Artificial intelligence for more efficient airport security checks

Security is a top priority at airports. Passengers have to undergo a security check before boarding their plane, during which their bags are scanned using x-ray equipment. Security personnel examine the images to ensure there are no prohibited items such as knives, firearms or bombs. In future, they will be supported by artificial intelligence (AI). A team led by Yanik Sterchi and Adrian Schwaninger at the FHNW School of Applied Psychology is conducting research in collaboration with airports, security service providers and equipment manufacturers. Sterchi explains: “For several years now, Automated Prohibited Item Detection Systems (APIDS) have been used to improve the efficiency of security checks on passenger baggage. As part of our applied research and development project, we’re supporting the rollout in Switzerland. We’re investigating how well APIDS work, how human-technology interaction is structured and how security personnel should be trained in the use of the new technology.”

Technology to support people

Security checks involve screening bags using x-rays of different wavelengths. X-rays penetrate the material to varying degrees depending on its density and its metallic or organic makeup. X-ray scanners automatically convert this density and material information into colour images so staff can better identify the different objects. Trained personnel use the shape and colour of the objects to identify prohibited items.

Security personnel have been supported by technology that detects explosives for many years. Explosives detection systems (EDS) detect explosives based on their density and material in x-ray images, which works very reliably, especially where computed tomography (CT) is involved. APIDS, on the other hand, are based on deep neural networks. In addition to the material, they use other information to identify prohibited items, such as their shape. Modern x-ray machines combine the three technologies – CT, EDS and APIDS – in one device. Schwaninger explains: «Combining these technologies allows security checks to be more efficient by eliminating the need to unpack liquids and laptops.»

Tests with proprietary AI

Working alongside researchers from the Zurich University of Applied Sciences (ZHAW), the project team developed their own APIDS algorithms for their experiments. They trained them with hundreds of thousands of images so that the AI can learn what a suspect object looks like from different angles in a wide range of luggage products. The researchers then challenged their algorithms in a simulator with multiple sets of 10,000 or so x-ray images of items of luggage. Sterchi: «The recognition rate of AI is influenced by various factors. The x-ray machine and the direction of exposure play an important role, as do the position and material of the prohibited items and the degree to which they are covered by other items.» APIDS manufacturers have confirmed to the researchers that these factors are also relevant for their own algorithms. In this respect, the researchers’ AI was comparable to professional applications. As Sterchi says: «We now have a better understanding of how APIDS algorithms can be improved and what to consider when testing the quality of these AI algorithms.»

How does the human-technology interaction work?

The research team and industry partners were particularly interested in the interaction between humans and AI. To this end, they conducted various studies on simulators involving airport security personnel. In one of these, the AI marked suspicious items in the x-ray image with a red frame. «By using AI, security personnel were able to detect prohibited items faster and with greater accuracy,» says Schwaninger. «As expected, this positive effect only applies to prohibited items on which the AI has been trained. In our current studies, these have been knives, firearms and sharp tools.»

In another study, the researchers wanted to find out how security personnel should be trained when introducing AI systems. For example, the APIDS tested have not yet detected pepper spray. «It’s very important to let security personnel know what AI can and cannot do,» explains Sterchi. «In addition, training with examples is useful in terms of developing a better understanding of the capabilities of AI.» The project team intends to investigate this in more detail in a follow-up study. «When introducing AI, staff should always be involved,» adds Schwaninger. «It improves acceptance and correct use of the new technology.»