Quality control is crucial in the food industry. Monitoring the product’s nutritive property contributes to protecting and increasing a brand reputation. Fat content is one of the properties that consumers look at first when buying meat. Its quantitative level needs to be precisely documented. Also, monitoring the fat content while transforming meat plays a cost-effective role. New automation technologies are therefore required to improve brand competitiveness.

For this study, Atria (Seinäjoki, Finland)  provided five minced meat samples. We obtained five extra samples by mixing the five original ones to include ten samples in this analysis. To know their fat content accurately, Specim ordered measurements from a 3rd party laboratory, Seilab. According to the Gerber method, they are certified in fat analysis, using a butyrometer (Seilab in Seinäjoki, Finland; method NMKL 181, 2005; see Table 1 below).

Measured value by SeilabMeasured value by FX17
Sample 10.6%0.9%
Sample 216%15.2%
Sample 3 *10%10.4%
Sample 4 *18%20.8%
Sample 575%75.1%
Sample 6 (mix)3%2.7%
Sample 7 (mix)6%5.5.%
Sample 8 (mix)11%12.8%
Sample 9 (mix)19%19.0%
Sample 10 (mix)24%23.5%
Table 1: fat content on each sample included in this study. Samples 3 and 4 were used for validation purposes.

We measured the samples with Specim FX17 hyperspectral camera (Fig.1). Hyperspectral imaging is a non-destructive method that combines spectroscopy and imaging. It collects NIR spectra for each pixel of the acquired image (900 – 1700 nm). Those can be converted into fat content employing relevant processing algorithms. Here a regression model was built and calibrated on eight samples and applied on the two remaining ones (indicated with * in Table 1). We used perClass Mira software to process the data.

Figure 1: FX17 on the 40×20 scanner (left) and example of a sample on the scanner sample tray (right).

The regression model results are presented in Table 1 and Fig.2. It clearly shows that the FX17 is a suitable tool to measure precisely the fat content in minced meat.

Regression plot of the quantitative model for fat content prediction. Red dots relate to calibration samples, whereas green ones relate to validation samples.
Figure 2: Regression plot of the quantitative model for fat content prediction. Red dots relate to calibration samples, whereas green ones relate to validation samples.

In addition to measuring the fat content in samples, hyperspectral imaging is suitable for measuring its distribution (Fig.3).

Figure 3: Example of fat distribution on a meat sample (here Sample 4).
Figure 3: Example of fat distribution on a meat sample (here Sample 4).

Conclusions:

Specim FX17 in machine vision systems will provide meat transformers crucial and accurate fat quantification information. This fast and non-destructive method is also suitable to detect other properties such as moisture and freshness. Furthermore, this method can sort contaminants such as pieces of wood and plastics. FX17 is a perfect tool for industrial quality control. Besides, the methods’ flexibility allows a fast adaptation to new regulations.

Hyperspectral imaging offers cost reduction and quick adaptation to new regulations by providing real-time information about the manufacturing process for decision making and real-time control of meat processing factory input and output being within specification.