You are here

Innovative Approaches for Aided Target Recognition (AiTR) of Army Targets


TECHNOLOGY AREA(S): Electronics, 

OBJECTIVE: Develop approaches for AiTR that advance the state-of-the-art for specific military applications, sensors, and operating conditions. 

DESCRIPTION: In recent years the research area of computer vision, a subfield of Artificial Intelligence (AI), has received significant attention and investments from both commercial and Department of Defense (DoD) sources. The principal trend has been to use deep learning-based approaches to automate both feature generation and object classification. While this approach has many advantages, especially in commercial applications, it has many disadvantages and limitations in DoD problem spaces. This is due largely to dynamic and unpredictable operating environments, uncooperative targets (e.g., partially concealed or in tree lines), and of course a shortage of data relevant to the problem space. The Army desires innovative approaches for improving performance, robustness, and/or training efficiencies for Army AiTR systems. Possible improvements could be, but are not limited to the following: • Improved probability of detection and false-alarm rates for AiTR algorithms against Army relevant targets with infrared (IR) imagery • Approaches to train algorithms using reduced amounts of data (e.g., transfer learning, synthetic data) • Manually created features that perform as good, or better, to deep learning automated features against Army relevant targets with IR imagery • Algorithm improvements to increase robustness to unpredictable and untrained on environments and backgrounds • Methods to quickly update a trained algorithm to a new target of interest without requiring the algorithm to retrain on previous target data Solutions proposed to this topic should describe an innovative and novel approach to current state-of-the-art AiTR algorithms. The proposal should explicitly describe the innovative aspect of the proposed solution and specify how and why this innovation is helpful to the Army AiTR mission. Where possible provide quantitative metrics. 

PHASE I: The proposer shall complete a proof-of-concept AiTR algorithm. This proof-of-concept shall demonstrate the proposed solution and the improvement it provides against the current state-of-the-art. For this phase the proposer shall use their own data. Government Subject Matter Experts (SMEs) will evaluate the proposer's design and results to determine utility against Army problem sets. 

PHASE II: In Phase 2 the proposer will be provided IR data against targets and environments relevant to Army operations. During this phase the proposer will mature the AiTR approach to a TRL 6 level. In this phase the proposer shall evaluate improvements provided by this solution over current state-of-the-art commercial approaches. Army SMEs will evaluate the final solution against an Army sequestered dataset to determine improvements over current Army approaches. 

PHASE III: Transition the developed approach to Army programs of record (PORs) and Army Futures Command (AFC) Cross Functional Teams (CFTs). In this phase the algorithms will be integrated into on-board processing hardware and platform software systems. Additional maturation of the algorithms using actual platform sensor data will take place. 


1: Haykin, S., [Neural Networks and Learning Machines], Pearson Education, Inc., (2009)

KEYWORDS: ATR, AiTR, Artificial Intelligence, Machine Learning, Computer Vision, Image Processing 

US Flag An Official Website of the United States Government