3.4 Automatic Target Recognition
Automatic target recognition (ATR) addresses the following JWOC: (1) Information Superiority—ATR gives real-time identification of an adversary from high-bandwidth sensors (providing sufficient knowledge to neutralize the enemy) and yields enormous data compression for transmission on battlefield datalinks; (2) Precision Force—ATR's real-time identification of forces over a wide area compresses the C4I timeline for responsive sensor-to-shooter operations and enables timely reacquisition of target for strike platform; (3) Combat Identification—ATR gives beyond-visual-range ID to launch missiles at long range, enabling lethal enemy engagement, reduced fratricide, and ownship survival; (4) Joint Theater Missile Defense—ATR enables finding ground-based missile launchers in a timely manner consistent with elusive adversary operations, discriminating between RVs and decoys during reentry, and discriminating between cruise missiles and slow-moving, low-flying confusers; (5) Military Operations in Urban Terrain—ATR enables finding targets in cluttered urban environment to precisely identify targets allowing precision weapon employment resulting in minimal collateral damage; (6) Joint Readiness and Logistics—ATR synthetic scene generation and modeling provides capabilities for enhanced simulation and training; (7) Joint Countermine—ATR technology enables rapid detection of mines; and (8) Counter Weapons of Mass Destruction—ATR aids in timely bomb damage assessment. ATR is needed for both ISR and weapon-delivery systems. Transitions planned for JSTARS, P-3, S-3, U-2R, Tier 2, Tier 2+, Tier 3-, F-14, F-15, F-16, F-18, F-22, Apache, Comanche, AWACS, Abrams, Bradley, MSX, THAAD, Destroyer, CG-47, DDG-51, DDG-993, and DD-963.
The ATR subarea is related to battlespace environments and sensors of this DTAP. Knowledge of battlespace environments can facilitate ATR evaluation and utility studies. Sensors and ATR are tightly intertwined as sensors provide the data for ATR development, and ATR is a key objective of sensor design. The ATR subarea also leverages computing technologies reported in the Information Systems Technology area (Chapter III) and the ATR work reported in the Weapons area (Chapter X).
3.4.2.1 Goals and Timeframes. The ATR program goals are grouped into two categories: those organized by target class (land, sea, air) driven by the need to improve ATR performance, and those that are general ATR goals driven by the need to reduce both acquisition and life-cycle costs. The goals and timelines for ATR are shown in Table VII-5.
| Fiscal Year | Goals |
|---|---|
| FY97 | Ground targets—open targets/standard configuration. Airborne targets—10 target types. Surface targets—100 large combatant ship classes. Reentry vehicles—discriminate debris. Target insertion—2 months for classification, 6 months for ID.* |
| FY02 | Ground targets standard configuration—up to 30% target obscuration; 150x search area. Airborne targets—35 target classes. Surface targets—small craft; 20 classes. Reentry vehicles—discriminate crude decoys. Target insertion—24 weeks for target classification, 6 weeks for target ID.* Affordability—2x-4x reduction in development time, 2x reduction in software costs, 10x reduction in hardware costs. |
| FY07 | Ground targets—multiple target configurations including articulation, light CC&D;
1,000x search area. Airborne targets—100 target classes. Surface targets—small craft; 100 classes. Reentry vehicles—discriminate sophisticated decoys. Target insertion—48 hours for target ID.* Affordability—6x-10x reduction in development time, 6x reduction in software costs. |
*Varies significantly depending on whether target is captured or denied, target complexity, sensor, and recognition approach.
3.4.2.2 Major Technical Challenges. The major technical challenge for ATR is contending with the combinatorial explosion of target signature variations caused by permutations of target configuration (e.g., stores, articulation, manufacturing, wear/tear), target/sensor acquisition parameters (e.g., aspect, depression, squint angles), target phenomenology (e.g., cavity responses, glints, IR thermal behavior), and target/clutter interaction (e.g., foliage masking, camouflage). ATR systems must maintain low false alarm rates in the face of varying and complex backgrounds, and they must operate in real time. Another extremely important challenge for ATR is the evaluation and prediction of ATR field performance given the practical limitation that data sets cannot represent the extreme variability of the real world. The ability to rapidly insert new targets and to train algorithms on the fly in the field to support flexible and sustained employment of ATR are important challenges. A key technical challenge is the development of affordable ATR solutions that employ an open architecture. This will provide capability growth via expandable hardware and software insertion.
3.4.2.3 Related Federal and Private Sector Efforts. Image processing technologies are used in medical imaging, law enforcement, automated manufacturing, transportation sensing, remote sensing, environmental sensing, robotics, and multimedia. Commercial computer technologies are leveraged as well.
Technology is divided into four areas. Note that high-performance computing is a key enabler for each technical area. Also note that the Moving and Stationary Target Acquisition and Recognition (MSTAR) technology development is advancing the state of the art in all four areas.
Algorithm developments address the key technical challenge—combinatorial complexity of ATR. Approaches include the development of both data-driven and model-based approaches using single and multiple radar and EO sensors. Developments include multidisciplinary technologies utilizing advances in signal processing, decision and estimation theory, artificial intelligence, operations research, and computer science. Key programs include MSTAR, image understanding program, imaging ATR, and multisensor fusion.
