| Author | Message | |||
     David W. Hopkins (dhopkins) Senior Member Username: dhopkins Post Number: 91 Registered: 10-2002 |
Ooops, sorry I misspelled your name, Ian ! | |||
| David W. Hopkins (dhopkins) Senior Member Username: dhopkins Post Number: 90 Registered: 10-2002 |
Hi Iam, I sure am sorry you were not at Chambersburg last week, so we could have discussed this in detail. I'm not sure what is driving your question, as I am sure you know the bands where the fat, protein and moisture absorb maximally. It appears that you are measuring reflectance spectra, and the noise certainly looks very nice. I would recommend that you omit the regions at the ends of the spectra where there are few bands of interest, above 2400 nm and below about 1200 or 1300 nm. I think the best approach to minimizing the multiplicative effects of particle size will be to use a combination of 2Der and MSC (or 2Der and SNV) in that order. Recently I have been working with Martens and Stark's Extend Multiple Scatter Correction method(EMSC), and I expect that should work very well in milk powder samples. Are you familiar with the paper by Martens etal, Analytical Chemistry, Vol. 75, No. 3, February 1, 2003 394-404? I hope this helps. Do you still have any questions? Best regards, Dave | |||
| Gavriel Levin (levin) Member Username: levin Post Number: 14 Registered: 1-2006 |
Hi ian, If you do PLS1 regressions, and if your software allows you to analyze what the Unscrambler software calls loading weights for each PC you will see clearly which wavelengths contribute most for each constituent. Generally - moisture will be in the 1450's and 1930's, fat will be in the 1660 to 1730nm region and proteins in the 1550's you can also ask Foss to provide you with their wavelength and chemical groups wall chart. Word of advice - to determine if the 2200 and above contribute or degrade - pending on degree of noise - granted that your Vision software allows you - exclude first 2300 to 2500 nm from the PLS1 regression - if the regression gets better and the SEP goes down - it means they don't contribute more than they interfere, then chop 2200 to 2500 and see what happens. It is a practical, useful way that tells you the truth about your set of data, not a theoretical universal situation. I hope this helps a bit. Gabi Levin Brimrose Corp. | |||
| Mike Zapf (mzapf) New member Username: mzapf Post Number: 1 Registered: 1-2006 |
Hello Ian, I believe that FOSS-NIRSystems has a calibration for this, perhaps you might obtain some clues, at least. Good luck | |||
| Ian Murray (imurray) New member Username: imurray Post Number: 1 Registered: 8-2006 |
Can someone please tell me the Fat, Moisture and Protein wavelengths for milk powder analysis so I can more accurately set the maths parameters in FOSS Vision Software.Also can you please tell me the regions to avoid as they are noisy? I have attached a 2nd derivative spectra.
|
2nd Deriv Spectra