READS: Resource efficient attack detection system for drones

Unmanned Aerial Vehicles (UAVs), commonly known as aerial drones, have gained popularity due to their versatility and applications in various industries. However, their increasing use has raised concerns about cybersecurity and the vulnerability of networked UAVs to malicious attacks. A deep learning-based Attack Detection System (ADS) is essential for mitigating cyber threats to UAVs. It can provide proactive protection even against unknown attacks. Also, for resource constraints of UAVs, ADS must be resource-efficient to optimize the deep learning model for embedded hardware, making a low code footprint. Therefore, this work focuses on developing a Resource Efficient Attack Detection System (READS) for drones that uses a lite deep learning-based network activity classification algorithm for detecting cyber threats on UAVs. The entire system has been implemented in three phases, where the first phase includes dataset training. The second phase involves running multiple conventional models to choose the best conventional model for intrusion detection using accuracy, precision, and recall. This is followed by phase three, which converts the selected conventional Dense Neural Network (DNN) model into a lite weight (Tiny) model that works on UAVs. The selected Lite DNN model is evaluated on a real-time testbed. The optimized model could lower the storage space and generates 97.69% accuracy. The proposed system consumes 30% of the original power. Hence, it allows efficient threat detection while minimizing the impact on UAV resources.