Machine Learning for Wide Area Surveillance from Aerial Platforms

Legacy detection approaches in persistent wide area surveillance (WAS) rely on tractable stochastic models to predict clutter and target behaviour to allow the generation of detector structures. Unfortunately, actual clutter properties encountered during radar and optical WAS are frequently observed to diverge significantly from the proposed statistical models. This results in degraded detection performance. machine learning (ML) is proposed as a potential technique to help capture complex clutter behaviour so as to improve detection performance. In this paper some prior motivating applications of ML for WAS are discussed. Ongoing challenges in the application of ML techniques to WAS are identified along with recommendations for future implementations.