Author |
Message |
Gabi Levin (gabiruth)
Senior Member Username: gabiruth
Post Number: 28 Registered: 5-2009
| Posted on Wednesday, December 09, 2009 - 12:55 am: | |
Hi Wipawee, I am not sure that straight PCA can give you what you want, even if you have a larger set of data. PCA is looking for variance, and it is not feasible to "force" it to recognize only specific reason for variance - it will pick on any reason for variance. It could be that you will need to use a little different way of classification - you may have to assign a totally untransformed starch the arbitrary value of 1, the fully transformed starch an arbitrary value of 2, and if you happen to have samples with known 50% transformation the arbitrary value of 1.5, and perform a PLS1 regression. This will enable the PLS1 model to focus on spectral changes that are unique to the starch transformation that you are looking for, and ignore moisture variations. This is true however, only of your set will have a random variation of mositure in all samples, but if you happen to have a situation where one type will consistently have a different level of moisture it will not work well. Gabi Levin |
Wipawee Pongsuwan (wipawee)
New member Username: wipawee
Post Number: 2 Registered: 12-2009
| Posted on Tuesday, December 08, 2009 - 7:55 pm: | |
Howard and Gabi, thanks to both of you for the advice. However, when PCA analysis was performed, it seems that significant data influenced the separation between group of samples are from variation of water rather than degree of starch transformation as I wanted it to be. However, that might resulted from the reason that I used quite few number of samples in the model calibration. Anyway, thank yo so much for the answer! |
Gabi Levin (gabiruth)
Senior Member Username: gabiruth
Post Number: 27 Registered: 5-2009
| Posted on Tuesday, December 08, 2009 - 12:37 pm: | |
Hi Wipawee, To expand on what Howard said - what is really and practically needed is a set of samples in which moisture will vary over most or all of of the anticipated range you may see in the process itself. If the samples which are part of the calibration set include all the anticipated variation of mositure - the chemometric software will "learn" how to ignore the moisture changes by assigning the necessary factors and coefficients to the appropriate wavelengths in the whole spectrum. Thus, from practical point of view you will have to use a relatively large number of samples for your calibration set to make sure you cover all possible variability of the product. I hope this helps. Gabi Levin |
Howard Mark (hlmark)
Senior Member Username: hlmark
Post Number: 300 Registered: 9-2001
| Posted on Tuesday, December 08, 2009 - 9:25 am: | |
Wipawee - water absorbs virtually everywhere in the NIR, therefore you can't find a wavelength region that will "exclude it out". It is for reasons like that that chemometric analysis is so critical to the success of NIR analysis. Since the water absorbance can't be excluded, what needs to be done is to use the variations of the water bands at some wavelengths to compensate for the variations of the water absorbance in the wavelength regions where your analyte absorbs. \o/ /_\ |
Wipawee Pongsuwan (wipawee)
New member Username: wipawee
Post Number: 1 Registered: 12-2009
| Posted on Monday, December 07, 2009 - 7:22 pm: | |
Hi everyone, has anybody here used to study the application of NIR to monitor the transformation of starch after extrusion process? If so, which region of NIR spectrum should be considered as significance? and also in the case that moisture content of samples is quite varied, should the water region be excluded out? Thank you in advance! |
|