基于回归CNN的烟叶近红外光谱模型研究
|
摘要
为了更大程度提取近红外光谱中的深层次关键特征,论文应用改进的回归式卷积神经网络算法,将传统的卷积神经网络架构中池化层移除,最顶层的线性分类层用回归层进行替代,构建卷积神经网络回归(CNNR)模型。为了验证该算法的有效性,论文经多次实验、对比评价指标,筛选出最佳模型:总糖、总烟碱和氯离子最佳CNNR模型的相关系数R分别为0.9318,0.941,0.933,交叉验证的RMSECV分别为0.7052,0.0710,0.0971。实验结果表明:CNNR模型抽提的特征光谱数据对三个指标有很强的解释能力,对烟叶化学成分的有较好的预测性能和综合表达能力。
In order to extract the deep critical features of near-infrared spectroscopy to a greater extent,a novel predictionmethod,namely,regression-based convolutional neural network(CNNR),which removes the pooling layer and substitute the re-gression layer for the linear softmax classification layer at the top of the general CNN's structure,has been proposed to develop thequantitative model for the tobacco constituents. In order to verify the effectiveness of the CNNR algorithm,the models for total sug-ar,total nicotine and chlorine in tobacco are built for which correlation coefficient R are 0.9318,0.941,0.933,respectively andRMSECV of cross validation were 0.7052,0.0710,0.0971,respectively. The results indicate that the extracted features have astrong ability to interpret the spectral data and have better prediction performance and comprehensive expression ability of tobaccochemical components.
In order to extract the deep critical features of near-infrared spectroscopy to a greater extent,a novel predictionmethod,namely,regression-based convolutional neural network(CNNR),which removes the pooling layer and substitute the re-gression layer for the linear softmax classification layer at the top of the general CNN's structure,has been proposed to develop thequantitative model for the tobacco constituents. In order to verify the effectiveness of the CNNR algorithm,the models for total sug-ar,total nicotine and chlorine in tobacco are built for which correlation coefficient R are 0.9318,0.941,0.933,respectively andRMSECV of cross validation were 0.7052,0.0710,0.0971,respectively. The results indicate that the extracted features have astrong ability to interpret the spectral data and have better prediction performance and comprehensive expression ability of tobaccochemical components.
引文
[1]张灵帅,邢军,王卫东,等.近红外光谱分析技术进展及其在烟草行业中的应用[J].光谱实验室,2009,26(2):197-201.ZHANG Lingshuai,XING Jun,WANG Weidong,et al. De-velopment of Near Infrared Spectroscopic Technology andIts Applications in Tobacco Industry[J]. Chinese Journalof Spectroscopy Laboratory,2009,26(2):197-201.
[2]祝元元,陈永宽,刘志华,等.近红外光谱技术在烟草行业的应用进展[J].应用化工,2010,39(11):1750-1753.ZHU Yuanyuan,CHEN Yongkuan,LIU Zhihua. Advanceon near infrared spectroscopic technology in tobacco indus-try[J]. Applied Chemical Industry,2010,39(11):1750-1753.
[3]纪鹏,冯剑,李志伟.近红外分析技术在烟草行业的应用研究进展[J].安徽农业学,2011,39(16):9848-9850.JI Peng,FENG Jian,LI Zhiwei. Development of Near In-frared spectroscopic in tobacco industry[J]. Journal of An-hui Agricultural Sciences,2011,39(16):9848-9850.
[4]申钦鹏,张涛,刘春波,等.近红外光谱定量分析技术在烟草和烟气化学成分分析中的研究进展[J].广东农业科学,2015,4(6):77-85.SHEN Qinpeng,ZHANG Tao,LIU Chunbo,et al. Re-search advances in application of quantitative analysismethod of near infrared spectroscopy in chemical compo-nents of tobacco and cigarette smoke[J]. Guangdong Agri-cultural Sciences,2015,4(6):77-85.
