TECHNOLOGY AREA(S): Weapons

ACQUISITION PROGRAM: PMA 280 Tomahawk Weapons Systems

The technology within this topic is restricted under the International Traffic in Arms Regulation (ITAR), 22 CFR Parts 120-130, which controls the export and import of defense-related material and services, including export of sensitive technical data, or the Export Administration Regulation (EAR), 15 CFR Parts 730-774, which controls dual use items. Offerors must disclose any proposed use of foreign nationals (FNs), their country(ies) of origin, the type of visa or work permit possessed, and the statement of work (SOW) tasks intended for accomplishment by the FN(s) in accordance with section 5.4.c.(8) of the Announcement. Offerors are advised foreign nationals proposed to perform on this topic may be restricted due to the technical data under US Export Control Laws.

OBJECTIVE: Develop an innovative automatic target recognition (ATR) system that leverages state-of-the-art machine learning technology to automatically find and extract a ship�s salient features from its inverse synthetic aperture radar (ISAR) images for high-speed weapons applications.

DESCRIPTION: Despite significant DoD investments in radar target recognition over the past 30 years, very few radar automatic target recognition (ATR) systems have transitioned into widespread use in weapon applications.� The reasons are many including: (1) they are too computationally complex for weapon platforms (e.g., template matching); (2) legacy ATR algorithms cannot achieve the required recognition update rate of 5Hz or higher; (3) poor false-alarm performance resulting in inadequacy to sort among a complex scene in time for correct target engagement while avoiding collateral damage to non-targets; and (4) they cannot or have difficulty in evolving to learn to recognize new targets.� This requires significant time and data resources (e.g., significant amounts of measured training data and/or high-fidelity models).� Due to the nature of the problem and the technology sought, significant in this context, data sets could vary from tens of thousands to hundreds of thousands.� Another aspect of this problem is the need to determine the amount of data necessary to effectively train the algorithm for the task.

Recent results in machine learning, as applied to synthetic aperture radar (SAR) images of land targets, show that it is possible to reduce run-time storage and computational complexity of new recognition algorithms by 50- to 100-fold compared to conventional recognition algorithms.� Deep learning algorithms have successfully been used to adaptively determine the salient features of a target by researchers.� Furthermore, these salient features can be used to uniquely represent a target thereby increasing the utility of this capability.

The problem with current maritime ATR and maritime classification aide algorithms is the reliance on heuristic features, which are not sufficient for classifying vessels beyond using vessel length.� Heuristic image processing techniques that attempt to extract features (e.g., superstructures) break down when the ISAR image quality degrades.� The reasons for the degradation vary but include: low Signal to Interference plus Noise Ratio (SINR), poor viewing geometry, poor environmental conditions, and insufficient observation time.� Moreover, as new data is observed in different operating conditions heuristic image processing techniques will require constant manual retraining to discover the required features.� Furthermore, heuristic approaches require identification of the features a priori with development of the image processing technique necessary to extract the features.� Finally, in choosing these features and developing the image processing technique first, the hardware is limited to relying on these features.� Ultimately, we need to identify the other features that technology can leverage to differentiate similar vessels.� By using advanced machine learning technology to develop an innovative approach that automatically finds the target�s salient features, we may find that we can differentiate similar vessels.� The technology must be robust to deliver image quality without reliance on heuristic image processing approaches.� The technology must also ensure that as the image quality degrades the algorithm can still function well to perform its classification task.

The end-product should be a weapon flight-tested hardware implementation of the ATR algorithms running in real-time on a low-power processor that meets the Government performance objectives.� The algorithm design must be modular and adaptable to adding new ships and target classes on a frequent schedule.� Once the core training is completed, that solution will be adequate for a physical ship design until the ship class is decommissioned.� Assessment of the ATR algorithm will be a function of ship type, operational environment (e.g., sea state, wind condition, etc.), radar parameters (e.g., bandwidth, frequency, etc.), and ISAR image quality obtained from different missile trajectories.

