Aerial Intruder Interception Based on Threat Classification for Enhanced Situational Awareness

Intent classification of Unmanned Aerial Vehicles (UAVs) either as a standalone vehicle or in swarm is extremely important in the current world. With the rise in autonomy of these platforms, there have been several instances where UAVs have, instead of providing benefit, been a nuisance and created chaos and confusion. Activity classification for threat assessment of UAVs is a relatively new area where different types of machine learning models are been tested. In this paper, the authors have provided a methodology to passively perform intent classification and a comparative study of different classification models namely Hidden Markov Model (HMM), Convolutional Neural Network - Long Short Term Memory (CNN-LSTM), Gramian Angular fields (GAF) with CNN and Transformers have been performed for action classification for threat identification. Additionally, the authors have developed a simulator using ROS and Gazebo to deploy these models in simulation along with a resource allocation and target interception framework to demonstrate an actionable and intelligent as well as automated Counter-UAS system.