IEEE Boston Section Tech Site IEEE Logo

Serving Eastern Massachusetts

Signal Processing Society [SP-01]

6:00 PM, Tuesday, 15 April

Interrupted SAR: Sparse Reconstruction for Reliable Change Detection

Leslie M. Novak, Ph.D, Consultant, 1037 Main Street, Dunstable, MA, USA, novakl@charter.net

An important advance in today’s radar technology is the development of systems that use active array antennas; systems such as the APG-77 or F-35 JSF employ active arrays which allow the radar to accommodate multiple modes for search, track, target recognition, SAR/ISAR imaging, etc. These radar modes may compete for radar resources, resulting in severe timeline demands due to multi-mode operational requirements. One critical radar function is the collection of data to form high-resolution SAR imagery. Since the coherent integration time required to form such imagery may be several tens of seconds, the radar may not be able to dedicate an uninterrupted period of time of this magnitude solely for SAR data collection. Other high-priority radar modes must be performed in a timely manner, depending on the operational situation; thus it may necessary to interrupt the SAR data collection randomly or periodically to perform other radar modes. Such interruptions will leave data-gaps in the coherent SAR phase-history data which can significantly degrade the quality of the resulting SAR imagery. As a consequence, novel image formation algorithms are required to mitigate the artifacts introduced into SAR imagery as a result of interrupted SAR data collections.

This talk will focus on Compressive Sensing approaches for reconstructing complex SAR imagery from interrupted SAR phase-history data collections. A baseline approach using Basis Pursuit DeNoising (BPDN) is shown to accurately reconstruct SAR images from phase-history data having arbitrary data gaps, including single contiguous gaps, periodic gaps, random unit-width gapping patterns, and combinations of these gapping patterns. Accuracy of the reconstructed imagery is sufficient for performing SAR change detection (coherent and non-coherent change detection) as well as for visual target recognition by image analysts and computer algorithms. Excellent change detection performance results are demonstrated using BPDN-reconstructed two-pass high-resolution complex SAR imagery gathered by AFRL's Gotcha airborne sensor. Another approach extends the BPDN to one that jointly processes multi-pass change detection images using a group sparsity (GS) constraint; GS-processed reconstructions are shown to improve the performance of the baseline BPDN approach.

Figure 1: SAR images from Periodic Interrupts;
(Left) 2DFFT-processed image; (Right) BPDN-processed image
(Left) 2DFFT-processed image; (Right) BPDN-processed image

Figure 2: SAR images from Random Interrupts;
(Left) 2DFFT-processed image;  (Right) BPDN-processed image 
(Left) 2DFFT-processed image; (Right) BPDN-processed image

Compressed Sensing Image Reconstruction Example: Figure 1 shows an example of SAR images constructed from phase-history data that has been interrupted by a periodic gapping pattern; Figure 1 (Left Image) shows the SAR image obtained by processing the interrupted SAR phase-history data using conventional 2D-SAR image formation processing (Taylor side lobe weighting followed by 2D-FFT processing). Figure 1 (Right Image) shows the corresponding SAR image obtained by processing the interrupted phase-history data using BPDN image formation. Note that the BPDN-processed SAR image clearly shows a set of 5-vehicles located in a homogeneous clutter background (a grassy field). Also visible in the image are two high-quality corner reflectors (metal spheres); note the sharpness of these scatterers.

Les Novak received his PhD degree in Electrical Engineering from the University of California, Los Angeles. He worked at MIT Lincoln Laboratory from 1977 to 2003 where he held the position Senior Technical Staff in the Sensor Exploitation Group. Initially he worked on the development of target detection algorithms, target discrimination algorithms, and target classification/recognition algorithms for synthetic aperture radar (SAR) systems. He also performed studies of multi-polarization radar signal processing algorithms and super-resolution signal processing algorithms. He developed the full-polarization change detection algorithm (POLCD) used on the DARPA foliage penetration (WATCH-IT) program; the full-polarization change detection algorithm was transitioned to the foliage penetration airborne SAR system where real-time, in-the-air, full-polarization change detection was demonstrated; the change detection algorithm was transitioned by DARPA to the NGA Image Exploitation System.

From December 2003 to February 2006 he worked at Alphatech, Inc. (now BAE) in Burlington, MA where he developed a full-polarization coherent change detection algorithm -- and he also investigated the effects of SAR image quality on the performance of DARPA's model-based target recognition system (MSTAR); he demonstrated significantly improved target recognition performance achieved via improved image quality obtained by applying phase-gradient refocusing to cross-range blurred SAR imagery.

He joined Scientific Systems Company (SSCI) as an independent consultant in February 2006 where he performed research on SAR coherent and non-coherent change detection algorithms; he investigated the detection performance of these change detection algorithms using imagery compressed via a simple Block Adaptive Quantization algorithm versus various wavelet-based algorithms and Compressive Sensing algorithms. While at SSCI, Dr. Les Novak proposed and was awarded many Small Business Innovative Research (SBIR) programs -- he was principal investigator on numerous SAR-based SBIRs, including: (1) "Interrupted Synthetic Aperture Radar (SAR)", (2) "Cross-platform Image Quality Metrics for SAR ATR", (3) "Compressive Sensing for DCGS-N", and others. Les was also a lecturer for a NATO technical lecture series (2011-2013) on ATR and NCTR.

At Raytheon Company, Bedford, MA, he worked on the design of the digital signal processor for the Patriot Radar system; he also developed correlation algorithms for Raytheon's Pulse Doppler radar map matching system. At Hughes Aircraft Company he developed Extended Kalman Tracking Filter Algorithms for the TPQ-36 and TPQ-37 artillery and mortar locating radar systems.

Figure 2 shows an example of SAR images constructed from phase-history data that has been interrupted by a randomly interrupted pattern; as the figure shows, this phase-history data set has resulted in numerous cross-range streaks in the SAR image (Left Image). Again, the BPDN processed image (Right Image) has excellent image quality.

Meeting will be held at the MIT Lincoln Laboratory Cafeteria, 244 Wood Street, Lexington, MA

Directions to Lincoln Laboratory: (from interstate I-95/Route 128)

From Exit 31B

Take Exit 31B onto Routes 4/225 towards Bedford - Stay in right lane

Use Right Turning Lane (0.3 mile from exit) to access Hartwell Ave. at 1st Traffic Light.

Follow Hartwell Ave. to Wood St. (~1.3 miles).

Turn Left on to Wood Street and Drive for 0.3 of a mile.

Turn Right into MIT Lincoln Lab at the Wood Street Gate

Have a valid driver’s license to present to security.

From Exit 30B

Take Exit 30B on to Route 2A - Stay in right lane

Turn Right on to Mass. Ave (~ 0.4 miles - opposite Minuteman Tech.).

Follow Mass. Ave for ~ 0.4 miles.

Turn Left on to Wood Street and Drive for 1.0 mile.

Turn Left into MIT Lincoln Lab at the Wood Street Gate

HHave a valid driver’s license to present to security.

All attendees must present a valid driver's license to MIT Lincoln Laboratory security. To get to the Cafeteria, proceed toward the Main Entrance of Lincoln Laboratory. Before entering the building, proceed down the stairs located to the left of the Main Entrance. Turn right at the bottom of the stairs and enter the building through the Cafeteria entrance. The Cafeteria is located directly ahead.

Figure 1: SAR images from Periodic Interrupts; (Left) 2DFFT-processed image; (Right) PDN-processed image

Figure 2: SAR images from Random Interrupts; (Left) 2DFFT-processed image; (Right) BPDN-processed image