Drones enable previously undreamt-of possibilities for data generation e.g., during inspections. Depending on the sensor system used, enormous amounts of data are generated for high-quality recordings. It is, therefore, not uncommon for uncompressed images to quickly account a volume of several gigabytes.

Depending on the complexity of the examined object, hundreds of images are necessary for later state evaluation or photogrammetric processing. The fact that the images can fill entire hard disks is rather common.

Drone above – AI below

High quality of captured images, which are necessary for later analysis, therefore also brings a high data volume with it. True to the motto „garbage in, garbage out“, poor data quality usually also produces poor results.

A typical example in this context is facade inspections, which require the acquisition of a large amount of image data and its evaluation, e.g., to detect cracks in the masonry or defects in lightning conductors. To obtain suitable images, drones with appropriate camera optics are used. This saves, among other things, the need to rent and erect scaffolding for the building, and the survey can be completed in less than an hour on average.

Figure 1: Crack in a facade – the inspection of a building can easily require more than 1,000 high-quality images, which must then be analyzed.

Drone and AI: How Do I Make the Technology Work for Me?

Artificial Intelligence (AI) provides incomparable added value in the analysis of a wide range of data, such as image data. The advantages are particularly evident when relevant findings, e.g., test reports, are to be generated very quickly from the existing data stock.

The FlyNex Enterprise Suite solves two problems. It offers unlimited and secure storage space to facilitate and accelerate processing. In addition, the acquired images can be connected to analysis or AI software, and the processing steps of the images can be extended as required.

Where Do I Use Artificial Intelligence in Unmanned Aerial Systems?

Overall, there are no limits to AI applications when it comes to detecting objects or damage in numerous images. As soon as a manual evaluation takes a lot of time due to the amount of data, the use of AI is worthwhile. This is especially the case in the area of error detection during inspections in the energy sector, such as during the regular inspection of pipeline networks.

Software with AI-based image analysis is also regularly used for infrastructure inspections or real estate as part of the comprehensive documentation of traffic safety obligations, as well as for condition inspections, e.g., damage to house roofs, facades or lightning conductors.

 Figure 2: Road Damage – Copyright: Procopter Gmbh

When it comes to AI, we still encounter many uncertainties. In the field of data analysis – if used correctly – it is an indispensable tool for fast and accurate results.

To understand the topic professionally, we invited DENKweit CEO and AI expert, Dr. Dominik Lausch, from the Fraunhofer-Gesellschaft to talk about artificial intelligence and neural networks in the data processing.

What Is Denkweit Doing in the Field of Artificial Intelligence, and How Did You Come up With the Idea?

We founded DENKweit in 2018 and are currently a team of 16 employees, including six AI programmers. We try to make the topic of AI as easy as possible for users to use. We have a lot in common with FlyNex in terms of the uncomplicated use of our products. At that time, I worked for a solar company in a project for image recognition. Unfortunately, this project failed. The reason was the approach on how to recognize an object. At that time, linear algorithms were chosen due to missing or not fully developed technologies. A procedure which is very complex, error-prone, and not very effective. With this painful experience, we at DENKweit have concentrated from the beginning on automating object recognition with AI, even without the user needing prior knowledge in the field of AI.

What Do You Think Is So Exciting About AI?

For me, AI is a technology that will make our everyday life much easier. That is why I am so fascinated by it. It solves problems that we were able to solve in part before, but now we are able to do it faster, better, and more accurately. To what extent AI will open up a new world for us remains to be seen. But one thing is clear: It will become increasingly important, especially in industry, because it makes processes simpler, faster and above all, more cost-effective.

What Exactly Is AI? and What Added Value Does AI Provide in Data Analysis?

First of all, I would like to say that AI can do a lot, but as so often in life, it is not a panacea. In the field of image recognition, AI-based technologies, especially artificial neural networks, are unrivaled. This means that whenever it comes to recognizing specific objects on image data, there is actually no way around AI.

Especially where data can change, the use of AI can be compelling. If you think of a defective power pole, for example, there are many different types of power poles, and the background of the drone recording can vary depending on the season. In autumn, an underlying field can show the mast in front of a dark background and in summer rather in front of a green background. Here an AI must still be able to detect the power pole and its possible defects. This is not easy, but it is possible with the use of the right technologies.

The variety and the added value of AI systems are also our motivation to improve our technology and to achieve excellent results with as little output data as possible. One of the most significant added values is that AI systems can work precisely 24 hours a day.

How Does AI Work? and Is Artificial Intelligence Really Intelligent?

In contrast to linear programming, which I mentioned at the beginning, the advantage of AI is that it seeks its own solution. As a user, I tell it what a crosswalk is, and I do so repeatedly by using different images. In concrete terms, I have to mark a crosswalk for the AI on different pictures. At some point, the AI itself can recognize and mark the crosswalk on all pictures to be analyzed.

Usually, the way of the AI to the target is always a black box. We have developed technologies to recognize where and “with what” an AI recognizes the crosswalk. We also use this to optimize the AI.

Is AI intelligent? A good example is a pedestrian traffic light. If someone next to me stops at a red light, I don’t know whether he does it because he sees the red color, because he sees that the light is on at the top or because the light gives an acoustic signal.

In short: AI is not intelligent; it is primarily based on statistical considerations. However, it can suggest an intelligence to humans, because it has sought or is seeking the way to the goal itself. Without getting into in-depth discussions here, AI is, from a factual point of view, mathematics and above all, a lot of statistics.

The partial complexity in the function, which is not really necessary for the application, also causes high prices on the market. This is one reason why many companies still refrain from taking up AI projects. We want to offer the simplest possible access here. Therefore, I see great synergies with the FlyNex motto to benefit from the advantages of drones – and why not use drones and AI together?

Can You Tell Us Briefly What Makes Your AI So Special?

The handling of our system is very simple. You only have to mark the object you want to recognize, and our AI adapts to the problem, the error, or something to be identified. When it comes to computing power, you don’t need to worry with us. The processing is done on Fraunhofer-secured cloud servers with a generous amount of computing power so that the evaluations are available in the shortest possible time. In the area of quality assurance for solar cells, for example, our AI must be able to detect a defect within 50 milliseconds with almost perfect accuracy. I think this underlines the performance of our AI technologies.

We would like to thank Dr. Dominik Lausch warmly for this exciting interview.

If you have any questions concerning the topic drone, data generation, or data analysis, please feel free to contact us.

Your FlyNex Team

The questions were asked by Michael Petrosjan, founder and CFO FlyNex GmbH.

Contact us:
info@flynex.de
Phone: +49 (0) 341 331 760