How Ukrainian Air Defense Learned to Shoot Down Drones Using Neural Networks

In Ukraine’s skies, artificial intelligence is rewriting the rules of air defense, enabling lightning-fast detection and destruction of drones.

Artificial Intelligence Conquers New Frontiers in Aerial Warfare

Russia’s war against Ukraine is not only a contest of artillery calibers and the depth of echelons. It is also an experimental testing ground for the future war of algorithms. In 2024–2025, Ukrainian air defense forces transitioned from manual target visualization to hybrid models, where artificial intelligence began to play a pivotal role. Neural networks capable of recognizing the thermal signatures of drones like Shahed or Lancet have reduced the detection-target-destruction cycle to mere seconds. In the night sky, these milliseconds are decisive: either a UAV hits a children’s hospital, or it is neutralized before reaching its target.

Technological Evolution Gains Momentum: From Radars to Machine Learning Models and New Rules for Night Defense

Previously, traditional air defense operated on a straightforward principle: a radar detects a target, an operator classifies it, and then a decision is made to launch a missile. In the case of drones, this sequence was often too slow. Drones are low-speed, fly low, and leave little time for reaction, especially at night. Since 2023, Ukrainian forces have been experimenting with neural networks trained on samples of drones’ thermal, infrared, and acoustic signatures. These are not commercial models like ChatGPT but highly specialized systems deployed on edge devices – computers integrated directly into air defense platforms. No cloud, no internet, no lag.

Real-world combat testing showed that a neural network can identify a UAV’s type in 1.2–1.4 seconds, compared to 5–7 seconds for a human operator. In practice, this means not only saving the targeted object but also optimizing ammunition use. A human might mistakenly launch an expensive NASAMS missile at a cheap drone. An algorithm does not. Moreover, neural networks have demonstrated the ability to identify drones even in challenging weather conditions, where an operator might see only a blur on the screen. All the operator needs to do is confirm the target lock, and the system takes over from there.

Until recently, nighttime was considered the drones’ advantage. The gray contours of the sky, thermal spectrum camouflage, and delayed operator reactions played in their favor. But now, it is in the dark that neural networks reveal their full potential. Modern models distinguish engine types (piston or jet), separate the thermal fluctuations of a bird from the engine of a Geran drone. Tactically, this allows Ukrainian air defense to build defenses not based on flight paths but on drone behavior patterns. The system learns from each raid: how a drone approaches, at what altitude, and what maneuvers it uses to evade. This is no longer a fight against an “object” but against its behavioral pattern.

Humans Cannot Yet Be Removed from the Equation, and Here’s Why

Full automation is still undesirable: artificial intelligence may be faster, but it is not always more accurate. The risk of misclassifying “friend or foe” remains. For this reason, the model primarily operates in a “recommendation” mode rather than “automatic launch.” In this setup, human control is preserved, but decision-making time is minimized. Ukrainian air defense operators no longer search for targets – they receive them on a tablet, already classified, prioritized, with a countdown in seconds.

Several Ukrainian companies at the intersection of defense and IT have already integrated these models into their projects. For example, modules based on FPGA processors can be installed on both air defense systems and kamikaze drones for automatic target designation. This represents a comprehensive evolution: not only defense but also offense is shaped through real-time analysis of enemy behavior patterns. This is a new stage of technological confrontation – where algorithms understand the enemy as well as intelligence does.

The Sky Sentinel Case: A Technological Breakthrough into a New Era of Military Technology
A striking example of this technological leap is Sky Sentinel, an autonomous combat module developed in Ukraine, one of the first fully human-independent air defense systems. This is a heavy turret with 360-degree rotation, equipped with artificial intelligence capable of independently detecting, classifying, and engaging aerial targets – from Shahed-type drones to cruise missiles. During combat trials, the system demonstrated the ability to destroy objects five times smaller than an Iranian kamikaze drone at speeds up to 800 km/h.

Sky Sentinel does not require operator intervention: it suffices to receive target coordinates from a radar, and the AI independently executes the full cycle – from locking onto a thermal signature to firing a series of shots, accounting for wind resistance, target speed, and real-time trajectory corrections. Solving several engineering challenges – such as eliminating mechanical backlash and adapting stabilization to recoil – enabled the creation of a platform that combines machine intelligence with high precision. One Sky Sentinel prototype has already proven itself on the front lines, destroying four Shahed drones at a cost significantly lower than a single NASAMS or IRIS-T missile.

The greatest value of this system lies in its scalability: Sky Sentinel is designed for mass production, enabling the creation of entire “fields of AI turrets” to protect cities and logistical hubs. This shifts the air defense paradigm from a “human-AI observer” to a full-fledged “AI fighter” conducting autonomous battles against drones on the front lines. In this approach, artificial intelligence not only accelerates decision-making but also becomes the foundation of the defense architecture. Ukraine, outpacing Russia and many Western allies, has practically reached a new level – automated, cost-effective, and adaptive air defense.

Russians Are Falling Behind the Pace of Technological Progress

The Kremlin is still trying to replicate this model, but its technological base is limited. After Western bans on exporting processors and cameras, Russian air defense systems remain dependent on Chinese matrices and outdated algorithms like YOLOv3. This is yesterday’s technology. In Ukraine, neural networks operate on open-source frameworks adapted for combat tasks, allowing rapid adaptation to new target types, such as Geran-3 or Orlan-30.

The use of neural networks in air defense is not just a technological upgrade. It is a fundamentally new level of situational awareness that changes the logic of defense. Previously, the primary resource was the missile. Now, it is the microsecond. Whoever detects and classifies first wins. Ukraine has learned to do this faster. And it is precisely because of this that dark nights are no longer a problem – they have become an advantage.

Bohdan Popov, Head of Digital at the United Ukraine Think Tank, communications specialist and public figure