Project Summary
The project aims to use Machine Learning techniques to allow cheaper and unattended updating of drone threat libraries within drone detection sensors and networks
Project Achievements
Initially the project developed and tested ML techniques for drone detection and classification. It then migrated the ML model onto drone detection hardware. State-of-the-art methods were then used to assess the level of confidence in predictions made by ML models. The project also proposed a concept archictecture to allow an entire system to function without requiring manual library updates.
Conclusions
Overall, the project proved that the concept is viable. A proof of concept system architecture was generated to allow the idea to be tested. The investigation into confidence scores, used to identify if an unseen threat was present, took significant effort. Results proved the idea is sound, and in specific applications identification of unseen threats is possible. However, additional work is needed in order to obtain consistently useful confidence scores and raise the technology readiness level (TRL). This is expected through further projects, and the concept is expected to be deployed into operational systems during 2021. This will offer a world-class, cost effective and practical drone detection solution for the challenge outlined.
