The Near-IR spectra that are used for quality control of food are quite complex to analyze. They contain overtones and combination spectra which are broad-band and a few orders of magnitude lower in intensity as compared to the Mid-IR intensity of the fundamental band1,2.
Many suppliers of food and nutritional products are interested in a handheld near-IR instrument that can measure the ingredients of their raw supplies and also sort them. This is not only a hardware challenge but measuring the reflectance spectrum by the hardware and converting it to the absorption spectrum by the app is only the first step of this process. The most difficult part is to use a chemometrics model such as a Principal Component Analysis (PCA) for sorting/identification or a Partial Least Squares (PLS) model for quantitative predictions. Each model should have its own set of calibration spectra which should preferably be accessible through a cloud platform. Lastly an app is needed which could have a two way communication with the cloud. The measurement taken by the instrument is directly saved on the cloud and the latter, in turn, runs the specific model and sends back the results of evaluation back to the user. This way, for example an unskilled worker could perform all the measurements by the press of a button and classify or quantify the objects under test rapidly and accurately.
Allied Scientific Pro (ASP) offers the Nirvascan spectrometer which is a handheld spectro-photometer that measures the reflectance spectrum of an object in the range 900-1700 nm.
Although developing a cloud based app is a near future goal, ASP has laid the foundation by developing a PC-run cloud called NirvaCloud. It is possible for a user to upload spectra and make a classification of a group of spectra under a NIR Profile. One can plot the NIR profiles containing hundreds of spectra of say wine grapes all at once.
One can plot the NIR profiles containing hundreds of spectra of say wine grapes all at once. Figure 3 shows the plotting of a grape NIR profile. One can make a PCA model or a PLS model from different sets of NIR profiles and apply them to a test set.
One can do cloud matching by applying a model to an unknown spectrum. In the example below, a model made of vitamins, tomato and grapes has been matched against a vitamin spectrum (The grey square) and has identified it correctly.
Nirva-Cloud is also capable of doing quantitative predictions by creating PLS models from a calibration data set and applying it against a validation set. It is important to note that the pre-processing of data will improve the R-squared of prediction because it gets rid of the offsets and makes absorption peaks more visible.
The PLS model made out of the pre-processed grape spectra was applied to the pre-processed grape validation set and the following predictions were obtained.
If the pre-processing had not been done, the R-Squared would have been nearly 0.5.
The cloud service is available to customers for a 3 months free trial and after that subscription fees will be charged. Customers also have the option of contributing to the cloud’s library by sharing the data and possibly collecting royalty charges.
References:
1- Dahm DJ, Dahm KD. 2001. The Physics of Near-Infrared Scattering. In Williams P, Norris K, editors. Near Infrared Technology in the Agricultural and Food Industries, 2nd ed., Saint Paul: American Association of Cereal Chemists, p 19-37.
2- C.E. Miller, “Chemical Principles of Near Infrared Technology”, Chapter 2 in Near Infrared Technology: In the Agricultural and Food Industry, P. Williams and K. Norris (Editors), Amer. Assn. of Cereal Chemists; 2nd Ed. (November 15, 2001) .