Application of Interval Partial Least Squares with Differential Evolution Algorithm in Wavelength Selection of Near Infrared Spectroscopy for Fishmeal
  
View Full Text    Download reader
DOI:
KeyWord:near infrared spectroscopy  fishmeal protein  feature extraction  interval partial least squares  differential evolution
  
AuthorInstitution
ZHANG You-you,CHEN Wei-hao,TANG Zhi-min,GU Jie,MO Li-na,CHEN Hua-zhou 1.College of Science,Guilin University of Technology;2.College of Electromechanical and Information Engineering,Chongqing College of Humanities Science and Technology;3.Center for Data Analysis and Algorithm Technology,Guilin University of Technology
Hits: 1554
Download times: 662
Abstract:
      Protein content is an important indicator for the evaluation of the quality of fishmeal.In this paper,a near infrared(NIR) spectral analysis technique combined with a feature selection method was adopted to establish a rapid quantitative analytical model detecting the protein content of fishmeal samples.Combining the interval partial least squares(iPLS) with the differential evolution(DE) algorithms of binary mutation strategy,a novel optimization mode,ie.interval partial least squares differential evolution(iPLS-DE) was established for the wavelength selection of the NIR spectral data for fishmeal samples.9 optimal feature wavebands were first selected by iPLS-DE through adjusting the number of equally divided intervals in iPLS,and then the discrete characteristic wavelength combinations in the optimal wavebands were further to screened out by the DE algorithm of binary mutation strategy.According to the evaluation indexes for the model,the optimal model of iPLS-DE was determined,and compared with the optimal model of iPLS.Results showed that,when the full spectrum was equally divided into 5 intervals,50 discrete characteristic wavelengths were screened out by iPLS-DE to establish an optimal model.The prediction root mean square error and relative prediction derivation of the iPLS-DE optimization model for the test set samples were 1.033% and 4.058,while the prediction root mean square error and relative prediction derivation of the iPLS optimization model for the test set samples were 1.131% and 3.855,respectively.In comparison with the common iPLS models,the iPLS-DE model is more feasible to improve the predictive ability of NIR analytical model applied to the quantitative detection of fishmeal protein.
Close