Detection of the wooden breast defect

Introduction

The concept behind HACCP is to provide a management tool for food safety assurance at all stages, from production to consumption [1]. This tool is now standard worldwide for maintaining and enforcing food safety standards.

Table 1: The seven steps of HACCP principles [2].

Detection of foreign material and dangerous foodborne diseases is part and parcel of food production in accordance with HACCP hygiene management’s seven steps (see in Table 1).

It is through a combination of spatially-resolved simultaneous analysis, of multiple analytes in real-time, and a powerful ejection steering unit in-camera, that EVK has managed to develop an ideal in-line solution for food production line management – that is in accordance with HACCP guidelines.

EVK’s HELIOS cameras, when combined further with the proprietary software package SQALAR, are capable of solving many of the HACCP defined tasks, through the in-camera capabilities of detection and ejection steering. HELIOS is therefore highly suitable in many respects, for instance: Principle 3 is essentially covered as is principle 6 (through data statistics), and principles 4 and 5 are certainly covered – these are the monitoring and sorting tasks at which HELIOS excels.

The solution

EVK developed a solution for an industrial partner, using quantitative hyperspectral analysis to increase sorting specificity, that sorts in real-time and at the production line. This application example illustrates how EVK collaborates with partners, to develop sensor-based solutions for them that are tailored to their particular food safety, and process optimization, challenges.

Sensor-based quality assurance removes the need for manual intervention, and in addition, performs real-time selective removal of objects that do not meet quality criteria.

The detection of this defect requires a more sophisticated analysis approach than that used for dealing with simple, one-to-one correlations of a measurement value to an observable. Hence, detection using colour cameras (see Figure 1), or using NIR simple spectra, is insufficient for sorting good quality from bad quality, chicken breasts. The reason is that many observable chemometric differences, between normal and wooden chicken breasts, do not necessarily correlate with the actual defect. Many of the differences are natural variations that are not related to the product quality. In fact, the wooden defect is hard to define and is therefore actually done subjectively, by experienced workers, using haptic touch.

The solution to this complex problem requires a new approach: one which uses of detection of multiple chemometric features, that are correlated to defect expression and are quantitatively measured. A further step of logical feature combination creates a chemometric model, which has the precision and accuracy to detect wooden chicken breasts.

The chosen optical sensor system is a HELIOS NIR G2 CLASS near-infrared hyperspectral camera, with halogen lighting, and positioned to monitor a conveyor belt. Data analysis and the creation of a chemometric model are performed with the EVK SQALAR software packet.

Relevant parameters, for a which a spectral feature correlation to wooden breast defect was found, include: Hardness, moisture, protein content, calcium concentrations and manual visual assessment. These were measured separately and used as reference inputs for creating a chemometric model. Hardness was measured using a Durometer, moisture measured using microwave dehydration unit and weighing, and calcium/protein content was measured by an external food laboratory. Manual assessment was carried out at the production line by experienced personnel.

Only with these parameters is it possible to solve this classification and sorting problem, in an industrially relevant manner. The logical combination of defect correlated features is what delivers the reliable and predictable performance necessary for removing wooden chicken breasts from the production line.

An illustrative example of such a logical combination is shown in figure 2. In this test, ten bad and ten good examples of chicken breasts were inspected. Two quantitative parameters are represented using two colour channels: red and green. The resulting colour hue in the chemical image shows the relative dependence, between the two parameters; with a greenish hue showing the location of the defect.

The final model that solves the detection problem is more complicated. Parts of the modelling are shown in figure 3 – with algorithm results shown in false colours, and wooden breast defects shown in percentages.

Conclusion

The example discussed here is a solution that uses HSI analysis for quality inspection. It is a problem that cannot be solved with a straightforward application of sensors, including even hyperspectral spectral cameras. Instead, the wooden breast defect is only solvable through the wealth of application experience accrued by EVK and the quantitative chemical imaging processing it developed [3] for its HELIOS range. Today, this application example is performing successfully on a production line of a European chicken meat processor.

Authors: Dr. Eduard Gilli, Dr. Matthias Kerschhaggl, Alexander Fetz