Contract Year 1 Review Signal and Image Processing

1
Contract Year 1 ReviewSignal and Image Processing

• Dr. Stan Ahalt
• The Ohio State University
• 15 May 2002

2
SIP Strategy

• SIP consists of three communities 1)
  real-time/embedded 2) algorithm develop test,
  and 3) simulation and verification.
• The HPC needs for each community differ
  sometimes dramatically. Further, this community
  is spread among MSRCs, DCs,and the service labs
  and test facilities.
• Our strategy is to to serve the intersection of
  these three communities, AND to serve the special
  needs of each community where resources allow. 

3
SIP Strategic Issues

• New algorithm design, for conditioning,
  processing and information extraction.
• Improved transmission and storage of data, since
  sensors today can produce copious amounts of
  data.
• Improved processing speed, in the design,
  simulation, verification, and implementation
  phases (real-time requirements).
• Incorporating life cycle software tools to manage
  exploration, design, implementation, deployment
  using a common framework.
• Increased use of COTS tools.
• Training in new software tools and methodologies.

4
SIP Strategic Directions

• The SIP team serves its communitys objectives by
  striking a balance between
• specialized training e.g., MATLAB and Simulink
  as life cycle tools.
• leadership through outreach, dissemination, and
  demonstration DCs, HPEC, VSIPL communities,
  CHSSI.
• significant project delivery image compression,
  3D SAR image formation, featurebased ATR.
• long-term vision XML and object oriented
  approaches for data transmission, distribution,
  modeling.

These components have been implemented in Y1 as
follows
5
FAPOC core activities as compared to SOW

• Section 1.1.5
• Technical support e.g., assistance to 46th Test
  Wing.
• Productivity enhancements e.g., MATLAB/MPI
  3DSAR.
• Training Content e.g., MATLAB, Simulink, Signal
  Mining, Image Processing.
• Community Development e.g., SIP Forum, GOMAC,
  HPEC
• On-Site Support Aug 01 at ARL/Adelphi, ASC on
  Oct 01
• Single POCs Ahalt
• Section 2.4.3, Task 8
• Technology transfer/transition, with
  recommendations to the Government e.g.,
  Simulink VSIPL
• Coordination with other FAs e.g., SIP/CEA/CEN
  Forum and with OKC (requirements and early
  testing).
• Ensure full MSI participation submitted
  independent proposal Y1, not funded submitted
  two integrated proposals Y2
• Informal trip reports completed to date.
• Presentation materials transmitted to Government
  completed to date.

6
User Codes Impacted by SIP

• sar3d unaliased Rewrote and optimized baseline
  3dsar MATLAB code to interact with HPC platforms
• sar3d aliased increased functionality by
  compensating for aliased data. Provided a
  calibration preprocessing option
• MATLAB/MPI new code originating from MIT/LL.
  Added security features, tested on alternative
  platforms and OSs
• VSIPL ATR systems require response times of
  military utility (1 minute or less). VSIPL is
  being tested as a portable solution.
• Hyperspectral Learning Vector Quantizier
  Algorithm ATR Codes (POC Bruce Weber and Joseph
  Penn (AMSRL-SE-SE / ARL-ALC)
• Synthetic Imagery code, CREATION (POC Bruce
  Weber and Joseph Penn (AMSRL-SE-SE / ARL-ALC)
• G. Stolovy HIE Code from CHSSI HIE portfolio

7
User Contacts and Assistance

• Contacted and assisted approximately 75 users
  across 3 services, various defense industries,
  and multiple universities.
• AFIT, Univ. Michigan, Purdue, MIT, Univ. Buffalo
• AFRL (22)
• ALC
• ARL(6)
• ASC(4)
• DIA
• LLNL (2)
• SPAWAR(3)
• 46th Test Wing
• Ball, Boeing, Lockheed Martin (3), Mathworks,
  Miltec

8
Tools introduced to community

• MATLAB/MPI enhanced and extended original
  toolbox introduced by Jeremy Kepner (MIT/LL).
• Content-Based-Compression SAR image compression
  tool.
• Junshui Ma, Yi Zhao, and Stanley Ahalt, "OSU SVM
  Classifier Matlab Toolbox",available at
  http//eewww.eng.ohio-state.edu/maj/osu_svm/.
• Video Annotation codes prototype (R. Williams,
  SNAS).
• VSIPL installed and enhanced on ARL MSRC
  platforms.
• Tested MATLAB 6.1 for CSC.
• MATLAB compilers added to SGI, COMPAQ, and IBMs.

