| Author | Message | ||
     Tony Davies (td) Moderator Username: td Post Number: 280 Registered: 1-2001 |
Hello Don, I think we are still waiting to be told a bit more about the sample. Is this a plastic sorting problem, something similar or quite different. Do we need to give an answer for each sample? Best wishes, Tony | ||
| Donald J Dahm (djdahm) Senior Member Username: djdahm Post Number: 72 Registered: 2-2007 |
This is not to take issue with anything that is said here. A validated technique is to be respected regardless of theory. However, I would like to talk about the term �spectral subtraction�. It implies that we have a spectra from the "belt" and a spectra from the sample, and if we scale the belt spectra properly (presumably to account for the changes in sample thickness), we can just subtract the log(1/R) values to get a corrected spectra. In fact, there can be light scattering back and forth between the belt and the sample. This changes the shape of the response curve to concentration of absorber. The mathematics for this has been worked out, subject to assumptions that may or may not be appropriate for a particular situation. | ||
| David W. Hopkins (dhopkins) Senior Member Username: dhopkins Post Number: 215 Registered: 10-2002 |
Hi Balen, While it may be a useful technique to use the belt as a reference, I think you need to be careful. How constant are the results of measurements of the belt at various locations? I think your calibrations and measurements should be based on very reproducible reference measurements, or your data will be highly variable. You may need to review what kind of variation this would introduce into your spectra. A second problem which you need to consider is the varying amount of sample being measured, if you say that you are sometimes seeing the belt. As Tony might have been headed, it might be a good idea to qualify the spectra before accepting them for prediction. That way, you would only get results displayed when there was sufficient sample on the belt. So, that leads to a final question: Is it possible for you to design the sample measurement so that the sample is "conditioned" to a constant sample depth and a constant distance from the top of the sample to the input optics of the instrument? On line measurements can be very tricky, but they can be very reliable, too. They require a lot of attention to details. Best regards, Dave | ||
| venkatarman (venkynir) Senior Member Username: venkynir Post Number: 154 Registered: 3-2004 |
Hi Balen ; Your Question: Did you use your belt as reference? (so (your sample spectra - dark)/(belt spectra - dark) )? Ans: No. If I take belt spectra then when PVC samples present in on-line , we would not get reality what we want. So I used reference plate to avoid this complication. Tony I do agree with you plastic samples were as per your spectification .But when it goes on -line. Please think off | ||
| Tony Davies (td) Moderator Username: td Post Number: 278 Registered: 1-2001 |
Hello Venkatarman and Belen S, Our plastic bags were made to our specification and supplied with the instrument. They did contribute to the spectral signature of the sample but as this does not vary very much it is not selected by the PLS calibration. This situation seems quite similar to me. The white platform will have a reasonably constant signal. The variation in sample size will be a difficulty. In your application, do you have to give an analysis of every sample? Best wishes, Tony | ||
| Belen S (belen) New member Username: belen Post Number: 2 Registered: 2-2012 |
Thank you both. So venkatarman, your problem sounds pretty close to mine. what do you say you did? Did you use your belt as reference? (so (your sample spectra - dark)/(belt spectra - dark) )? | ||
| venkatarman (venkynir) Senior Member Username: venkynir Post Number: 150 Registered: 3-2004 |
Hi Tony Davies ; I read your answer, Any plastic bag has influence in NIRS spectra .It is very difficult hold same nature of Platic bags always . Based on polymer density the spectra has its own impact. I don't know how you assume same nature palstic used for all calibraiton samples. We have the problem in plastic sorting as base belt is made of PVC . We eleminated it used by taking refernce spectra and we got the result. | ||
| Tony Davies (td) Moderator Username: td Post Number: 277 Registered: 1-2001 |
Hello Belen S, Welcome to the group. I worked on a meat analyser in which the sample was held in a plastic bag. We tried to minimise the influence of the bag and then ignored it in the calibrations. We were able to meet the targets for the the performance of the calibrations. It sounds as if you will have a more difficult problem because of the sample size. If the background is spectrally different to the sample then I think it will work but perhaps not too well. Good luck! Tony | ||
| Belen S (belen) New member Username: belen Post Number: 1 Registered: 2-2012 |
Hello. I am conducting some experiments taking data from a white moving platform. When the amount of sample is small, chances are that I take also spectral information from the platform. Although the best solution would be changing the material, I have to stick with it. The fact is that It reflects light and has its own spectra...I have seen in some cases that background spectra does not affect when developing calibrations. Other people subtract the background spectra. For instance there was a study done with animal feces where the authors scanned samples inside plastic bags and removed the plastic bag spectra from the sample + bag. Does this approach look feasible when we talk about white reflective surfaces? i am going to make some tests scanning samples covering the area in black, but I would rather being able to work with the original surface, and not wait until I send samples for references (thus waste money) to know if the approach works. many thanks. B. |