Article is available in full to IFST members and subscribers.

Register on the FST Journal website for free

Click the button to register to FST Journal online for free and gain access to the latest news


If you are an IFST member, please login through the Members Area of the IFST website.













Sensors break out of the lab

Isabel Hoffman, Mark Bloore, Behafarid Darvish and Zoltan Kovacs of Canada and UK-based Tellspec describe a new, miniaturised food sensor able to detect compounds in food at a molecular level. They explain the potential of the sensor for detecting melamine adulteration of foods.

Spectral analysis

Spectral analysis is a standard technique for discovering the chemical composition of substances. Different parts of the spectrum are suited to different types of analysis. The near-infrared (NIR) spectrum is particularly good for studying organic substances, because it responds to bonds between their different types of atoms and thus to the makeup of entire molecules.

A spectrum is obtained by shining NIR light onto a sample and recording the intensity of reflected light at each wavelength in the NIR range. Different atomic bonds will absorb light at different wavelengths and by different amounts creating a pattern, which can be examined for data on the bonds present in a sample.

Until recently, this type of analysis had to be performed in a laboratory with large and expensive equipment to get spectra of adequate quality. However, in 2014, Canada and UK-based Tellspec designed a new sensing system that combines a hand-held miniature NIR spectrometer with a smart phone linked to a cloud-based collection of machine-learning algorithms that can be trained to detect compounds, such as adulterants and contaminants in foods, at a molecular level.

Rapid, mobile food sensing

The internal light source in the miniaturised sensor focuses a beam of light through the front window into the food. Light reflected from the sample is then collected through the same window. This light is dispersed onto a micro-mirror device and measured by an optimised detection system. A digital electronic spectrum is produced, characteristic of the composition of the food.

The digitised spectrum of the food is transmitted wirelessly from the scanner to the Tellspec analysis engine in the cloud. The algorithms analyse the spectrum for information about the food and send the results back to a smartphone in seconds. A combination of machine learning, bioinformatics techniques and traditional spectroscopy provides the ability to extract nutritional information from a spectrum, the unique fingerprint of the food.

Tellspec has built an extensive food database of reference spectral scans and data on the quality and authenticity of the food from all points in the food supply chain, from farm to fork. The patented real-time cloud analysis can help monitor events of food fraud as well as food contamination locally and in specific regions, thereby helping consumers and authorities to make choices to prevent the onset of health issues related to food.

The food sensor (Tellspec Enterprise Food Scanner) uses digital light processing technology developed by Texas Instruments, which improves the scanner performance due to a higher signal-to- noise ratio as well as a more accurate spectrum acquisition. It is targeted at the B2B market and is suitable for rapid, non-destructive food quality and food fraud detection. The company is currently finalising another miniaturisation of the future Tellspec Food Sensor generation 1 (to be launched Q2 2018).

Detecting melamine in infant formula

In 2008 there was an infamous incident of infant formula adulteration in China with the industrial chemical melamine. This allowed producers to dilute the formula but still have it pass protein-content tests. At least six infants died of kidney failure and tens of thousands were sickened.

That was not the first incident. Even today, pet foods, livestock feed and commercial flour shipments are found with added melamine. There is a strong need for quick and easy detection of such adulteration at all levels in the food chain.

Melamine has a characteristic spectrum in near-infrared light (Figure 1), due in part to the many nitrogen-hydrogen bonds present. In addition to the obvious series of peaks, there are subtler features present, which are still significant. Mixing with other substances alters all of these features and may overlay them with the spectra of other substances (Figure 2). Machine learning provides a powerful array of techniques for finding spectral features that distinguish the melamine signature hidden within the complex mix of substances found in any usual foodstuff.

Complicating the detection of melamine in powdered infant formula is the fact that the spectra are different depending on whether the melamine was mixed into the formula before or after it was dried to powder. The melamine signature is less prominent when it is mixed into liquid formula. A detection method must be able to recognise both cases, or even a combination of them.

A rapid, portable sensor able to detect melamine in foods could find wide application in routine food analysis.

A combination of machine learning, bioinformatics techniques and traditional spectroscopy provides the ability to extract nutritional information from a spectrum.'

Figure 1, Melamine NIP spectrum
Figure 2 Adulterated infant formula spectrum

Validation testing

Different infant formulas were contaminated with various doses of melamine (0-10%) and samples were scanned with Tellspec Enterprise Food Scanners. Spectra were recorded in the 900-1700 nm interval, with 2nm spectral step. Partial least squares regression (PLSR) was used for quantitative models to evaluate the relationship between the melamine concentration and NIR spectra. The PLSR models were optimised with cross-validation, where data of single samples with their repeats were left out of the calibration and were used for validation, iteratively.

Average absorbance spectrum of melamine shows peaks at 1021, 1473, 1494 and 1522nm (Figure 3), which is in line with results obtained by other researchers[1].

The accuracy of the results obtained was assessed by using statistical algorithms for validation. A high coefficient of determination (R2) was found in calibration (0.9821) and in cross-validation (0.9808). The errors of calibration (RMSEC) and cross-validation (RMSECV) were 0.3898 and 0.4032% respectively (Figure 4a). Independent prediction, using newly-prepared samples, closely matched the known concentrations of melamine in samples indicating the accuracy of the built model with 0.9774 R2 and 0.4491% error of prediction (Figure 4b). The results indicate that the food scanner was effective at determining melamine concentration in infant formula down to 1%.

Similar performance testing was carried out in mixed samples of wheat gluten and urea contaminated with melamine up to a concentration of 18%[2]. Concentration measurements were found to be accurate using statistical models and independent prediction based on data from two scanners.

Figure 3 Smoothed and normalised average NIR spectra of melamine and infrant formula with different concentrations of added melamine
Figure 4 Results of the partial least squares regression calibration (blue and +) and cross-validations (red and •) for melamine (a), and results of the independent prediction (b)


The hand-held food scanner can warn consumers, as well as commercial buyers, of melamine adulteration in infant formula, flour and gluten supplies, pet foods and any other foodstuff that might benefit from a seeming boost in protein content. This has already been demonstrated for powdered infant formula.

It could be a useful tool for users, buyers, inspectors and regulators for rapid melamine detection. It offers the potential for manufacturers to prevent contamination of their products, regulators to track contamination to its source and consumers to be confident in the quality of their food.

Isabel Hoffman, founder and CEO, Mark Bloore, Behafarid Darvish and Zoltan Kovacs

Tellspec, 7B Pleasant Blvd, Suite 991, Toronto, Ontario, Canada M4T 1K2

Email Web


1. Cantor, S.L., Gupta, A. & Khan, M.A., 2014. Analytical Methods for the Evaluation of Melamine Contamination. Journal of Pharmaceutical Sciences, 103(2), pp.539–544. Available at: [Accessed January 8, 2017].

2. Kovacs, K., Bazar, G., Darvish, B., Nieuwenhuijs, F., Hoffmann, I., 2017. Simultaneous detection of melamine and urea in gluten with a handheld NIR scanner, OCM 2017, 3rd International Conference on Optical Characterization of Materials. pp. 13 – 23.


View the latest digital issue of FS&T or browse the archive


Click here

Become a member of the Institute of Food Science and Technology


Application handbook: Food, Beverages, Agriculture

IFST Twitter Feed