Key to information superiority:
Artificial Intelligence

From data to dominance

HENSOLDT has long relied on machine learning (ML) and artificial intelligence (AI) to improve products and solutions. Close cooperation with the German Armed Forces and NATO has resulted in in-depth knowledge of the specific needs and procedures of the military. This, together with the vast amount of data available, provides extensive opportunities for the development of new capabilities for the future.

November 7, 2024
info@hensoldt.net

CATEGORIES:
Innovation

The importance of Artificial Intelligence

Software is an essential component of modern military operations in the sense of cross-force multi-domain operations (MDO). Artificial intelligence topics are among the most important technology trends of recent years and are therefore one of the most important national key technologies. AI offers great opportunities in the defence sector in particular and logically also plays a major role at HENSOLDT.

This is particularly due to the fact that deep neural networks have now found widespread application thanks to their impressive capabilities, aided by large amounts of training data and optimised hardware.

The position paper Software Defined Defence from the BDSV, BDLI, Bitkom and BMVg underlines the importance of the topic:

"For AI-based services, new patterns (e.g. camouflage, radio signals) must be generated as well as identified, services must be retrained, tested, certified and rolled out to the tactical level. It should be possible to implement these models quickly and with little integration effort in the respective platform. One use case, for example, is the utilisation of data from all sensors and other sources (C4I systems) for AI-based functions (connectivity and interoperability)."

Unlocking potentials

HENSOLDT has been using artificial intelligence methods for decades, in particular machine learning methods. Recently, deep learning algorithms and generative AI methods have been added. These improve the automated analysis of sensor data in many ways and strengthen the product portfolio through the further development of technologies in areas such as radar, spectrum dominance, optronics and avionics.

The huge amounts of data (gigabytes in seconds) generated by modern AESA radars, digital receivers or multispectral sensors can hardly be converted into usable information ("actionable intelligence") without AI processes.

The experts at HENSOLDT use state-of-the-art algorithms to structure filtered sensor data in a data model, train the latest AI methods and interpret the information on the basis of extensive domain knowledge from military operations. This interaction in terms of AI integration enables HENSOLDT to reliably and rapidly extract information from data that can contribute to information superiority on the battlefield and make the decisive difference.

Unlocking the potential of these volumes of data is challenging: on the one hand, they form the indispensable basis for information on which decisions can be made. On the other hand, the data is initially unstructured and finding relevant information can be like looking for a needle in a haystack.

To limit the complexity of sensor systems and solutions, it is necessary to integrate sufficient computing power into the sensor to ensure that the data volume is at least pre-processed and filtered near or within the sensor. The pre-processed data from individual sensors is compiled into a comprehensive common operational picture of the situation through sensor data fusion and enrichment with additional metadata, which would not be possible without sophisticated algorithms and AI due to the vast amount of data.

The integration of artificial intelligence in HENSOLDT solutions: The AI-enabled portfolio

HENSOLDT solutions are further enhanced by harnessing advanced embedded systems that can efficiently deploy AI capabilities in particular. Sensor systems traditionally rely on a data pre-processing unit close to the sensor or servers in a headquarters for data analysis.

The idea for next-generation systems is that each sensor can take over a large part of the data analysis itself. For this purpose, a dedicated data pool of real and simulation data is utilised and further expanded, which can be used to train and validate AI processes. Specific optimisations enable efficient operation on different hardware, so that sensors and more complex systems benefit equally from AI-supported functions.

The domain expertise and the enormous amounts of data that are already available from decades of hardware development are a major advantage over providers that only deal with AI. At the same time, it is important to keep an eye on the environment and build an innovative ecosystem through smart partnerships.

Thus, HENSOLDT is working closely with external partners, for example various Fraunhofer Institutes, in the areas of sensor data analysis, explainable AI and ethical considerations for the military use of AI functions. Furthermore, HENSOLDT actively participates in a range of research projects within the European Defence Fund Initiative, working on emerging military and security topics and cooperating with renowned research institutions and commercial players.

The AI ecosystem is complemented by partnerships with large digital transformation companies and start-ups with novel but already industrialised technologies.

AI in HENSOLDT solutions: Information superiority for the user

The experts in the various HENSOLDT areas integrate AI into HENSOLDT solutions to ensure the information superiority of the armed forces on the battlefield.

