Target Recognition and Identification

1
Target Recognition and Identification

• Using ENVI

Cherie S. Darnel
2
Target Recognition

• ENVI Overview
• Necessary components for Target Recognition
• Target Recognition Methods in ENVI
• Case studies

3
Target Recognition

• ENVI is the ultimate tool to extract information
  from your geospatial data.

4
Target Recognition
Visualize
Customize
5
Target Recognition

• Factors that must be considered when doing
  accurate target detection include

• What kind of data is available for analysis?
• What algorithms, or software packages are
  available to produce accurate results from your
  acquired data?
• Successful target recognition is a blend of these
  two factors

6
Detecting physical features Taliban caves in
Afghanistanusing high resolution commercial
imagery
High Resolution Target Recognition
Unclassified
7
Thermal (IR) Target Recognition

• ENVIs thermal IR capabilities can be enhanced
  with quantitative analysis algorithms to support
  specific IR applications
• Standard thermal processing utilities include
• Radiance Calibrations
• Emissivity Calculations
• Radiant to Kinetic Temperature Conversion
• Bulk Temperature Measurements

8
Trafficability Analysis
Example Land cover and trafficability What is
the capability of the terrain to bear traffic ?
Very dense vegetation, impassible for vehicles
9
RADAR Data Target Recognition
Detecting Vehicles with SAR

• RADAR can see through clouds and be used at
  night.
• Radar can penetrate tree canopies, and camouflage
  netting.
• Return bright signals for man-made structures and
  vehicles using polarization return

Left Airborne SAR Image with camouflaged and
hidden vehicles Right Vehicle map (TARGET)
derived from SAR imagery
TARGETS Identified
10
RADAR Target Recognition

• SAR can provide sufficient resolution to
  distinguish terrain features and to recognize
  man-made targets.
• SAR offers the capability for penetrating
  materials which are optically opaque, and thus
  not visible by optical or IR techniques.
• Target material identification using the unique
  responses of man-made materials against a
  vegetation background

Ordinance
Fuel Tanks
Unidentified Metal Objects
11
Hyperspectral Data Target Recognition

• Hyperspectral data is well suited for detecting
  and identifying concealed and camouflaged
  military hard targets such as vehicles and
  weapons systems.
• Uniquely suited for verifying targets and
  detecting and identifying decoys.
• Increases the accuracy and efficiency of target
  identification, and reduces false alarms.

12
Spectral Data for Target Recognition

• Each material has a unique spectrum
• Using the spectral information in the images you
  can identify target materials, even at the
  sub-pixel level so that very small targets can be
  successfully resolved and identified.

13
ENVI Algorithms for Target Recognition
ATR (Automatic Target Recognition) with ENVIs
state-of-the-art analysis tools Identify targets
at the whole-pixel and sub-pixel level.
14
ENVI Algorithms for Target Recognition
Build a library of known and unknown spectra to
increase the accuracy of target detection
15
Spectral Exploitation in Tactical Time Lines
Unclassified
16
Spectral Exploitation in Tactical Conditions
Using HSI data with ground truth information to
detect tanks concealed under vegetation Targets
were not resolved using high resolution imagery
Unclassified
17
Spectral Exploitation in Tactical Conditions
High Resolution Line Scanner Data
HSI Processed Data
Ground Truth Image
Line Scanner Data
Unclassified
18
Spectral Exploitation in Tactical Conditions
SCUD Missile Launcher Detection
Hyperspectral Data with ENVI algorithms allows
analyst to detect heavily concealed targets
Targets were SCUD TELs concealed under trees and
camouflage
Scud Detections Identified in Yellow
Unclassified
19
Spectral Exploitation in Tactical Conditions

• TACREPS Generated in
• lt 90 Minutes
• 95 Target Detection Rate
• Zero False Alarms
• . Spectral exploitation was the only method that
  provided high detection rates and provided
  identification for concealed targets.