9

• Presentations
• Ahalt received a letter of appreciation for a
  paper presented at the SCI2001, 5th World
  Multiconference on Systemics, Cybernetics, and
  Informatics, July 22-25, 2001, Orlando, FL. The
  paper was cited as one of the best presented
  papers.
• J. Nehbass made presentation to approx. 20 MATLAB
  users from SIMAF and NAIC at MATLAB Day held at
  WPAFB.
• Conferences  (Total of 9 papers 2 additional
  SPIE papers plus 4 UGC papers)
• H. Li, J. Ma, Y. Zhao, and S. C. Ahalt,
  Kernal-based Feature Extraction, Proceedings of
  SPIE Vol. 4726, Automatic Target Recognition X!!,
  Aerosense, April 2002, Orlando, FL
• Y. Zhao,, H. Li, and S. C. Ahalt, Multiclass
  Kernal-based feature extraction, Proceedings of
  SPIE Vol. 4730, Data Mining and Knowledge
  Discovery Theory, Tools, and Technolgy IV,
  Aerosense, April 2002, Orlando, FL
• Workshops
• SIP Forum 2001 held in conjunction with GOMAC
  2001.
• Chaired Session in MATLAB on many Processorsat
  HPEC 2002.
• Seminars
• S. Ahalt invited speakerVideo Annotation and
  Query, Madrid, Spain, Oct. 2001.
• Training
• Introduction to MATLAB, ASC, Feb, 2002 (
  students 21) 6 additional courses scheduled.

10
Technical Reports and Articles

• Junshui Ma, Xun Du, and Stanley C. Ahalt, 2D HRR
  Radar Data Modeling and Processing, accepted for
  2002 publication in Multidimensional Systems and
  Signal Processing Special Issue on Radar Signal
  Processing and its applications.
• Adriana Dapena and Stanley C. Ahalt, An hybrid
  DCT-SVD image coding algorithm, in IEEE
  Transactions on Circuits and Systems for Video
  Technology, vol. 12, no. 2, pp 114-121.
• Jayanth Anantharaman, Ashok Krishnamurthy and
  Larry Feth, Spectral Centroid Effect in Dynamic
  Tones, accepted for publication in Journal of
  the Acoustical Society of America, 2002.
• S. Moore, D. Pressel, J. Chaves, Reality Check on
  OpenMP Implementations, ARL Technical Report
  (2001).
• Larry A. Lambe, Richard Luczak, and John W.
  Nehrbass, Symbolic Computation in
  Electromagnetic Modeling, in preparation.

11
SIP 003 Detection, Classification, and
Recognition

• Fundamental problem in ATR is number of in-class
  target variants, and fully characterizing all
  possible targets is infeasible. Challenge is to
  deal with extremely large search space, and
  inferring a class or target ID from relatively
  sparse data.
• Solution identify and refine critical components
  and parallelize

• Funding 79,775
• Expenditures 38 (Mar 02)
• 1 Nov 2001 30 Sep 2002
• PI Stan Ahalt

• Deliverables and Timetable (note dates reflect
  original Sep 01 start date)
• Feature Based ATR
• Identify and code components green
• XML (7/1) yellow Work initiated 4/1.
• Change detection study (6/1) green
• ATR MATLAB/Simulink Library
• Identify and develop (8/15) green

SAR image of tank, Compressed using CBC.
12
SIP 005 Image Compression, Detection, and
Formation

• Funding 103,474
• Expenditures 39 (Mar 02)
• 1 Nov 2001 30 Sep 2002
• PI Stan Ahalt

• Objective In-class target variants make full
  characterization of targets is infeasible.
  Thus, in order to deal with deal with large
  search space, we seek methods to simplify and
  regularize common ATR tasks to achieve robust
  solutions.
• Soln identify reusable components and
  parallelize, test and refine.

• Deliverables and timetable
• Reports on code scalability (5/1) green
• Parallel codes (8/15) green
• Documentation of codes (8/30) green
• docs on platforms (9/01) green
• docs for compression codes 8/01 green

13
SIP 006 Parallel MATLAB/Simulink

• Funding 99,160
• Expenditures 29 through Mar 02
• 1 Nov 2001 30 Sep 2002
• PI Stan Ahalt

• Issue MATLAB and Simulink are widely used
  development packages inside and outside of the
  DOD. But, no well supported, full-featured,
  parallel versions of either MATLAB or Simulink.
• Soln promote set of technical solutions that
  realize a viable, complete, parallel, and
  portable version of MATLAB and/or Simulink.