SETAS See-Through Armour System

The 360° all-round vision system See-Through Armour System (SETAS) paired with the CERETRON data evaluation and fusion system can simultaneously observe and analyse the entire area around an armoured vehicle. Deep learning algorithms for object recognition increase the degree of automation in data evaluation. This can ensure the early detection of hazards, for example.

One of the core competences of the Optronics & Land Solutions division is to run such processes directly integrated in the sensor. In the case of cameras, this is known as embedded computer vision, which brings "intelligence directly into the camera" and reduces problems such as latency times, explains Dr Michael Teutsch, AI Lead: “Data processing close to the camera increases efficiency and usually reduces the size, weight and power consumption requirements of sensor systems and solutions.”

The maturity of this product development was impressively demonstrated in May 2023 during a combat exercise under real conditions at the Bundeswehr Combat Training Centre (GÜZ) in Gardelegen. Enemy camouflaged land vehicles were automatically detected and reconnoitred in the field. This reconnaissance information was geo-referenced within seconds and distributed to allied troops.

Similar approaches are also being pursued for the optronic mast systems (OMS) installed in submarines as well as for airborne self-protection and camera systems. To this end, the professionalisation of the acquisition and utilisation of relevant data is being consistently driven forward. In addition to automating the annotation of real training data using state-of-the-art foundation models, the focus is also on generative AI to generate synthetic training data from the highly scalable simulation.
Radar-based Automatic Target Recognition Ground-based air defence

In the radar sector, machine learning methods are used to enable automatic target classification and thus a quick overview of the situation for the operator. This ranges from methods for classifying the jet engine modulation effect, for unambiguous identification of aircraft/fighters, to micro-Doppler methods for differentiating between different air and ground targets, through to conventional and modern methods for automatic target recognition on synthetic aperture radar (SAR) images.

For example, the SPEXER 2000 3D MkIII radar offers the classification of small and slow-moving objects in an urban context. In particular, the differentiation between small drones and birds ensures a comprehensive picture of the situation, which makes the operator's work considerably easier. In addition, the differentiation of a large number of target types for different radar types takes place in the HENSOLDT portfolio and is constantly being further developed.

In the field of automatic target recognition on SAR images, it has also been shown that modern deep learning methods can be used not only to detect and classify targets, but also to successfully segment entire SAR scenes. This enables methods based on this for the semantic analysis of scenarios for which SAR images are available, for example to automatically recognise battle formations or attack formations.

Current developments enable the detection of swarms of drones, almost instantaneous detection of targets that are only visible for a short time, such as combat helicopters that only ascend briefly to drop their weapons, or the suppression of clutter caused by wind turbines, complex urban structures and increasing occupancy of the electromagnetic spectrum. It can even be used with radars with low Doppler resolution, so that the entire HENSOLDT portfolio benefits and enables a fast, reliable overview of the situation.
Kalaetron Integral Air SIGINT

AI can also support SIGINT systems such as Kalaetron Integral in all mission phases. Intelligent algorithms help to parameterise the highly complex SIGINT sensors. This is based on cognitive AI processes that are based on the perception-action cycle and use reinforcement learning methods to optimise the sensor technology.

During the mission, intelligent cluster and classification algorithms recognise pulses from radar systems that belong together or use spectral analyses to detect and classify communication and radar signals. In the next steps, radar devices and modes can be identified by recognising pulse patterns.

Through close coordination and cooperation between the individual areas in terms of AI integration, HENSOLDT is able to reliably and rapidly extract information from data that contributes to information superiority on the battlefield.

Artificial intelligence in the defense industry means responsibility

As a company in the defense industry, HENSOLDT not only asks itself how AI can be used, but also how it can be used responsibly. In complex threat scenarios, deterrence and superiority - and therefore survivability and effectiveness - can only be achieved if one's own forces are enabled to go through the "find-fix-track-target-engage-assess cycle" faster and more effectively than the enemy. Responsibility means that the human being always remains involved in the control process, but is nevertheless comprehensively supported by automation at all times in order to act in a timely, effective and resource-saving manner and to avoid collateral effects as far as possible.

HENSOLDT is aware of its responsibility: in order to optimally enable highly complex defense systems, artificial intelligence must be mastered in its entire spectrum. The defense systems of tomorrow are created at the interface of artificial intelligence with the in-depth knowledge of our customers' application scenarios.