[5]张翼,杨征宇,葛炯,等.近红外分析技术在烟草中的应用[J].计算机与应用化学,2016,33(2):251-254.ZHANG Yi,YANG Zhengyu,GE Jiong,et al. The appli-cation of near-infrared spectroscopy in tobacco[J].Computers and Applied Chemistry,2016,33(2):251-254.
[6]陈达,王芳,邵学广,等.近红外光谱与烟草样品总塘含量的非线性模型研究[J].光谱学与光谱分析,2004,24(6):672-674.CHEN Da,WANG Fang,SHAO Xueguang,et al. Re-search on the Nonlinear Model of Near Infrared Spectrosco-py and the Total Sugar of Tobacco Samples[J]. Spectrosco-py and Spectral Analysis,2004,24(6):672-674.
[7]李世勇,王芳,邵学广.最小二乘支持向量回归与偏最小二乘回归建立烟草总糖NIR预测模型比较[J].烟草科技,2006(11):45-48.LI Shiyong,WANG Fang,SHAO Xueguang. Comparisonof Tobacco Total Sugar NIR Prediction Models Developedwith Least Square Support Vector Regression and PartialLeast Square Regression[J]. Tobacco Science&Technolo-gy,2006(11):45-48.
[8]Janik L J,Forrester S T,Rawson A. The prediction of soilchemical and physical properties from mid-infrated spec-troscopy and combined partial least-squares regressionand neural networks(PLS-NN)analysis[J]. Chemomet-rics and Intelligent Laboratory Systems,2009,97(2):179-188.
[9]Schmidhuber J,Meier U,Ciresan D. Multi-column deepneural networks for image classification[C]//Computer Vi-sion and Pattern Recognition,2012:3642-3649.
[10]Krizhevsky A,Sutskever I,Hinton G E. ImageNet classi-fication with deep convolutional neural networks[C]//Neural Information Processing Systems,2012,25(2):1097-1105.
[11]Li H,Lin Z,Shen X,et al. A convolutional neural net-work cascade for face detection[C]//Proceedings of theIEEE Conference on Computer Vision and Pattern Recog-nition. Boston,USA:IEEE,2015:5325-5334.
[12] He K,Sun J. Convolutional neural networks at con-strained time cost[C]//Proceedings IEEE Conferenceon Computer Vision and Pattern Recognition. Boston,USA:Computer Society,2015:5353-5360.
[13]Sermanet P,Eigen D,Zhang X,et al. Overfeat:Inte-grated recognition,localization and detection using con-volutional networks[J]. Computer Vision and PatternRecognition,2014.
[14]Taigman Y,Yang M,Ranzato M,et al. Deepface:Clos-ing the gap to human-level performance in face verifica-tion[C]//Proceedings IEEE Conference on ComputerVi-sion and Pattern Recognition. Washington D. C.,USA:ACM,2014:1701-1708.
[15]Farabet C,Couprie C,Najman L,et al. Learning hier-archical features for scene labeling[J]. IEEE TransPat-tern Analysis and Machine Intelligence,2013,35(8):1915-1929.
[16]Savitzky A,Golay M J E,Anal Chem. Smoothing and dif-ferentiation of data by simplied least squares procedures[J]. Analytical Chemistry,1964,36(8):1627-1639.
[2]祝元元,陈永宽,刘志华,等.近红外光谱技术在烟草行业的应用进展[J].应用化工,2010,39(11):1750-1753.ZHU Yuanyuan,CHEN Yongkuan,LIU Zhihua. Advanceon near infrared spectroscopic technology in tobacco indus-try[J]. Applied Chemical Industry,2010,39(11):1750-1753.
[3]纪鹏,冯剑,李志伟.近红外分析技术在烟草行业的应用研究进展[J].安徽农业学,2011,39(16):9848-9850.JI Peng,FENG Jian,LI Zhiwei. Development of Near In-frared spectroscopic in tobacco industry[J]. Journal of An-hui Agricultural Sciences,2011,39(16):9848-9850.