• Deliverables and timetable
• advisory panel and meeting trip reports green
• report on Simulink and graphics standards (3/1)
  red. Work has been initiated. Stateflow needs
  to be incorporated. Spending profile projections
  reflect end-load of effort.
• codes and performance analysis 6/30 green

14
SIP 007 Standards and Middleware

• Objective Middleware is infrastructure. SIP
  community has made only nascent attempts at
  developing a collection of middleware standards
  that are designed and tuned to SIPs unique
  problems.
• Soln support middleware that provides SIP
  functionality (VSIPL), real-time and QoS
  c-constrained communications (MPI-RT, RT-CORBA),
  and embeddable and portable SIP coding standards
  (HPEC).

• Funding 101,170
• Expenditures 18 through Mar 02
• 1 Nov 2001 30 Sep 2002
• PI Stan Ahalt

• Deliverables and timetable
• trip reports green
• proto implementation of codes (4/15) greenJCC
• Develop/refine parallel hyperspectral codes
  (6/15) green
• Project is end-loaded to expend funds

15
SIP 008 Training

• Objective Development and delivery of training
  courses in both software and techniques that have
  been identified as being critical for research
  and development activities in the SIP community.

• Funding 90,788
• Expenditures 33 (Mar 02)
• 1 Nov 2001 30 Sep 2002
• PI Stan Ahalt

• Deliverables and Timetable
• MATLAB (2) green
• Simulink green
• Signal/Image Processing green
• Signal Mining green
• SIP Forum 2002 green

From A.K.Kevorkian and L.A.Parnell, DoD High
Performance Computing Modernization Program
Nurtures Skills At SSC San Diego, 2001.
16
SIP 017 Access, Distribution, and Mining of SIP
Data

• Funding 51,423
• Expenditures 57 (Mar 02)
• 1 Nov 2001 30 Sep 2002
• PI Ashok Krishnamurthy

• Objective SIP and IMT communities are awash in
  data. Tools required to access, distribute and
  process this data are inadequate.
• Soln (1) use XML as a common language for
  describing, distributing and visualizing SIP
  data, (2) develop data mining techniques.

• Deliverables and timetable
• Documentation for XML format for
• ATC data green.
• MATLAB codes for parsing XML
• documents green.
• MATLAB codes for data mining green.
• MATLAB to output XML green.
• codes for visualizing XML green.

17
Core Financial Summary
18
SIP Invoice Status
CLIN 1 CLIN 1 CUM CLIN 4 CLIN 4 CUM
JUN 0 0
JUL 0 0
AUG 0 0
SEP 0 0
OCT 0 0 0 0
NOV 0 0 0 0
DEC 0 0 0 0
JAN 143,101 143,101 1,636 1,636
FEB 37,847 180,948 0 1,636
MAR 0 180,948 0 1,636
APR 0 180,948 0 1,636
19
SIP Budget
March - 68 of Plan Forecast 100 of Plan
20
SIP Budget
March - 89 of Plan Forecast 100 of Plan
21
Staffing

• Number of positions filled (1) Juan Carlos
  Chaves, ARL/Adelphi (2) John Nehbass, ASC.
• Number of staff departed none.
• Length of time positions open lt 1 month.
• Administrative assistant essential team member.
• Academic Associate Ashok Krishnamurthy (key to
  continuity).

22
Summary

• SIP effort is coordinated and is having a
  measurable impact. Strong user base, engaged UAP,
  highly effective teamwork.
• Training is key MATLAB, Simulink, Stateflow
  life-cycle tools are being adopted by entire
  community. Need for EXTENSIVE HPC clear and
  growing rapidly.
• HPEC program evolved from SIP Forum.
• Significant management overhead.
• Some current PET processes are problematic
• Funding timing. Solution delayed starts?
• Administrative processes cross-FA activities
  equipment projects. Solution greater
  flexibility?

