|
|
vol. 6, nr. 1 (2005)
|
Artificial neural networks in compositional analysis of rapeseed meal from NIRS – assessment of applicability
|
|
|
Marek Wróbel1, Henryk Czarnik-Matusewicz2, Ryszard Siuda1
|
|
| (get PDF )
|
|
|
1 Institute of Mathematics and Physics, University of Technology and Agriculture, ul. Kaliskiego 7, 85-796 Bydgoszcz |
|
|
2 Research Institute of Clinical Pharmacology, Faculty of Pharmacy, Medical Academy, ul. Bujwida 44, 50-345 Wrocław |
|
|
vol. 6 (2005), nr. 1,
pp. 261-272
|
|
|
abstract:
Near Infra-Red Spectroscopy (NIRS) is a rapid and cost-effective method widely used for the determination of chemical composition of agricultural products. Reflectance spectra recorded in near infra-red region for a set of samples of known composition are used for establishing calibration model by use of one of standard multivariate calibration methods, like MLR (multiple linear regression), PCR (principal component regression) or PLS (partial least squares). Multivariate calibration is a task that can be tried to be solved with artificial neural networks (ANNs) as well. The present paper is aimed at assessing the applicability of artificial neural networks as a tool for the determination of the content of main nutritional components of rapeseed meal: protein, dry mass, fibre and oil, on the basis of NIRS measurements. To the knowledge of the authors, no paper has been published on modelling of the dependence of chemical composition of rapeseed meal and NIR spectra with ANNs. Two most popular types of ANNs are tried in this work: multi-layer perceptron (MLP) and radial basis function (RBF). The obtained results show that chosen types of ANNs can provide models of performance comparable to that characterizing models built with MLR.
|
|
keywords:
rapeseed meal, artificial neural networks, near infrared spectroscopy
|
|
original in:
English
|
|
|
|
|
|
