 568~1 618、1 777~1 827、1 921~1 971、2 123~2 173 刺猬紫檀 9 763~813、1 222~1 272、1 308~1 358、1 461~1 511、1 548~1 598、1 760~1 810、1 931~1 981、
2 092~2 142、2 211~2 261巴里黄檀 9 765~815、1 221~1 271、1 307~1 357、1 466~1 516、1 545~1 595、1 769~1 819、1 930~1 980、
2 087~2 137、2 219~2 269红檀香 9 753~803、1 223~1 273、1 309~1 359、1 463~1 513、1 558~1 608、1 771~1 821、1 932~1 982、
2 092~2 142、2 212~2 262破布木 9 763~813、1 222~1 272、1 317~1 367、1 463~1 513、1 551~1 601、1 772~1 822、1 933~1 983、
2097~2147、2214~2264木犀科豆瓣香 9 766~816、1230~1280、1317~1367、1468~1518、1554~1604、1775~1825、1940~1990、
2095~2145、2 216~2 266中美洲黄檀 9 753~803、1 218~1 268、1 305~1 355、1 457~1 507、1 544~1 594、1 769~1 819、1 928~1 978、
2 084~2 134、2 209~2 259黑檀 9 881~931、1 218~1 268、1 305~1 355、1 452~1 502、1 557~1 607、1 772~1 822、1 923~1 973、
2 092~2 142、2 218~2 268表 4 不同起始波段的SPA-BP神经网络平均识别率
Table 4. Average recognition rate of SPA-BP neural network with different starting bands
特征值数/个 起始波段/nm 10种木材提取特征波长分布/nm 平均识别率/% 10 895 364~368、2 141~2 144;402~410;418~426;324、2 135~2 142;375~383;432~440;400~408;476~484;420~428;1 452~1 460 89.7 10 1 445 478~586;410~418;423~431;500~508;405~413;436~444;418~426;693~701;891~899;888~896 90.4 10 1 605 133~135、2 137~2 142;891~899;891~899;2 135~2 142、2 132;419~427;819~827;420~428;446~454;892~990;893~901 90.1 10 15 2133~135、2 137~2 142;2 133~2 135、2 137~2 142;408~416;292、22 135~2 142;375~383;430~438;414~422;461~469;420~428;890~898 88.3 10 795 61~64、2 139~2 143;405~413;420~428;326、2 135~2 142;378~386;527~
535;403~411;478~486;420~422、1 453~1 458;1 350~1 35889.5 10 995 203~209、2 141~2 142;399~407;418~426;349~352、2 138~2 142;3381~389;434~442;421~429;485~493;527~535;1 452、1 454~ 1458、1 461~1 463 89.2 10 1 350 82~90;891~899;434~442;519~527;416~424;886~894;420~428;694~702;891~899;888~896 88.9 10 1 950 13、2 135~2 142;379~387;407~415;281、2 135~2 142;293~301;428~436;
1 058~1 066;450~458;413、1 452~1 459;1 452~1 46088.6 表 5 同一起始波段不同特征波段的SPA-BP神经网络平均识别率
Table 5. Average recognition rate of SPA-BP neural network with the same starting band and different characteristic bands
起始波
段/nm特征值
数/个平均识
别率/%起始波
段/nm特征值
数/个平均识
别率/%1 445 5 92.3 1 445 10 90.6 1 445 7 93.0 1 445 20 92.7 1 445 9 93.2 1 445 25 91.2 1 445 8 91.6 表 6 同一预处理方式10种木材的SPA-BP神经网络平均识别率
Table 6. Average recognition rate of SPA-BP neural network for 10 kinds of wood with the same pretreatment method
木材种类 平均识
别率/%木材种类 平均识
别率/%木材种类 平均识
别率/%红檀 90.9 巴里黄檀 94.2 中美洲黄檀 100.0 大果紫檀 100.0 红檀香 100.0 黑檀 100.0 檀香紫檀 90.7 破布木 94.6 平均 95.7 刺猬紫檀 95.1 木犀科豆瓣香 91.0 说明:预处理方式为移动平均法+SNV,起始波段为1 445 nm,
特征值数为9个 -
[1] LEE H T, SOON L L, KEVIN K N, et al. DNA extraction from dry wood of Neobalanocarpus heimii (Diperocarpaceae)for forensic DNA profiling and timeber tracking [J]. Wood Sci Technol, 2012, 46(5): 813 − 815. doi: 10.1007/s00226-011-0447-6 [2] 陈利顶, 李秀珍, 傅伯杰, 等. 中国景观生态学发展历程与未来研究重点[J]. 生态学报, 2014, 34(12): 3129 − 3141. CHEN Liding, LI Xiuzhen, FU Bojie, et al. Development history and future research priorities of landscape ecology in China [J]. J Ecol, 2014, 34(12): 3129 − 3141. [3] JIAO Lichao, YIN Yafang, XIAO Fuming, et al. Comparative analysis of two DNA extraction protocols from fresh and dried wood of Cunninghamia lanceolata(Taxodiaceae) [J]. J Iawa, 2012, 4(33): 441 − 456. [4] 王宪, 沈华杰, 于清琳, 等. 基于IAWA的3种简易木材识别方法探究[J]. 