Identifying Drone Operator by Deep Learning and Ensemble Learning of IMU and Control Data

Task-Incremental Learning for Drone Pilot Identification Scheme

“…Specifically, the modules which want to utilize the drone flight attitudes for computation first subscribe To address the challenge mentioned above, we design a novel task-incremental learningbased drone pilot identification scheme. Motivated by the previous works [12,13,23], we first design a background service to collect drone flight data by subscribing to the topics from a micro object request broker (uORB) message bus. We then construct a module for extracting pilot behavioral traits from the flight data and establish a mapping between the extracted pilots' behavioral traits and their provided identities.…”

“…Despite the significant progress that has been made to reduce drone pilot impersonation attacks, a critical challenge still needs to be addressed: drone pilot membership dynamic management. Unlike the protocol-based authentication scheme, the ML-based identification scheme mentioned in the previous works [11][12][13] could not adapt to newly joined pilots for identification and authentication, only succeeding with the help of the current pilot's flight data. As illustrated in Figure 1, once the well-established pilot identification scheme is deployed on the flying drone, only the previous registered pilot's legal status can be verified during the flight procedure.…”

“…Through comparing with the state-of-the-art authentication protocols [9,10], the authors verified that their proposed scheme could well resist masquerade, replay, node tampering, and cloning attacks over the drone communication channel. On the other hand, many machine learning (ML)-based identification schemes [11][12][13] have also been designed for ensuring drone pilot legal status during the flight procedure. For example, shouFan et al in [11,12] first verified the pilot's legal status by monitoring the remote radio controller sending instructions.…”

“…After analyzing extracted control commands with Linear Discriminant (LD) [14], Quadratic Discriminant (QD) [15], Support Vector Machine (SVM) [16], weighted k Nearest Neighbors (k-NN) [17] and Random Forest (RF) [18] the best performance was achieved with RF approximating to 89% in accuracy. Alkadi et al in [13] further combined the onboard sensor measurements and received control instructions for improving drone pilot identification performance. Through analyzing with Long Short-Term Memory (LSTM), feature-based, and majority voting-based classification algorithms [19][20][21], the authors concluded that the combination methods could enhance the pilot identification performance.…”

Enhancing Data Discretization for Smoother Drone Input Using GAN-Based IMU Data Augmentation

“…Regarding IMU sensors, researchers have focused on enhancing data processing techniques and addressing the limitations associated with discrete measurements. Alkadi [13] explored using machine learning techniques to continuously authenticate pilots through sensor data and control signals obtained from drones, aiming to enhance the security of drone flight controls and ground stations. Ochoa-de-Eribe-Landaberea [14] introduced a novel landing assistance system for drones that utilizes a fusion of ultra-wideband (UWB), IMU, and magnetometer data to accurately locate the drone for safe landings, with improved performance compared to traditional UWB-based systems.…”