23
Backup
24
3D Synthetic Aperture RADAR (3DSAR)
10 hours processing time to create one image at
one angle on a 1GHz NT workstation.
5,000 times speedup on HPC resources. Generates
all 360 angles at 1 degree resolution. 41
minutes total time
IMPACT Targets Under Trees (TUT) Chartered to
develop a capability for performing the Find, Fix
and Engage portions of the kill chain process on
targets employing concealment with camouflage and
foliage. Foliage Penetration (FOPEN ) RADAR with
Change Detection, information fusion,
geo-registration, terrain characterization,
weapons fuze and human-system interface
technologies. Change Detection Techniques Targets
will be identified by associating the detections
with fused intelligence information products and
held at risk while in hide with upgraded foliage
penetrating munitions.
25
SIP 007 Standards and Middleware,Optimization
of HIE using VSIPL

• Detection algorithms used on Hyperspectral Image
  Exploitation (HIE) applications require low
  latencies to process an incoming hyperspectral
  image cube.
• Objective for these ATR systems response times
  of military utility (1 minute or less).
• Soln VSIPL is being investigated as a portable
  solution to accomplish the required performance.
• Approach
• VSIPL Organization Liaison participation at
  February VSIPL Tutorial and User's Group Meeting
  
• Profiling, Analysis and optimization of HIE codes
  
• Testing and validation of HIE codes
• Preliminary results upcoming at DoD UGM 2002.
• POC Gary Stolovy (ARL ALC)

26
Porting and Optimization ATR codes at ARL ALC

• ALC ATR system train and test.
• Problem 1) Improvements in design and
  performance of classifier is iterative process
  requiring test and adjustment. 2) Data sets are
  large and performance trends incomplete or
  difficult to adequately characterize due to the
  time required for computation (Around 3 days per
  data set on high end Linux or SGI workstation)
• Solution Port and optimize on the HPC MSRC
  platforms.
• Accomplishments
• Codes ported to Origin SGI HPC.
• Successful profiling, optimization and validation
  of codes
• Currently 11 hours per data set (around 654
  speedup)
• POC Bruce Weber and Joseph Penn (ARL ALC)

27
Profiling of Synthetic Imagery Codes CREATION
code at ARL ALC

• CREATION is a 3-dimensional multi-spectral
  (infrared, visible) computer scene generation
  program that produces high-resolution synthetic
  imagery for signature phenomenology research,
  perception and search studies and ATR
  development, testing and evaluation.
• Problem 1) ARL researchers developed CREATION
  using the IrisGL graphics library and therefore
  it presents serious machine dependences.
  Currently it only runs on Iris SGI high-end
  workstations, as it requires a fully compliant
  IrisGL library. 2) ARL Modeling and Simulation
  Branch researchers are very interested in the
  porting and optimization using OpenGL.
• Soln Profiling and analysis of code.
• Accomplishments
• Initial profiling completed indicated heavy
  dependence on GL SGI features not only on the
  graphical display functionality but also on the
  basic computational engine.
• SIP team conclusions 1) porting of CREATION
  requires a major redesigning of the code
  separating the computational engine from the
  graphical display. 2) Rewriting of computational
  engine requires a porting from GL to OpenGL.
  Graphical display could be re-designed using the
  X Window system. 3) New code could be optimized
  allowing usage of multiple processor HPC
  platforms.
• ARL users looking for major funding as SGI high
  end Visual Workstations where they run CREATION
  are being decommissioned and they require
  generating hundreds of synthetic images for a
  variety of applications.
• POC Bruce Weber and Joseph Penn (ARL ALC)

28
Network Centric VHF Change Detection

• Investigation
• Real-Time IFP HPC Implementation on RC-12
• Lockheed Martin Signal-Data Processor
• Fast Parallel Implementation of Mo-Comp,
  PSAP-type Processing, Digital Spotlighting,
  Model-Based Minimum-Entropy Compensation
  Functions, VHF Change Detection.
• Transmission of Real-Time RC-12 Imagery from
  FOPEN Ground Station to HPC Facility for
  Post-Processing.
• Transmission of VHF Change Detection Result from
  HPC Facility to FOPEN Ground Station.

29
User Contacts and Assistance
30
SIP User Advisory Panel

• Andy Sullivan (ARL/Adelphi, Govt CTA Lead)
• Keith Bromley (SPAWAR)
• Rich Linderman (Rome)
• Mike Bryant (WPAFB)
• Gary Stolovy (ARL/Adelphi)

31
Travel by SIP personnel