Computer Vision and Robotics for the agri-food industry

Computer Vision and Robotics for the agri-food industry

This expertise is focussed on non-destructive sensing, machine learning, automation and robotics. We apply advanced computer vision and robotics technologies to applied problems in agri-food processes. Wageningen Food & Biobased Research is a global leader in the development of technological solutions for the automated quality inspection of agri-food products.

Applied Sensing

Wageningen experts use a wide range of sensors to automatically and objectively measure product quality parameters such as colour, shape, firmness, ripeness, stress and diseases. These include:

2D imaging

We use cameras to capture the colour information and extract informative features to determine the quality of the products objectively. Objective measurements help in identifying any deviations from the standard and for determining the quality classification of the products.

3D imaging

Beyond colour, structure and shape are extremely informative features. We use multiple sensors like lasers, stereo and depth cameras, multiple 1/2-D cameras, to extract 3D information. WUR-developed ‘Marvin’-technology is being used for high-speed plant phenotyping, seedling sorting and other bulk quality assessment and sorting applications in industry.

NIR spectroscopy and Hyperspectral imaging

Many properties of fresh food – like Brix, dry-matter, internal damage, firmness – are not visible. Spectrometers and hyperspectral cameras can help in identifying these properties non-destructively. We specialize in correlating spectral data with internal quality parameters. This technology is becoming increasingly cost-efficient, fast and suitable for industrial applications like automatic bulk sorting.

Other methods
We also use sensing technologies such as XRT and Tera-Hertz imaging which provide considerable certainty regarding various internal quality aspects not caught by other sensors.

Machine (deep) learning

The data generated by the sensors is increasingly automatically analysed and interpreted via powerful machine learning methods. We have considerable experience in classical and modern learning methods. We frequently apply deep learning, as classical methods struggle to address the complexity of problems and data. These methods are used for identifying patterns in the data that correlate with the identified problems with higher reliability.

Automation and robotics

Several of our solutions are commercial products where we take lab-scale results and ready them for real-world applications. Here we look at how all the information can be made applicable for actual practical use, ranging from classification of a batch of agri-food products to steering a sorting machine or robot. A close cooperation with product experts, software developers and industrial machine builders is crucial to ensure the correct translation into practice.

Interested in the possibilities?

Contact us for an informal conversation.