Work produced in Phase II may become classified. Note: The prospective contractor(s) must be U.S. owned and operated with no foreign influence as defined by DoD 5220.22-M, National Industrial Security Program Operating Manual, unless acceptable mitigating procedures can and have been implemented and approved by the Defense Security Service (DSS). The selected contractor and/or subcontractor must be able to acquire and maintain a secret level facility and Personnel Security Clearances, in order to perform on advanced phases of this project as set forth by DSS and NAVAIR in order to gain access to classified information pertaining to the national defense of the United States and its allies; this will be an inherent requirement. The selected company will be required to safeguard classified material IAW DoD 5220.22-M during the advanced phases of this contract.

PHASE I: Leverage machine learning concepts and technology to automatically identify and extract key ship salient features from simulated and/or unclassified ISAR imagery and use these features to demonstrate robust automatic target recognition of the vessels.� Assessment of the algorithms will be performed against simulated and test radar data to identify the performance expectation against the ISAR image quality obtained from a missile platform.� Develop prototype plans for Phase II.

PHASE II: Further develop and demonstrate the performance of the prototype ATR system against collected radar data of maritime targets from applicable/relevant weapon radar seekers.� Perform ATR performance assessment as a function of ship type, operational environment (e.g., sea state, wind condition, etc.), radar parameters (e.g., bandwidth, frequency, etc.), and ISAR image quality obtained from different missile trajectories.� Once demonstrated, a key task is to develop real-time embedded software code of the ATR system and map the processing requirements for candidate processors selected by the PMA for flight testing demonstration.

It is probable that the work under this effort will be classified under Phase II (see Description section for details).

PHASE III DUAL USE APPLICATIONS: Finalize the hardware testing for the software code of the ATR system for the candidate processors.� Support integration on available weapon hardware that will be ready for integration with software.� The testing will include Hardware-in-the-Loop (HWIL) testing with synthetic target scene generation and a flight test to verify and validate performance.� The ATR algorithm would be beneficial to the Coast Guard for maritime target recognition at range in addition to any other applications that would require target recognition using ISAR imaging in a maritime environment, including military targeting and sensor aircraft.� This technology could also benefit those who need to track shipments and could provide improvements to facial recognition via algorithm discovery.

REFERENCES:

1. Michie D., Spiegelhalter, D., & Taylor C. (eds). Machine Learning: Neural and Statistical Classification, 1994. http://www1.maths.leeds.ac.uk/~charles/statlog/

2. Chen V. and Martorella M. Inverse Synthetic Aperture Radar Imaging: Principles, Algorithms, and Applications, 2014. https://books.google.com/books/about/Inverse_Synthetic_Aperture_Radar_Imaging.html?id=xWmABAAAQBAJ

3. Barth K., Bruggenwirth S., and Wagner S. �A Deep Learning SAR ATR System Using Regularization and Prioritized Classes.� IEEE Radar Conference, 2017. http://ieeexplore.ieee.org/document/7944307/

4. Li J., Mei X. and Prokhorov D. �Deep Neural Network for Structural Prediction and Lane Detection in Traffic Scene.� IEEE Transactions on Neural Networks and Learning Systems, Vol 28, Issue 3, March 2017. http://ieeexplore.ieee.org/document/7407673/

5. Ji K., Kang M., Leng X., et al. �Deep Convolutional Highway Unit Network for SAR Target Classification with Limited Labeled Training Data.� IEEE Geoscience and Remote Sensing Letters, Vol PP, Issue 99. http://ieeexplore.ieee.org/document/7926358/

6. Nguyen A., Xu J. and Yang Z. �A Bio-inspired Redundant Sensing Architecture.� 30th Conference on Neural Information Processing Systems (NIPS 2016), Barcelona, Spain. https://papers.nips.cc/paper/6564-a-bio-inspired-redundant-sensing-architecture.pdf

7. Cheng Y., Lu W., Zhai S., et al. �Doubly Convolutional Neural Networks.� Advances in Neural Information Processing Systems 29 (NIPS 2016). http://papers.nips.cc/paper/6340-doubly-convolutional-neural-networks

KEYWORDS: ISAR; Automatic Target Recognition (ATR); Machine Learning; Maritime; Deep Learning; Image Processing