Protection against the harmful use of drones is a relevant task for the protection of public spaces, critical infrastructure and assets. In addition to these public safety issues, drones have been introduced to the battlefield for ISR missions and armed attacks. The growth of anti-drone technologies is directly linked to the growth of the threat that drones pose in both military and civilian environments. In the military field, small drones have proliferated at such a rate as they have an impact on the battlefield. Other concerns are no longer hypothetical. Meanwhile, near-crashes between drones and manned aircraft have become a common occurrence in every crowded airspace system in the world. Air defense systems that have traditionally been used to protect airspace from manned aircraft are generally ineffective against drones.

Regardless of their weight class and categories, drones are considered threats if they have the potential to perform dangerous, harmful or unwanted acts. Drone threats must be adequately addressed by security systems, where the type and scope of mitigation techniques depend on the situation and the environment.

The project idea is to develop an “AI-framework” of novelty AI-based algorithms for the detection and classification of killer-drones. The project will build a shared database by collecting RF and EO/IR signatures of different types of drones, which can be used as training data and test set. This database will allow to compare different detection and classification AI-algorithms (i.e. ML, DL, CNN, ..).

Ai4CUAV intends to improve the Threat Evaluation Subsystem of an anti-drones through AI-based algorithms. Supposing the anti-drones composed by multiple heterogenous sensors, such as radar and EO/IR sensors, these algorithms “work” on radar signals and EO/IR images to enable the detection and classification of the killer-drones, as well as on drone trajectories to help to recognize a drone from another object.

Ai4CUAV will investigate all the key SOTA of AI techniques, including but not limited to, machine and deep learning. These techniques will be evaluated against the different use cases and scenarios, in order to assess the most adapted/promising ones.