Kevin Riordan, Head of Airports & Checkpoint Solutions, Smiths Detection explains how airports can harness AI to mitigate threats.
Artificial intelligence (AI) has become an all-pervasive aspect of everyday life and is being widely deployed to boost efficiency in business processes across all sectors.
The opportunity to use AI to deliver enhanced safety, security and efficiency has also been recognised by security solution providers and airports alike, particularly in the wake of the COVID-19 pandemic, with the need for competitiveness, efficiency, cost-effectiveness and contactless solutions becoming more critical than ever before.
In line with the growing acknowledgement of the vast potential of more automated screening solutions in airports, aviation security regulators in several countries are currently working towards implementing certification and test methods for AI powered object recognition solutions; these are requisite for safe and compliant deployment.
Taking airport security to the next level
Security screening processes generate very large volumes of data and AI has the potential to generate insights from this data which serve to optimise screening operations, outcomes and system performance. Through machine learning and its subset deep learning, algorithms can be developed which imitate the ways in which the human brain processes data and identifies patterns based on examples to inform decision making.
For the development of deep learning algorithms for security scanners, a library of X-ray images are shown to the algorithm so it can learn to identify patterns in the shape of items, such as guns, gun parts, ammunition and knives or other potentially dangerous items such as lithium batteries. While the list of objects that AI algorithms can detect is constantly expanding, deep learning is currently limited in that it cannot yet detect substances or items which are inconsistent in shape. However, traditional material property discrimination-based techniques, when combined with machine learning, can be powerful for detecting such objects.
Smart, adaptable deep learning algorithms that are available for the automatic detection of dangerous, prohibited and contraband goods and substances, significantly reduce the burden on security operators and are particularly helpful for less experienced image analysts. As algorithms cannot get tired or distracted and are impartial, they reduce the risk of human errors, resulting in improved security outcomes. In addition, automated screening processes can reduce operational expenditure because, with greater screening efficiencies and productivity, fewer staff are required for dealing with passengers and their belongings.
By increasing throughputs and reducing manual processes requiring physical contact, the security screening experience becomes significantly less stressful for passengers.
With a very high level of detection, these AI algorithms have the capability to drive down false alarm rates. There is also the potential to combine the automatic explosives detection capability of a scanner with object recognition to enable ‘alarm only viewing’ of X-ray images at checkpoints, further accelerating passenger flow and unnecessary interaction between operators and passengers. However, this concept, which has been in use for hold baggage screening at airports for several years, does not yet have regulatory approval at the checkpoint.
Across checkpoint, hold baggage and cargo scanning, customs authorities could also benefit from the automatic detection of currencies, endangered species, agricultural products and other contraband in baggage or cargo to prevent the entry of smuggled items. In some cases, information on scanned items could be shared between airports, countries and authorities using wide area networks (WANs). For example, X-ray images captured at a departure airport could be sent securely to the transfer and destination airport for review by local authorities, all while the plane is still in the air.
In this way, WANs are another way to increase productivity, providing an opportunity for other agencies to view data and even run additional algorithms to analyse the data.
Enabling new approaches to screening and maintenance
The benefits of AI-powered solutions go beyond image-based analysis, with broader advantages relating to the overall approach to screening and system operation and maintenance; AI and biometrics could be used to gather, combine and analyse comprehensive passenger risk profiles to allow for the more efficient and targeted screening of passengers.
Moreover, risk profiles could be built on inputs such as biometric data, travel destination, passport details, visa information, flight behaviour patterns as well as intelligence from government and security services. Through a biometrics-enabled checkpoint, individualised risk assessments could be generated by AI based on a unique identifier (biometric data) combined with available contextual information. By applying differentiated levels of screening and focusing operator resources on those with higher risk scores, the pressure on airport screening operators and costs are reduced – while also enabling more effective monitoring and the identification of potential threats.
AI algorithms also have another useful application for airport security scanners, delivering asset performance management insights to enable predictive maintenance; these algorithms can digest swathes of data and turn them into early warnings of potential system faults and actionable performance insights. This way, airport operations teams can monitor and manage screening system performance to minimise risk and operating costs whilst maximising system uptime.
Unlocking the full potential of AI through collaboration
For all its benefits, object recognition algorithm development and implementation does not come without risks. Close collaboration is therefore required between suppliers, airports and authorities to successfully develop algorithms which can then be integrated responsibly and seamlessly into workflows without compromising compliance or operational efficiency. For example, access to the image data is key for training algorithms, so the availability, variety and richness of X-ray images is critical to enhancing the reliability of object recognition algorithms.
The more that different airports from around the world share X-ray images, the more advanced algorithms can become; this requires the involvement and openness of government bodies and airports who need to share their data and access to prohibited items with security solutions providers.
Finally, for the implementation of algorithms, there needs to be a close working partnership between software developers and OEMs to ensure that the functionality of scanning hardware is not impacted by the integration of an algorithm, therefore compromising the security process.
There is some way to go before fully automated airport security scanners can be realised. Efforts are currently focused on adding further dangerous goods and prohibited items to the object recognition list.
Beyond this, the further advancement of AI capabilities relies on the growing adoption of computer tomography-based (CT) systems, which allow for 3D volumetric object recognition and open architecture approaches that facilitate the transfer of images between airports and countries as well as the integration of third party algorithms to scanning equipment. However, this adoption is contingent on an industry approved, regulated set of standards, tests and certification methods.
Only then will airports be able to fully take advantage of AI-powered solutions to tackle emerging and evolving security threats and effectively handle rising passenger numbers in the long term.
For more information, visit: https://www.smithsdetection.com/
This article was originally published in the January 2022 edition of Security Journal UK. To read your FREE digital copy, click here.