Search
Many bees in a beehive

What can we learn from digital hives?

Computer vision can count insects, be used for quality control in industry, and can even detect skin cancer

May 17 2022Patrick Köhler

What does AI vision tell us about bees?

Do you like bees? Like many of my colleagues, I have become an avid beekeeper. This is because insects fascinate me and I want to protect them. I like to use digital technologies and methods for this. And the best part is: by using our virtual beehive AI project as an example, we can show what is possible with AI vision – that is, AI-based image recognition – in many industries.

How can data generate intelligence?

Woman in protective suit films beehive

A few years ago, bee mortality spurred us to develop high-tech beehives at our Innovation Center in Munich. IoT sensors in the hive measure temperature, weight, and humidity. Additional environmental sensors e.g. ozone content. Beekeepers obtain these values on their smartphones via the Open Telekom Cloud, which is why we like to refer to it as a baby monitor for bees. AI means our digital hive will be able to do even more in the future. We have equipped some of the hives with a webcam and a special camera with an 830 nanometer wavelength to detect incoming and outgoing bees in the video stream. The cameras supply us with all kinds of data to feed our artificial intelligence. We use a computer vision application. This machine learning method can be trained to identify a variety of patterns. Visual object recognition systems can detect and process almost any object or living thing in digital photos or videos. 

If the bee disappeared off the surface of the globe then man would only have four years of life left. No more bees, no more pollination, no more plants, no more animals, no more man.

Albert Einstein

Why AI Vision?

For our application, the system must first be able to reliably distinguish bees from wasps and hornets. We have trained AI algorithms to reognize "bee patterns" with the help of a vast amount of data. Success: the artificial intelligence can now identify which insects are bees, and thus counts only these. But that's not enough for us. We are also using neural networks to track the activity of the bees in the connected hive. Linking this information to the measurement results obtained from the IoT sensors gives us a more accurate picture of the cause-and-effect relationships. This expands our knowledge of beekeeping. If you would like to see how temperature or air values such as ozone affect the activity of our insects, I recommend you visit our Bee Flight Data page.

Is there a faster way to detect skin cancer?

While more than 300 Telekom employees in the Green Pioneers community are now involved in beekeeping, among other things, our Innovation Center's "Digital Hive" project is of course about more than just bees. We can use the bee project to illustrate the value of deep learning and machine vision, since there are potential applications in almost all industries. We are currently working on a project that uses visual object recognition for skin screening. We can train AI to recognize changes in the skin. If we then integrate the software into an app, we would simply be able to photograph or film our moles. The app then uses the training data acquired from pictures and videos to decide whether it would be advisable to consult a dermatologist. Of course, we don't want this to replace medical examinations, but instead to ease the burden on specialist staff and speed up the early detection process.

Trend book: Sustainable Transformation

How can your enterprise benefit from sustainable transformation? Practical tips and expert knowledge in a nutshell.

AI vision: real-time checks while welding

Industrial image processing can currently be found in companies primarily in quality control, but also in warehouse inventory or spare parts management. An example: countless welds are needed to turn sheet metal into a car door. The human eye is not capable of detecting whether the machine is actually making the connection correctly in real time. This means that the quality of the seam has to be assessed afterwards. Unless you use machine learning, sensors, and images as training material. Like the AI project of an automotive manufacturer and the AI factory from T-Systems. Here, the quality control takes place directly during the welding process: This is possible because the robots have learned to assess the quality of their work using image processing. If something goes wrong, they can sort out the faulty pieces themselves.  

Why algorithms like more than just bees

AI vision platforms present very few limits for potential applications: This is because algorithms can be trained to deal with a vast range of objects – such as bees, skin cancer, and welding seams. Computer vision really shows its strengths in applications where we need to find or count objects or extract information from images. With processes like these, AI is paving the way for us to achieve fully automated and constantly self-improving factories. If, for example, a company wants to optimize its logistics processes in order to better utilize its warehouses, freight rooms, and containers, then image analysis systems based on computer vision lend themselves very well to the task. Or consider companies facing heightened security risks. Here, AI could trigger an alarm whenever an employee is not wearing the required helmet or when unauthorized persons are in security areas. Result: This type of digitaliszation makes it possible for companies to automate tasks that until now have tied up a lot of resources.

Is AI vision available as a service?

Implementing AI is not a trivial matter; many find it easier to obtain the software as a service. Our AI Solution Factory and the AI Vision platform from T-Systems MMS prove that this also works for computer vision. Purchasing an image processing system as a managed service means you don't have to worry about server architecture, computing power, or data preparation. When choosing a service provider, consider the following:

  • Does your provider have a reputation for trustworthy AI? Does the company have guidelines in place for dealing with artificial intelligence?
  • Does it offer you a cloud-based approach and a standardized environment for developing your AI solutions?
  • Can it connect an application suite to a deep learning factory?
  • Does it assure the integration of AI developments into your customer systems?

This level of diligence is important because, after all, you want to process data that is important to you – and not just count bees.

What the future holds

With managed services like these, companies can quickly and easily make use of systems that were previously too complex. Artificial intelligence and, more specifically, computer vision or AI vision will enable us to adapt end products more quickly, accurately, and precisely to customer needs in the future – such as when we are looking to develop customized medications. If you are curious as to what is already possible with AI today and what may be possible in the future, then you will likely enjoy the article by my colleague Pavol Bauer: “How humanity and AI merge to create masterpieces?” And there is one thing left to say about us. If you would like to know what a connected mug can tell us about the innovative power of digitalization, then take a look at this article of mine. I am happy to recieve your feedback and questions. You are also welcome to stop by in person in Munich. I would be pleased to welcome you to the Innovation Center.

Contact me: Patrick.Koehler@t-system.com.

About the author
Patrick Köhler, Senior Innovation Manager

Patrick Köhler

Senior Innovation Manager, T-Systems International GmbH

Show profile and articles

You might also find this interesting

AI vision workshop

Are you interested in AI and digital image recognition?
Please get in touch!
I look forward to starting a workshop with you.

Do you visit t-systems.com outside of Germany? Visit the local website for more information and offers for your country.