TUE Peter de With

Peter de With is a professor of video content analysis at Eindhoven University of Technology.

30 November 2022

AI and deep learning surveillance techniques are making inroads in food production, observes Peter de With.

Artificial Intelligence (AI) and deep learning for data-driven system solutions have revolutionized vision-based system designs in the past years at an ultrahigh pace. This high change rate isn’t expected to relax anytime soon. On the contrary, with the advent of transformers, a type of neural network that excels in image classification, a new wave of innovations is expected to come upon us. This technology can be inserted in just about every digital system or sub-system design in which signal patterns or signal behavior are learned and improved by data-driven algorithms.

Our research group started a few years ago to put surveillance in a broader perspective than just the behavior of people or other traffic participants. This is how we ended up working with several animal breeding companies, along with Wageningen University & Research. Those companies are nothing to sneeze at, with a turnover of billions of euros and research departments stocked with well-educated and highly skilled people who have an open mind toward new technical developments.

The objective of using surveillance in this context was to give animals a better life, lower the losses of bred animals for the companies and benefit society by shrinking the footprint of the food industry. The losses refer to animals not growing well because they have ‘bad neighbors’ or ‘bad leaders’ in their group, which pick on weaker animals and abuse them.

Surveillance would make it possible to reorganize animals into social groups in which the weaker personalities aren’t hunted down by the leaders. Instead, they can live a pleasant life in a social group with more friendly leaders. The previously significant losses of animal lives during the breeding periods may eventually go down to zero.

 advertorial 

Free webinar ‘Modernizing your code base with C++20’

As many production tool chains now adopt C++20 features, the potential this brings is unlocked. What advantages can recent versions offer to your code base? In this webinar we’ll look at the great improvements C++ has gone through and how features like concepts and ranges can transform your code. Register for 2 February, 4PM.

It takes little imagination to understand that AI and deep learning can be readily applied to this surveillance case. The detection of animals can, for example, be implemented with an object detector implemented in a neural network (like the You Look Only Once or Yolo detector), while animal tracking can be added in various ways. When animal walking behavior is registered by such an algorithm, the behavior of the aggressive leaders can be isolated. Every day, we’re getting closer to this target, thereby confirming that AI can potentially also revolutionize animal breeding to become a more data-driven industry that optimizes animal life and food production.

Surveying plants and not animals, the carbon footprint of food production can be further reduced by growing vegetables in a fully controlled environment instead of outside in the field. This way, climate, light and plant feeding can be highly optimized for maximum growth efficiency. But this is only achieved if there’s continuous monitoring of growth and development, so poorly developing plants are recognized at an early stage and replaced by a new seed insertion. Such a ‘green’ production hall can be fully automated and controlled by robots in a similar way as the logistics centers for storing and rerouting packaged goods.

In short, AI is going green and will contribute to agricultural food production with higher efficiency. This is an area where the Netherlands has phenomenal chances, as it has knowledge of both food production and AI with deep learning. The emergence of robotized food growing operations is a rather plausible scenario for Dutch society. Robots and AI are upon us, whether we like it or not.

In this respect, I don’t understand the negative perception of columnist Rosanne Hertzberger in NRC of 19 November. She complains that robots are failing in areas like healthcare and have developed much slower than she’d hoped or expected. Well, they’re most certainly not failing: it’s the overregulated healthcare domain that complicates every bit of progress, slowing down and frustrating innovation. In other domains, the development and growth of robots and AI are nothing less than spectacular and innovation speed is breathtaking. Healthcare will sooner or later follow.