西南林业大学学报, 2019, 39(6): 167 − 172. doi: 10.11929/j.swfu.201902073 WANG Xian, SHEN Huajie, YU Qinglin, et al. Research on 3 simple timber identification methods based on IAWA [J]. J Southwest For Univ, 2019, 39(6): 167 − 172. doi: 10.11929/j.swfu.201902073 [5] MUECHER S. , KLIJN J A, WASCHER D, et al. A new European landscape clasification(LANMAP): a transparent, flexible and user-oriented methodology to distinguish landscapes [J]. Ecol Indic, 2010, 10(1): 87 − 103. doi: 10.1016/j.ecolind.2009.03.018 [6] 王学顺, 黄安民, 孙一丹, 等. 基于BP神经网络的木材近红外光谱树种识别[J]. 东北林业大学学报, 2015, 43(12): 82 − 85, 89. doi: 10.3969/j.issn.1000-5382.2015.12.018 WANG Xueshun, HUANG Anmin, SUN Yidan, et al. Back propagation artificial neural network combine with near infrared spectroscopy for timber recognition [J]. J Northeast For Univ, 2015, 43(12): 82 − 85, 89. doi: 10.3969/j.issn.1000-5382.2015.12.018 [7] 谭念, 孙一丹, 王学顺, 等. 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[J]. 光谱学与光谱分析, 2017, 37(11): 3370 − 3374. TAN Nian, SUN Yidan, WANG Xueshun, et al. Research on near infrared spectrum with principal component analysis and support vector machine for timber identification [J]. Spectrosc Spectral Anal, 2017, 37(11): 3370 − 3374. [8] 陈远哲, 王巧华, 高升, 等. 基于近红外光谱的淡水鱼贮藏期质构品质的无损检测模型[J]. 激光与光电子学进展, 2021, 58(12): 491 − 499. CHEN Yuanzhe, WANG Qiaohua, GAO Sheng, et al. Nondestructive testing model for textural quality of freshwater fish instorage usingnear-infrared spectroscopy [J]. Laser Optoelectron Prog, 2021, 58(12): 491 − 499. [9] 郭文川, 朱德宽, 张乾, 等. 基于近红外光谱的掺伪油茶籽油检测[J]. 农业机械学报, 2020, 51(9): 350 − 357. doi: 10.6041/j.issn.1000-1298.2020.09.040 GUO Wenchuan, ZHU Dekuan, ZHANG Qian, et al. Detection on adulterated oil-tea camellia seed oil based on near-infrared spectroscopy [J]. J Agric Mach, 2020, 51(9): 350 − 357. doi: 10.6041/j.issn.1000-1298.2020.09.040 [10] 潘拓, 马鑫, 谢安, 等. 利用主成分分析法优化BP神经网络模型在砂砾岩岩性识别中的应用[J]. 新疆地质, 2020, 38(3): 417 − 420. doi: 10.3969/j.issn.1000-8845.2020.03.022 PAN Tuo, MA Xin, XIE An, et al. Application of the optimized BP neural network model based on principal component analysis in lithology identification of glutenite reservoirs [J]. Xinjiang Geol, 2020, 38(3): 417 − 420. doi: 10.3969/j.issn.1000-8845.2020.03.022 [11] ZHU Hongyan, CHU Bingquan, FAN Yangyang, et al. Hyperspectral imaging for predicting the internal quality of kiwifruits based on variable selection algorithms and chemometric models [J]. Sci Rep, 2017, 7(1): 1 − 13. doi: 10.1038/s41598-016-0028-x [12] ARAÚJO M C U, SALDANHA T C B, GALVÃO R K H, et al. The successive projections algorithm for variable selection in spectroscopic multicomponent analysis [J]. Chemometrics Intell Lab Syst, 2001, 57(2): 65 − 73. doi: 10.1016/S0169-7439(01)00119-8 [13] 董蒙, 栾希亭, 吴宝元, 等. 基于自适应遗传算法的电液伺服系统控制[J]. 机床与液压, 2019, 47(14): 78 − 83. doi: 10.3969/j.issn.1001-3881.2019.14.018 DONG Meng, LUAN Xiting, WU Baoyuan, et al. Control of electro-hydraulic servo system control based on adaptive genetic algorithm [J]. Mach Tools Hydraul, 2019, 47(14): 78 − 83. doi: 10.3969/j.issn.1001-3881.2019.14.018 [14] 冯国红, 朱玉杰, 徐华东, 等. 应用遗传算法-主成分分析-反向传播神经网络的近红外光谱识别树种效果[J]. 