AI-based framework for drone intent classification using a non-cooperative radar system

随着无人驾驶汽车（UAV）应用（如测量、医疗交付等）的普及，正在开发使其能够与载人航空共存并在受控空域内使用的系统和服务。为了确保安全，航空用户和操作人员必须警惕飞机之间以及飞机与无人机之间的潜在冲突。在欧洲，所谓的“U3先进服务”正在开发，包括容量管理和支持冲突检测和解决的系统，特别是使无人机能够操作。对于载人航空来说，一个被称为短期冲突警报（STCA）的成熟概念目前被用作安全网，提醒ATC操作员在任何情况下，任何一对合作监视（雷达）轨迹之间的用户定义的最小分离距离预计在短时间内（通常为2分钟）被违反。这是通过雷达显示器上的视觉警报来实现的，尽管有些系统也提供声音警报。该系统通常使用基于协同雷达的跟踪，即直接从应答飞机获取位置信息的系统。然而，这种解决方案不能应用于非应答飞行器，因此未来有人驾驶和无人驾驶飞行器的混合空域将需要一个更有弹性和更强大的解决方案，以充分释放无人机应用的潜力。该博士建议开发一种基于非合作雷达特征的强大冲突检测和警报机制，以对飞机意图进行分类，作为一种提高混合空域安全性的自动化方法，同时满足有人驾驶航空的错误警报率要求。为了改进使用的传统分类和冲突检测算法，将研究在意图分类和随后的潜在冲突检测及其解决中使用元启发式和深度学习（DL）技术。该解决方案将显著增强现有载人航空非合作雷达系统的能力，以及反无人机雷达系统，增强混合空域的安全性。

With the proliferation of Unmanned Autonomous Vehicles (UAV) applications (e.g. surveying, medical deliveries etc), systems and services to allow them to co-exist with manned aviation and to be used within controlled airspace are being developed. To ensure safety, it is paramount that aviation users and operators are alerted of potential conflicts between aircraft and between aircraft and UAVs as well. In Europe, so-called 'U3 advanced services' are being developed that include capacity management and systems that support conflict detection and resolution specifically to enable UAV operations. For the manned aviation case, a well-established concept known as Short Term Conflict Alert (STCA) is currently being used as a safety net to alert ATC operators of any situation where user defined minimum separation distances between any pair of cooperative surveillance (radar) tracks are predicted to be violated within a short look ahead time (usually 2 minutes). This is achieved via a visual alert on the radar display, though some systems also provide an audible alert. The system typically uses tracks based on cooperative radar, i.e. a system that acquires position information directly from a transponding aircraft. However, this solution cannot be applied to non-transponding air vehicles, and thus the future blended airspace of manned and unmanned air vehicles will need a more resilient and robust solution, to fully unlock the potential of UAV applications.This PhD proposes to develop a robust conflict detection and alerting mechanism that is based on non-cooperative radar signatures to classify the aircraft intent, as an automated way to improve safety in a blended airspace, whilst meeting the false alerts rate requirements of manned aviation. The use of Metaheuristics and Deep Learning (DL) techniques in the classification of intent and subsequent detection of potential conflicts and their resolution will be investigated, in order to improve the conventional classification and conflict detection algorithms used. The solution will significantly enhance the capabilities of existing non-cooperative radar systems in manned aviation, as well as counter UAV radar systems, enhancing the safety and security of the blended airspace.