Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

The quantitative prediction abilities of various multivariate calibration methods are compared on real datasets arising from near infrared reflectance spectroscopy analyses of agricultural and food products. The calibration methods compared include three traditional statistical methods (multiple linear regression with stepwise selection of variables, principal component regression and partial least-squares regression) and three neural networks methods based on the multilayer perceptron architecture. A theoretical and bibliographical study is given for each method taken into consideration. Comparisons are done on 14 dependent variables bound to 5 spectral collections. The comparison criterium is the standard error of prediction. A factorial experimental design is used to investigate the effects of three subsidiary factors : the calibration dataset size, three NIR spectra transformation methods and three levels of spectral simplification. The results show that neural networks methods often compete with, or outperformed traditional calibration methods. Nevertheless, no method is uniformly the best one. Practical recommendations based on the obtained results are given in order to help users in the calibration development process.

Spectrometry --- Measurement --- Statistical methods --- Analyse mathematique

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Forest resources --- Forest inventories --- Aerial photography --- Geographical information systems --- computer applications --- Forest mensuration --- Wallonie --- Strate --- Zone homogene

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Proefschriften

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Informatica --- Software

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

Computer software