Hi-res data and ENVIs algorithms were capable of
accurately detecting and identifying targets in a
variety of battlefield scenarios.
Unclassified
20
Target Recognition - BandMax

• Developed for target recognition and feature
  extraction with hyperspectral data.
• Automatically identifies spectral bands that
  provide maximum contrast between target and
  background.
• Reduces processing time, increases accuracy, and
  is appropriate for use by non-experts.
• Used specifically for target recognition
• ENVI Wizard guides the user to pick appropriate
  target and background data

21
Target Recognition - BandMax
The BandMax wizard automates all decisions and
produces a target map
Unclassified
22
Target Recognition - BandMax
HSI and MSI data were successfully used with ENVI
during the Militarys Joint Expeditionary Force
Experiment JEFX 2002
Target Detections Identified In Red
Tank Detection for The Hawkeye Kenney Battlelab
Initiative Ft. Polk
Unclassified
23
Target Recognition - BandMax
Field test
Unclassified
24
Spectral Target Detection Camo Netting
Spectral target detection using both in-scene
derived spectra and spectral libraries. Data was
atmospherically corrected to remove effects of
water vapor and other aerosols in scene.
Processed using ENVIs Mixture Tuned Match
Filtering algorithm.
Unclassified
25
Case Study
Ships displacement analysis using Formosat
data DATA A series of panchromatic Formosat
images from Norfolk port one image per day
during 4 days PREPROCESSING All images are
co-registered calibrated then saved in a same
multitemporal cube GOAL Analyze ships
displacement using ENVI spectral tools
(Z-profile, SAM, )
26
Methodology

• Build a temporal cube each band pan image at
  each date.
• Each pixel is characterised by its temporal
  signature.
• Changes are supposed to be only ships
  displacements. All the other changes are not
  considered
• Two cases ship is Absent (pixel value water)
  or Present (pixel value ship)

R May, 9 (Present-P) G May, 8
(Absent-A) B May, 7 (Absent-A)
may, 7 may, 8 may, 9 may, 10
Composite color image Aircraft carrier is red
27
Result temporal SAM classification

• Using the temporal signature of the aircraft
  carriers (Absent, Absent, Present, Present), we
  find all ships which have the same displacements
  (yellow class).
• We can also classify all ships displacements.

Aircraft carrier
Yellow class (A,A, P,P)
Tug boats
28
ENVI Zoom
29
ENVI Zoom
30
ENVI Zoom
31
ENVI Zoom
32
Successful Target Recognition

• Essential components of accurate and successful
  target recognition

• Appropriate, accurate and calibrated data is
  essential.
• Multiple data sources often provide better
  results than a single source.
• Using a proven software solution such as ENVI
  will ensure that you have the best analysis
  techniques and results.

33
Successful Target Recognition

• Essential components of accurate and successful
  target recognition

• Appropriate, accurate and calibrated data is
  essential.
• Multiple data sources often provide better
  results than a single source.
• Using a proven software solution such as ENVI
  will ensure that you have the best analysis
  techniques and results.

www.ITTVIS.com
34
Target Detection and Identification

• A primary goal of using multispectral/hyperspectra
  l/SAR remote sensing data is to determine and
  classify the identities and abundance fractions
  of materials.
• In remote sensing image analysis, the difficulty
  arises in the fact that a scene pixel is
  generally mixed linearly or nonlinearly by
  different material substances resident in the
  pixels.
• RSIs ENVI software includes specialized
  algorithms and techniques that have been
  developed and adapted to assist in targets
  detection and target identification.
• ENVI includes a variety of whole pixel and
  sub-pixel analysis techniques that can detect and
  identify man-made features and equipment.
• Often there is no need to identify a specific
  target the need is only to detect the presence
  of a target that is anomalous to the background.

35
ENVI

• ENVI provides support for virtually any type of
  panchromatic, multispectral, hyperspectral,
  thermal, single band and polarimetric SAR data.
• ENVI includes state-of the art algorithms for
  feature extractions and target recognition.
• ENVI can be easily extended using C, C,
  FORTRAN, JAVA and IDL, so that custom algorithms
  can easily be added.
• ENVIs can be customized to create mission
  specific workflows and batch processing