东北林业大学学报, 2020, 48(6): 56 − 60. doi: 10.3969/j.issn.1000-5382.2020.06.011 FENG Guohong, ZHU Yujie, XU Huadong, et al. Using near infrared spectrum to identify tree species by GA-PCA-BP neural network [J]. J Northeast For Univ, 2020, 48(6): 56 − 60. doi: 10.3969/j.issn.1000-5382.2020.06.011 [15] 许锋, 付丹丹, 王彬, 等. 基于MCCV-CARS-RF建立红提糖度和酸度的可见-近红外光谱无损检测方法[J]. 食品科学, 2018, 39(8): 149 − 154. doi: 10.7506/spkx1002-6630-201808024 XU Feng, FU Dandan, WANG Bin, et al. Nondestructive detection of sugar content and acidity in red globe table grapes using visible near infrared spectroscopy based on Monte-Carlo Cross Validation-Competitive Adaptive Reweighted Sampling-Random Forest (MCCV-CARS-RF) [J]. Food Sci, 2018, 39(8): 149 − 154. doi: 10.7506/spkx1002-6630-201808024 [16] ROMERO-TORRES S, PÉREZ-RAMOS J D, MORRIS K R. Raman spectroscopic measurement of tablet-to-tablet coating variability [J]. J Pharm Biomed Anal, 2005, 38(2): 270 − 274. doi: 10.1016/j.jpba.2005.01.007 [17] 于慧伶, 门洪生, 梁浩, 等. SA-PBT-SVM的实木表面缺陷近红外光谱识别[J]. 光谱学与光谱分析, 2018, 38(6): 1724 − 1728. YU Huiling, MEN Hongsheng, LIANG Hao, et al. Near, infrared spectroscopy identification method of wood surface defects based on SA-PBT-SVM [J]. Spectrosc Spectral Anal, 2018, 38(6): 1724 − 1728. [18] 殷勇, 王光辉. 连续投影算法融合信息熵选择霉变玉米高光谱特征波长[J]. 核农学报, 2020, 34(2): 356 − 362. doi: 10.11869/j.issn.100-8551.2020.02.0356 YIN Yong, WANG Guanghui. Hyperspectral characteristic wavelength selection method for moldy maize based on continuous projection algorithm fusion information entropy [J]. J Nucl Agric Sci, 2020, 34(2): 356 − 362. doi: 10.11869/j.issn.100-8551.2020.02.0356 [19] 朱淑鑫, 顾兴健, 杨宸, 等. K均值算法结合连续投影算法应用于土壤速效钾含量的高光谱分析[J]. 江苏农业学报, 2020, 36(2): 358 − 365. doi: 10.3969/j.issn.1000-4440.2020.02.015 ZHU Shuxin, GU Xingjian, YANG Chen, et al. K-means algorithm combined with successive projection algorithm for hyperspectral analysis of soil available potassium content [J]. J Jiangsu Agric Sci, 2020, 36(2): 358 − 365. doi: 10.3969/j.issn.1000-4440.2020.02.015 [20] 陈伟, 李创, 唐荣年. 应用间隔随机蛙结合连续投影算法检测橡胶树叶片氮含量[J]. 河南科技大学学报, 2019, 40(5): 51 − 56. CHEN Wei, LI Chuang, TANG Rongnian. Application of interval randomfrog combined with successive projections algorithm to detecting nitrogen content in rubber tree leaves [J]. J Henan Univ Sci Technol Nat Sci, 2019, 40(5): 51 − 56. [21] 熊智新, 房桂干, 梁龙, 等. 近红外光谱结合连续投影算法检测综纤维素含量[J]. 中国造纸学报, 2019, 34(4): 46 − 51. doi: 10.11981/j.issn.1000-6842.2019.04.46 XIONG Zhixin, FANG Guigan, LIANG Long, et al. Full cellulose content in composite optical fibrous in combination with continuous projection algorithm [J]. Transac China Pulp Paper, 2019, 34(4): 46 − 51. doi: 10.11981/j.issn.1000-6842.2019.04.46 [22] 明曼曼, 陈芳, 孙恺琦, 等. 基于集群算法优化BP神经网络的NIRS树种识别研究[J]. 西部林业科学, 2020, 49(5): 124 − 128. MING Manman, CHEN Fang, SUN Kaiqi, et al. NIRS tree species identification based on cluster algorithm optimized BP neural network [J]. J West China For Sci, 2020, 49(5): 124 − 128. [23] 汪紫阳, 尹世逵, 李耀翔, 等. 基于可见/近红外光谱识别东北地区常见木材[J]. 浙江农林大学学报, 2019, 36(1): 162 − 169. doi: 10.11833/j.issn.2095-0756.2019.01.020 WANG Ziyang, YIN Shikui, LI Yaoxiang, et al. Identification of common wood species in northeast China using Vis/NIR spectroscopy [J]. J Zhejiang A&F Univ, 2019, 36(1): 162 − 169. doi: 10.11833/j.issn.2095-0756.2019.01.020 -
-
链接本文:
https://zlxb.zafu.edu.cn/article/doi/10.11833/j.issn.2095-0756.20210377

计量
- 文章访问数: 129
- 被引次数: 0