Selecting optimal wavelength intervals for an optical sensor: A case study of milk fat and total protein analysis in the region 400-1100 nm

Abstract A broad-band optical sensor analyzer, based on a set of light-emitting diodes (LED), for milk fat and protein analysis has been simulated and optimized using full-spectrum data in the wavelength range 400–1100 nm obtained in a designed experiment. Genetic Algorithm (GA) has been adapted to find an optimal set of wavelength intervals to be used for analysis in order to get the best prediction accuracy. Weighting and averaging of the spectral variables within the chosen intervals has been applied to take the LED emission spectra and integrating diode detection into account. Partial least-squares (PLS) regression models built on seven and six selected intervals for fat and protein, respectively, exhibit no performance loss compared to the corresponding full-spectrum models. Suggested approach is universal and can be used to customize any LED-based or similar optical sensor system for a specific analytical problem prior to the construction. The GA-based algorithm of searching optimal de-resolved spectral intervals can be used as a general variable selection method for multivariate calibration.

[1]  A. Bogomolov,et al.  Fat Globule Size Effect on Visible and Shortwave near Infrared Spectra of Milk , 2013 .

[2]  Rudolf W. Kessler,et al.  Quantitative determination of fat and total protein in milk based on visible light scatter , 2012 .

[3]  Riccardo Leardi,et al.  Genetic algorithms in chemometrics , 2012 .

[4]  Tahir Mehmood,et al.  A review of variable selection methods in Partial Least Squares Regression , 2012 .

[5]  Alejandro C. Olivieri,et al.  Wavelength Selection for Multivariate Calibration Using a Genetic Algorithm: A Novel Initialization Strategy , 2002, J. Chem. Inf. Comput. Sci..

[6]  Yanbo Cao,et al.  Portable analyzer for rapid analysis of total protein, fat and lactose contents in raw milk measured by non-dispersive short-wave near-infrared spectrometry , 2013, Chemical Research in Chinese Universities.

[7]  Alejandro C. Olivieri,et al.  Visible/near infrared-partial least-squares analysis of Brix in sugar cane juice: A test field for variable selection methods , 2010 .

[8]  A. Bogomolov,et al.  Scatter-based quantitative spectroscopic analysis of milk fat and total protein in the region 400-1100nm in the presence of fat globule size variability , 2013 .

[9]  Jiewen Zhao,et al.  Real-time monitoring of total polyphenols content in tea using a developed optical sensors system. , 2014, Journal of pharmaceutical and biomedical analysis.

[10]  C. Ruckebusch,et al.  Genetic algorithm optimisation combined with partial least squares regression and mutual information variable selection procedures in near-infrared quantitative analysis of cotton-viscose textiles. , 2007, Analytica chimica acta.

[11]  Kássio M. G. Lima,et al.  A portable photometer based on LED for the determination of aromatic hydrocarbons in water , 2012 .

[12]  Harald Martens,et al.  A multivariate calibration problem in analytical chemistry solved by partial least-squares models in latent variables , 1983 .

[13]  Nikolai V. Tkachenko,et al.  Optical Spectroscopy: Methods and Instrumentations , 2006 .

[14]  E. Gorritxategi,et al.  Optimization of the multivariate calibration of a Vis–NIR sensor for the on-line monitoring of marine diesel engine lubricating oil by variable selection methods , 2014 .

[15]  Sergey Kucheryavskiy,et al.  Determination of fat and total protein content in milk using conventional digital imaging. , 2014, Talanta.

[16]  T. De Beer,et al.  Process analytical tools for monitoring, understanding, and control of pharmaceutical fluidized bed granulation: A review. , 2013, European journal of pharmaceutics and biopharmaceutics : official journal of Arbeitsgemeinschaft fur Pharmazeutische Verfahrenstechnik e.V.

[17]  Andrey Bogomolov,et al.  Multivariate process trajectories: capture, resolution and analysis , 2011 .