Author |
Message |
Natalie (natalia)
New member Username: natalia
Post Number: 5 Registered: 2-2010
| Posted on Wednesday, February 24, 2010 - 2:51 am: | |
Hello to all, First of all, thanks for all your helps. I think I should try to do a standardization because of two problems: 1. I have to deliver some results in the next days, so if it is possible to resolve my problem doing a good job, even it is a little bit complicate, I�ll prefer it. 2. We made reference analysis in the old samples, and we don�t have access to them, they are like they have been disappeared. So for this reason I think to do a new calibration couldn�t be the best solution... So, I�ll try to do standardization. Has anybody have literature about that? And Scott, what is your e-mail? Thanks again, Natalie |
Scott Ramos (lsramos)
Junior Member Username: lsramos
Post Number: 7 Registered: 1-2007
| Posted on Tuesday, February 23, 2010 - 5:37 pm: | |
Natalie, Performing standardization on NIR spectra is a little complicated but there are software packages that include standardization and make it pretty straightforward. Since I work for one of the software companies, if you want to pursue this, I suggest you contact me off list. Regards, Scott Ramos scott_ramos at infometrix dot com |
Bruce H. Campbell (campclan)
Moderator Username: campclan
Post Number: 121 Registered: 4-2001
| Posted on Tuesday, February 23, 2010 - 1:54 pm: | |
I would be very cautious about using your oak samples of years ago for a correction. Unless you are sure your samples are stable with time, you could be misled. In my opinion, you would probably be better off doing a recalibration. If you do, then consider scanning a substance that you know is absolutely stable with time and would have bands in the wavelength region you are scanning. Then you could correct spectra if you find a change in your system. You could also use the "reference" spectra periodically to verify your system hasn't changed. Of course, you wouldn't use the reference spectra in the calibration. Bruce |
NATALIA VP (natalia)
New member Username: natalia
Post Number: 4 Registered: 2-2010
| Posted on Tuesday, February 23, 2010 - 1:47 pm: | |
Hello Scott, Thanks for all your different ways to achieve a solution to my problem. I think that the best solution for me it would be to create a transformation matrix. Is it very complicate? Because I�m too novice in NIRS. Thanks again, Natalie |
NATALIA VP (natalia)
New member Username: natalia
Post Number: 3 Registered: 2-2010
| Posted on Tuesday, February 23, 2010 - 1:33 pm: | |
Dear Ana, I�ll try to answer your questions as well I can. The instrument with I�m working is a Foss 6500 transport. The material which I�m working with are diferents portions of oak. And about the last question, I don�t know if there was any scanned reference or standard sample. But I could find out about this. I hope the answers give you some clues. Thanks a lot Natalie |
Scott Ramos (lsramos)
Junior Member Username: lsramos
Post Number: 6 Registered: 1-2007
| Posted on Tuesday, February 23, 2010 - 1:04 pm: | |
Natalie, As Ana mentioned, standardization may work to allow you to use the calibration model from the old spectra to make predictions on the new spectra. Let me offer some additional questions and comments. 1. Based on your description, you have a collection of spectra from both before and after changing the lamp. Since I presume you will continue working with the new lamp, can you make a model from the new spectra that could then serve for predictions on any additional spectra collected? 2. If you feel that it is nevertheless essential to use the model from the old spectra, are there spectra in common from before and after the lamp change (that is, in which the Y composition(s) are the same)? These could serve as the transfer spectra used to create a transformation matrix in the standardization algorithm. 3. Some have suggested that a more robust method, under changes like in your situation, would be to add some of the new spectra to the old spectra set and make a new model on the combined data. The concept is that changes due to the different lamps would be incorporated into the model. Regards, Scott Ramos |
Ana Garrido-Varo (garrido_0)
New member Username: garrido_0
Post Number: 2 Registered: 1-2006
| Posted on Tuesday, February 23, 2010 - 12:19 pm: | |
I am not sure if I got all your previous e-mail . May be the best solution will be to do a standardization of your instrument (using as master, samples analyzed before the lamp was changes and as satellite samples scanned after that). Your 34 samples scanned before and after could be very useful. But before to advise you a more concrete action plan, please could you answer to the following questions: - Which instrument brand and model are you working with? - Which product are you working with? - Do you scanned previously some reference/standard sample in your instrument, which material? Regards Ana |
NATALIA VP (natalia)
New member Username: natalia
Post Number: 2 Registered: 2-2010
| Posted on Tuesday, February 23, 2010 - 10:50 am: | |
Thanks to both, I think I should explain better the problem. According to what I think I have made the next things, I have passed exactly 34 samples of the old ones (samples of 2007); so now i�ve got 34 samples passed in 2007 and 34 samples in 2010. But the problem is that I think I might do something with them in the way to allow me to use the equation I did to predict, because my equations were made with old samples, and now I�ve got new samples and I think I can�t predict really good with this problem. And the clue is the lamp of the equip has been changed in that period. What should I do? :-) Thanks a lot Natalie |
David Russell (russell)
Senior Member Username: russell
Post Number: 45 Registered: 2-2001
| Posted on Tuesday, February 23, 2010 - 10:42 am: | |
No question that more reference samples is always better. I simply wished to convey that if recalibration is not possible, that correction of the result would be less difficult for a novice to accomplish than attempting transformation of the spectra. Dave Russell |
Howard Mark (hlmark)
Senior Member Username: hlmark
Post Number: 306 Registered: 9-2001
| Posted on Tuesday, February 23, 2010 - 9:39 am: | |
Don't mean to contradict you, Dave, especially since I agree with your underlying premise that a skew-and-bias correction will bring Natalie's results in, but I would recommend at least ten samples. Three is theoretically the minimum to use, but I would prefer her to use a few more, to get a little robustness into the correction. Howard \o/ /_\ |
David Russell (russell)
Senior Member Username: russell
Post Number: 44 Registered: 2-2001
| Posted on Tuesday, February 23, 2010 - 9:06 am: | |
The short answer is "it depends". You may have to do a new calibration of the instrument. However, if you have samples where the values are known (minimum of 3, low, medium, and high) you could attempt to calculate a calibration correction with those results. Hard to comment any further without knowing the full details of your work. Dave Russell |
NATALIA VP (natalia)
New member Username: natalia
Post Number: 1 Registered: 2-2010
| Posted on Tuesday, February 23, 2010 - 4:14 am: | |
Hello, I�m having a problem. I�m very amateur in NIRS. I�ve been passing samples, and in this project the samples started to be passed through the NIRS in 2007, so now in this equip has been a changes of the lamp. The problem is the new spectras are desplazed from the oldest, and i have seen that they are similar. So I think that may be there is a way to make a matrix to transform the old spectra, i mean, try to standarize the old spectra wiht the new ones? Anybody can help me? Thanks a lot. Natalie |