[4]申钦鹏,张涛,刘春波,等.近红外光谱定量分析技术在烟草和烟气化学成分分析中的研究进展[J].广东农业科学,2015,4(6):77-85.SHEN Qinpeng,ZHANG Tao,LIU Chunbo,et al. Re-search advances in application of quantitative analysismethod of near infrared spectroscopy in chemical compo-nents of tobacco and cigarette smoke[J]. Guangdong Agri-cultural Sciences,2015,4(6):77-85.
[5]张翼,杨征宇,葛炯,等.近红外分析技术在烟草中的应用[J].计算机与应用化学,2016,33(2):251-254.ZHANG Yi,YANG Zhengyu,GE Jiong,et al. The appli-cation of near-infrared spectroscopy in tobacco[J].Computers and Applied Chemistry,2016,33(2):251-254.
[6]陈达,王芳,邵学广,等.近红外光谱与烟草样品总塘含量的非线性模型研究[J].光谱学与光谱分析,2004,24(6):672-674.CHEN Da,WANG Fang,SHAO Xueguang,et al. Re-search on the Nonlinear Model of Near Infrared Spectrosco-py and the Total Sugar of Tobacco Samples[J]. Spectrosco-py and Spectral Analysis,2004,24(6):672-674.
[7]李世勇,王芳,邵学广.最小二乘支持向量回归与偏最小二乘回归建立烟草总糖NIR预测模型比较[J].烟草科技,2006(11):45-48.LI Shiyong,WANG Fang,SHAO Xueguang. Comparisonof Tobacco Total Sugar NIR Prediction Models Developedwith Least Square Support Vector Regression and PartialLeast Square Regression[J]. Tobacco Science&Technolo-gy,2006(11):45-48.
[8]Janik L J,Forrester S T,Rawson A. The prediction of soilchemical and physical properties from mid-infrated spec-troscopy and combined partial least-squares regressionand neural networks(PLS-NN)analysis[J]. Chemomet-rics and Intelligent Laboratory Systems,2009,97(2):179-188.
[9]Schmidhuber J,Meier U,Ciresan D. Multi-column deepneural networks for image classification[C]//Computer Vi-sion and Pattern Recognition,2012:3642-3649.
[10]Krizhevsky A,Sutskever I,Hinton G E. ImageNet classi-fication with deep convolutional neural networks[C]//Neural Information Processing Systems,2012,25(2):1097-1105.
[11]Li H,Lin Z,Shen X,et al. A convolutional neural net-work cascade for face detection[C]//Proceedings of theIEEE Conference on Computer Vision and Pattern Recog-nition. Boston,USA:IEEE,2015:5325-5334.
[12] He K,Sun J. Convolutional neural networks at con-strained time cost[C]//Proceedings IEEE Conferenceon Computer Vision and Pattern Recognition. Boston,USA:Computer Society,2015:5353-5360.
[13]Sermanet P,Eigen D,Zhang X,et al. Overfeat:Inte-grated recognition,localization and detection using con-volutional networks[J]. Computer Vision and PatternRecognition,2014.
[14]Taigman Y,Yang M,Ranzato M,et al. Deepface:Clos-ing the gap to human-level performance in face verifica-tion[C]//Proceedings IEEE Conference on ComputerVi-sion and Pattern Recognition. Washington D. C.,USA:ACM,2014:1701-1708.
[15]Farabet C,Couprie C,Najman L,et al. Learning hier-archical features for scene labeling[J]. IEEE TransPat-tern Analysis and Machine Intelligence,2013,35(8):1915-1929.
[16]Savitzky A,Golay M J E,Anal Chem. Smoothing and dif-ferentiation of data by simplied least squares procedures[J]. Analytical Chemistry,1964,36(8):1627-1639.