NIRS feature extraction based on deep auto-encoder neural network

Abstract As a secondary analysis method, Near Infrared Spectroscopy (NIRS) needs an effective feature extraction method to improve the model performance. Deep auto-encoder (DAE) can build up an adaptive multilayer encoder network to transform the high-dimensional data into a low-dimensional code with both linear and nonlinear feature combinations. To evaluate its capability, we experimented on the spectra data obtained from different categories of cigarette with the method of DAE, and compared with the principal component analysis (PCA). The results showed that the DAE can extract more nonlinear features to characterize cigarette quality. In addition, the DAE also got the linear distribution of cigarette quality by its nonlinear transformation of features. Finally, we employed k-Nearest Neighbor (kNN) to classify different categories of cigarette with the features extracted by the linear transformation methods as PCA and wavelet transform-principal component analysis (WT-PCA), and the nonlinear transformation methods as DAE and isometric mapping (ISOMAP). The results showed that the pattern recognition mode built on features extracted by DAE was provided with more validity.

[1]  Zhang Jianping,et al.  NIRS-based Pattern Recognition of Domestic Flue-cured Tobacco , 2006 .

[2]  Ronald R. Coifman,et al.  The prediction error in CLS and PLS: the importance of feature selection prior to multivariate calibration , 2005 .

[3]  Qing-Bo Li,et al.  [A dimension reduction method applied in spectrum analysis]. , 2013, Guang pu xue yu guang pu fen xi = Guang pu.

[4]  Lucas J. van Vliet,et al.  Recursive implementation of the Gaussian filter , 1995, Signal Process..

[5]  Rasmus Berg Palm,et al.  Prediction as a candidate for learning deep hierarchical models of data , 2012 .

[6]  Liu Jin-yun.  SIMCA pattern recognition of tobacco leaves based on near infrared spectra , 2009 .

[7]  Hui-Hua Yang,et al.  [Isomap-PLS nonlinear modeling method for near infrared spectroscopy]. , 2009, Guang pu xue yu guang pu fen xi = Guang pu.

[8]  S. Wold,et al.  PLS regression on wavelet compressed NIR spectra , 1998 .

[9]  Yoshua. Bengio,et al.  Learning Deep Architectures for AI , 2007, Found. Trends Mach. Learn..

[10]  Xie Yun,et al.  Analysis of Tobacco by Near-infrared Spectroscopy and Support Vector Machine , 2009 .

[11]  Keum-Shik Hong,et al.  Cortical brain imaging by adaptive filtering of NIRS signals , 2012, Neuroscience Letters.

[12]  XU Fei-fei.  Robust ISOMAP insensitive to singular value , 2007 .

[13]  Zhiqiang Gao,et al.  Text Window Denoising Autoencoder: Building Deep Architecture for Chinese Word Segmentation , 2013, NLPCC.

[14]  Shao Yongni,et al.  Discrimination of tobacco varieties using vis/near infrared spectra by wavelet transform and back propagation neural network model. , 2009 .

[15]  Xueguang Shao,et al.  A general approach to derivative calculation using wavelet transform , 2003 .

[16]  Dong Yu,et al.  Context-Dependent Pre-Trained Deep Neural Networks for Large-Vocabulary Speech Recognition , 2012, IEEE Transactions on Audio, Speech, and Language Processing.

[17]  Song Bai,et al.  Deep learning representation using autoencoder for 3D shape retrieval , 2014, Proceedings 2014 IEEE International Conference on Security, Pattern Analysis, and Cybernetics (SPAC).

[18]  Yi Jianhua,et al.  On-line chemical determination in threshing and redrying process and its application in quality control , 2009 .

[19]  J. Tenenbaum,et al.  A global geometric framework for nonlinear dimensionality reduction. , 2000, Science.

[20]  Geoffrey E. Hinton,et al.  Reducing the Dimensionality of Data with Neural Networks , 2006, Science.

[21]  Geoffrey E. Hinton,et al.  Using very deep autoencoders for content-based image retrieval , 2011, ESANN.

[22]  Peter E. Hart,et al.  The condensed nearest neighbor rule (Corresp.) , 1968, IEEE Trans. Inf. Theory.