Integration of Automatic Dependent Surveillance

Extracting patterns from Automatic Dependent Surveillance-Broadcast (ADS-B) data to identify air corridors and detect anomalous behavior could provide crucial information for both commercial and military applications. Advancements in machine learning (ML) allow for identification of complex patterns and adaptive anomaly detection. Since flight routing is constantly changing due to temporal variables such as weather, identifying air corridors and detecting anomalous aircraft behavior requires an online approach that takes into account such variables. Utilizing miniaturized ML accelerated hardware, a pre-trained network can be used for online anomaly detection. With such hardware it is feasible to develop a small, inexpensive, ADS-B Anomaly Detector that operates independently or in a network, on stationary and mobile platforms. In the Phase I effort, Skyward, Ltd (Skyward) focused on obtaining data, processing data, selecting a machine learning architecture, identifying benchmarks, and evaluating detector options, with the goal of minimizing risk for the Phase II development.  In Phase I, Skyward developed a prototype anomaly detector unit and the corresponding software for anomaly detection and corridor recognition using ADS-B data collected in situ by Skyward.  In Phase II, Skyward will further develop the anomaly detection algorithms to improve performance.  The corridor recognition algorithm will be expanded with additional parameters to allow more complicated analysis. Skyward will develop software for automated ship-based or supercomputer-based training servers capable of supporting a limited or unlimited amount of training data, respectively.  Hardware prototypes will be developed to support ship-based, forward operations, and other roles. Skyward will also develop the needed software to allow the inclusion of real-time ADS-B anomaly detection and air corridor data into a database to subsequently be viewed in a common environment. Together these elements will form a scalable low-cost solution that can progress into a commercial product following Phase II.