Artificial neural network calibrations

16. Apr, 2018
By Richard Mills
ANN calibrations are powerful calibrations covering very large datasets. Get an overview the advantages of ANN calibrations and when to use them in this short article and video interview.

An Artificial Neural Network (ANN) is a calibration model that is based on the neural structure of the human brain. ANN calibration models have been developed since the early 1990’s in line with advances in computing power that have enabled calibrations with very large datasets.

The dataset aspect is relevant where many variables such as varying raw material, different product varieties and multiple parameters are involved. A single ANN calibration used in a dairy application can cover all products within a product group. For instance, the FOSS FoodScan cheese calibration covers hard and semi-hard cheese, soft, cream and processed cheese and a range of parameters such as Fat, Moisture / Total Solids, Fat in dry matter, Salt and Protein.

Several thousands of data points might be included in the calibration. To take another example, the FOSS ANN Meat calibration covers the parameters Fat, Mois¬ture, Protein and Collagen and was developed using approximately 20,000 spectra collected globally from the more than 1.000 FOSS dedicated meat analysers installed worldwide since 1989. The huge number of spectra makes the calibrations robust and versatile.

When to use ANN calibrations?

A major advantage with ANN compared to other calibration methods is that you can cover a large range of parameters without having to switch between a number of individual calibrations. For instance, when testing meat, the operator can stay with the same calibration for different fat ranges and different products instead of having to make a decision about changing to another calibration and then implementing it on the instrument.

Where ever they are used, ANN calibrations save time and make life easier for the operator while also reducing the risk of operator error.
Using ANN also means there will be less calibrations to verify, minimizing the validation time and costs involved. When many products and parameters are involved, this aspect can save significant costs compared to using other calibration methods.

Machine learning and ANN
There is a lot of excitement these days about the possibility to program all manner of machines (including analytical instruments) to handle ever-complex tasks.
As explained in this video, the work has actually been going on in relation to food and agri analysis for many years using ANN calibrations. The technique is particularly valuable for handling the variable nature of samples such as meat and cheese that need to be tested in the meat and dairy industry.

“If you want to be able to handle many factory sites in many regions, then you need to use the neural network model.”

FOSS Chemometric specialist, Lars Nørgaard explains the potential of Artificial Neural Network (ANN) calibration models including the ability to handle sample variability in meat and cheese.

See more industry-related articles and videos

back to top icon
The content is hosted on (Third Party). By showing the content you accept the use of Marketing Cookies on You can change the settings anytime. To learn more, visit our Cookie Policy.