Scene Geometry Aided Automatic Target Recognition (ATR) for Radar

OUSD (R&E) MODERNIZATION PRIORITY: Artificial intelligence/machine learning (AI/ML), autonomy TECHNOLOGY AREA(S): Information systems, sensors, electronics OBJECTIVE: Develop and demonstrate synthetic aperture radar (SAR) ATR that reduces false alarm rates by incorporating modern artificial intelligence and geometry of the imaged area. DESCRIPTION: The focus of this research incorporates geometry (building, tree, and road networks, etc.) of the imaging scene for radar ATR so that false alarms can be reduced. When an area of interest has been interrogated with SAR, the imagery includes the targets' signature, layover, and signatures of surrounding objects. All of the unwanted signatures (other than the targets' signature) contribute to false alarms. Hence, the goal of this research is to investigate novel radar ATR that reduces false alarms. Radar ATR technologies have evolved over time from one-dimensional signal (range profile) to three-dimensional (3D) signal (i.e., 3D imagery) or even four-dimensional information. ATR has also evolved from a template-based approach to modern AI/ML that is, deep learning based recognition. During ATR technology evolution, we have also seen significant improvement in classification accuracy. In particular, it was shown that target variations, articulations, and various operating conditions are problematic for the template-based ATR approach because this approach relied heavily on correlation templates. In a sense, the template-based approach works best when thousands of target templates are provided. Recent deep learning techniques overcame many of the shortcomings of the template-based approach. However, ATR technology with reduced a false alarm rate (FAR) is very important for precision target engagement. Along with advanced sensors (high resolution, multiple polarizations, etc.) and AI/ML technology, the foundation of the imaging scene may provide additional information to reduce FAR. It is important that researchers of this topic have significant experiences in SAR imaging (two-dimensional (2D) and 3D imaging), layover issue, moving target signature, radar clutter mitigation, and deep learning-based target classification. Understanding SAR datasets such as the Gotcha radar data from the Air Force Research Laboratory (AFRL), moving and stationary target acquisition and recognition (MSTAR) datasets, and implementing deep learning techniques to MSTAR targets will be helpful. As needed, the Government will work with the performer to find relevant synthetic and measured datasets. PHASE I: Research, develop, and demonstrate concepts for deep learning and scene geometry aided (SGA) ATR that contribute to reducing false alarms. PHASE II: Implement geometry-aided ATR algorithms using synthetic aperture radar datasets and imaging scenes. Evaluate performance of SGA ATR and quantify false alarm reduction. PHASE III DUAL USE APPLICATIONS: Transition SGA ATR technology by implementing the algorithms on relevant measured SAR datasets. REFERENCES: John F. Gilmore, Knowledge-based target recognition system evolution, Optical Engineering 30(5), 1 May 1991. https://doi.org/10.1117/12.55829. B. Bhanu, Automatic target recognition: State of the art survey, in IEEE Transactions on Aerospace and Electronic Systems, vol. AES-22, no. 4, pp. 364 379, July 1986, doi: 10.1109/TAES.1986.310772. U. K. Majumder, E. P. Blasch, and D. A. Garren, Deep Learning for Radar and Communications Automatic Target Recognition. Norwood, MA, USA: Artech House, 2020. Timothy D. Ross, Steven W. Worrell, Vincent J. Velten, John C. Mossing, and Michael Lee Bryant. Standard SAR ATR evaluation experiments using the MSTAR public release data set, Proceedings SPIE 3370, Algorithms for Synthetic Aperture Radar Imagery V, 15 September 1998. KEYWORDS: Automatic target recognition (ATR), synthetic aperture radar (SAR), false alarm rate (FAR), deep learning, artificial intelligence/machine learning (AI/ML), scene geometry, knowledge-based ATR