Affordability developments leverage open architecture initiatives and design tools. Algorithm tools focus on a common environment to reduce ATR development and evaluation cost and improve algorithm performance via shared and distributed algorithm design, reuse of software, and decoupling of software development from real-time high-performance computer (HPC) architectures. The key program activity is integrating the development of Khoros and the image understanding environment as the ATR standard. The hardware component of affordability is addressed by leveraging commercially developed multichip modules to design and demonstrate a family of affordable, miniaturized, high-density, high-performance image and digital signal processors. These efforts use HPC processors and RASSP-developed design tools to meet cost and performance objectives.
Database development, including signature modeling and scene synthesis efforts, is the backbone of ATR progress. Databases support the development and evaluation of ATR algorithms for single/multisensor EO and radar systems. Signature modeling is critical to rapid target insertion capability and provides a cost-effective complement to measured data to evaluating multispectral ATR. Synthetic data also provide a practical means of exploring complex multi-sensor ATR design spaces. Scene synthesis efforts uniquely provide high-fidelity models for distributed, interactive simulations to assess the warfighting payoff of new technologies such as ATR, including performance of postulated advanced sensor systems against future conventional and low-observable (LO) threats. Key developments include Electronic Terrain Board, Xpatch, and Creation.
Scientific evaluation promotes accelerated and orderly ATR development by providing statistically significant performance feedback that pinpoints algorithm deficiencies to developers and provides valid field performance data to the respective service users for use in transition decisions. Scientific evaluation includes the development of performance estimation/bounding theories to guide ATR development. Strong ATR predictive theories will provide developers the tools to focus efforts at the knee of the curve in the highly complex ATR design space. Standard metrics and evaluation procedures are jointly developed as part of the efforts of the ATR Working Group. Collaboratively developed evaluation methodologies and shared data sets enable direct comparison of algorithms/processors among developers across the services and development agencies.
3.4.3.1 Technology Demonstrations.
Semiautomated Imagery Processing ACTD (JWSTP Information Superiority DTO A.09). This technology demonstration develops template-based ATR coupled with terrain and force structure analysis, object-level change detection, image/map registration, human computer interface, and interactive target recognition. Model-supported exploitation technology developed by the RADIUS program will be integrated and applied to site monitoring. ATR-aided data compression technology developed by the Clipping Service program will be used to reduce datalink requirements. The final demonstration delivers software modules integrated into imagery exploitation migration systems. Key imagery platforms include Tier 3- and U-2R.
Reentry Vehicle Discrimination Technology Demonstrations. These efforts develop discrimination algorithms to separate RVs from debris and sophisticated decoys using radar and EO sensors. Collection and analysis of missile flight data are used to validate signature models and develop/evaluate discrimination algorithms.
Advanced ID ATD (JWSTP Combat Identification DTO C.03). The objective of this DTO is to develop radar signal processing algorithms that provide reliable identification of noncooperative maneuvering aircraft at all target/sensor aspect angles from long standoff ranges. The technology approach is to develop algorithms to continually adapt radar's ID processing to target dynamics and mission demands on the radar system by fusing multiple radar/RF modes—ESM, RSM, and HRR and by performing advanced ISAR imaging techniques via an adaptive range/Doppler imaging process. Potential technology insertion platforms include F/A-18, F-14, F-15, F-16, F-22, and AWACS.
3.4.3.2 Technology Development.
Affordable ATR via Rapid Design, Evaluation, and Simulation (DTO SE.19.03). The objective of this DTO is to reduce the cost and development time for ATR systems including single and multisensor ATRs for land and air targets. Technology advancements will be made in the areas of high-fidelity, real-time synthetic signature and scene simulation; image and performance evaluation metrics, standards, facilities, and tools; large, high-quality, ground-truthed, multi- sensor databases; algorithm development tools and environments; integrated design environments, and high-performance computing. Standardized methodologies and databases will be integrated with industry and academia via the ATR Working Group.
ATR for Reconnaissance and Surveillance (DTO SE.20.01). This DTO will develop the capability to automatically recognize targets using high-range-resolution radar and ISAR for moving targets and high-resolution SAR for stationary targets. Advances in high-resolution imaging for both stationary and moving targets and advances in hybrid ATR algorithms using elements of both template and model-based approaches will be developed. This advanced imagery exploitation capability will be demonstrated using imagery from a number of reconnaissance/ surveillance platforms to meet service-specific exploitation needs. Demonstrations are planned for JSTARS, U-2R, Predator, S-3, and P-3.
3.4.3.3 Basic Research. ATR is a key focus for the 6.1 community. Basic research investment in ATR subarea technologies is estimated at $11 million. Important research themes include multi-resolution processing, fusion, advanced and nonlinear signal processing, computational electromagnetics, algebraic invariance, artificial intelligence and knowledge-based systems, advanced imaging techniques and inverse processing, and distributed/parallel computing. Recent key initiatives include the reduced signature target recognition effort focused on advanced algorithm and computational electromagnetic research; the Federated Lab effort focused on research partnerships between university, industry, and service labs; the signal processing and AI program; the single-sensor ATR program; and the Center for Imaging Science and the sensor processing program.
