Maritime Target Classification from Inverse Synthetic Aperture Radar (ISAR) Using Machine Learning/Toyon Research Corp.

Toyon Research Corp. proposes to develop innovative algorithms to perform automatic target detection (ATR), localization, classification, and sub-classification of maritime targets in ISAR imagery. The proposed algorithms are based on a careful analysis of reflection-mode ISAR imaging systems that aims to address complications that are unique to the images produced in this modality, such as poor resolution due to motion and range uncertainty in various environmental or constrained observation geometries. Specifically, this proposal presents a novel method for the direct processing of sparse spatial-frequency samples (e.g. I/Q pairs) using neural networks that is expected to significantly increase the performance of ATR algorithms in scenarios involving airborne radar platforms that experience limited observation time along their flight path, such as for airborne surveillance or weapon systems. The resulting system is highly tunable, via a number of objective-function parameters, and adaptable to various target geometries, environmental conditions, and radar bands. Furthermore, the proposed deep learning system is uniquely capable of real-time performance, with immediate extensions amenable to massively parallel GPU, CPU, and FPGA hardware implementations.