Marine Surveillance with RADARSAT-2: Ship and Oil Slick Detection

1
Marine Surveillance with RADARSAT-2 Ship and Oil
Slick Detection

• Gordon Staples, Jeff Hurley, Gillian Robert,
  Karen Bannerman
• MDA GSI
• Richmond, BC
• gstaples_at_mda.ca

2
Outline

• Project Objectives
• Ship Detection
• Oil Slick Detection
• Conclusions

3
Introduction

• Canadian Space Agency EOADP-funded project that
  started in June 2007 and finished in March 2011.
  The overall objective was to investigate the use
  of dual-polarized and quad-polarized SAR data for
  maritime surveillance, initially in preparation
  for RADARSAT-2 using ENVISAT and SIR-C data and
  then follow-on with RADARSAT-2 data.
• The project had two focus areas ship detection
  and oil slick detection
• The objectives of the ship detection were to
• Assess the use dual-polarized (and quad pol) as a
  function of incidence angle
• Assess detection and ship orientation
  with-respect-to radar look direction
• The objectives of the oil slick detection study
  were to
• Investigate the use of the polarimetric entropy
  for oil slick characterization
• Understand the application of incoherent target
  decomposition algorithms using RADARSAT-2
  quad-polarized wide-swath modes

4
Ship Detection Study Site

• Study site was the Strait of Georgia, BC
• Access to ship-tracking information from Canada
  Coast Guard shore-based radars
• Ships have predictable routes (e.g. BC Ferries),
  so it was possible to image the same ship, in the
  same orientation, but using a different incidence
  angle
• Variable wind speeds, but nothing too extreme
• Wind speeds were typically less than 10 m/s

Strait of Georgia study site showing ship Routes
for ferry traffic between the BC Mainland and
Vancouver Island
5
Ship Validation Data

• Vessel tracked from the Maritime Communications
  and Traffic Services Centre operated by the
  Canadian Coast Guard (CCG)
• The CCG data provided
• continuous tracking
• ship name, Lloyds registry, type, length, speed,
  direction, etc.
• The time difference between the CCG ship data and
  the RADARSAT-2 acquisition was typically less
  than a few minutes, so positive identification
  was possible

6
Dual-Polarized Data

• The data were acquired in ascending mode since
  previous work with ENVISAT data indicated that in
  general there was more ship traffic at 6 PM
  versus 6 AM.

7
HH

• Wide 1 image (HHHV) acquired August 4 (left) and
  wide speed derived from the HH image (top). Note
  the high wind speeds that appear as bright
  returns in the HH image, but less pronounced in
  the HV image.

HV
8
Wide 1 Wide 2 Results (HH HV) 19.1? - 39.4?
HV TCR constant with incidence angle HH TCR
increases with increasing incidence angle

• TCR near range HV gt HH
• TCR far range HH gt HV
• At 30? incidence angle, HV and HH TCR are
  similar
• Results based on the BC Ferries with lengths
  between 80 m and 160 m, with the same ships in
  Wide 1 and Wide 2

9
Wide 1 Wide 3 Results (VVVH) 19.1? - 31.1?
and 38.7? - 45.3?

• Seven ships between 26 m and 133 m were detected
  in Wide 1 and Wide 3
• TCR near range HV gt HH
• TCR far range HH gt HV

10
Ship Orientation and TCR

• CCG data provides accurate heading/course
  information on each vessel
• All of the vessels /- 45 from parallel were put
  in the parallel category, while the remaining
  were classed as orthogonal
• Ships greater gt 50 m in length were selected and
  the TCR estimated for ships that were parallel
  and perpendicular to the radar look-direction

As the ship length decreases, it was difficult to
discern orientation (depends on radar resolution
30 m for this example)
11
As ship length increased, there was a trend for
the TCR (VV or VH) to be larger when the ship was
oriented perpendicular to the radar look
direction vs. parallel
12
Study SiteCantarell Oil Seep

• Cantarell oil seep (left) and a subscene of a
  RADARSAT-2 ScanSAR Narrow image (right) acquired
  March 5, 2011 showing the Cantarell oil seep.
  The inset shows an overlay of the FQW15 acquired
  March 1, 2011. RADARSAT-2 quad-polarized Fine and
  Standard Wide swath data (50 km az x 25 km rg)
  were acquired.

13
Differences in the Cloude-Pottier entropy were
observed between oil types Intermediate Fuel
Oil (IFO) and oil from the Cantarell seep
SIR-C
RADARSAT-2
The entropy increased with oil viscosity (IFO gt
Cantarell), but was this increase related to oil
properties or incidence angle?
14

• RADARSAT-2 data were acquired at larger incidence
  angles
• Sigma-0 divergence for oil-ocean with increasing
  incidence angle for co-polarized, but invariant
  for cross-polarized data

• Entropy divergence correlates with co-polarized
  divergence
• Entropy increased with incidence angle, but
  entropy is derived from the coherency matrix
  which is derived from the scattering matrix, so
  the relationship makes sense ? oil-type
  dependency is suspect

15
Entropy

• Entropy for FQW2 (top) and FQW15 (bottom)
• The incidence angle range for FQW2 is 19.1? -
  22.7? and 34.4? 36.0? for FQW15
• The scale on the right is from 0 to 1, with low
  entropy (blue) and high entropy (red)

16
Noise Floor and Target Decomposition

• Cross-polarized return for oil and ocean and the
  FQW15 and SQW15 noise floor
• The cross-pol (HV) for oil and ocean is at or
  below the noise floor
• For low return targets, the use the
  cross-polarized data may be noise-limited and
  impact results for target decomposition

17
Eigenvalues

• Eigenvalues ?1 and ?2 (top) and ?3 (bottom) for
  oil and ocean calculated from the 3x3 coherency
  matrix
• Ocean scattering
• ?1 invariant with incidence angle
• Oil scattering
• ?1 dominates to about 25? incidence angle, and
  then ?2 increases (?1 gt ?2)
• For both oil and ocean, ?3 ltlt (?1 and ?2)
• ?3 ? 0 (HV polarization)

18
Target Decomposition3x3 vs. 2x2 Coherency Matrix

• Incoherent target decomposition (e.g.
  Cloude-Pottier, Touzi) use co-polarized and
  cross-polarized data derived from (usually) the
  3x3 coherency matrix
• The dominance of the first and second eigenvalue
  (at smaller incidence angles) suggests that the
  cross-polarized terms in the 3x3 coherency matrix
  can be neglected
• To assess the impact of neglecting the
  cross-polarized terms, the 2x2 symmetric
  coherency matrix was formed by setting the
  cross-polarized terms, SHV 0

19
Dominant eigenvalue (left) and Touzi scattering
type phase (right) derived from 3x3
coherency matrix (top), 2x2 coherency matrix
(middle), and the difference (bottom). The
differences are mainly in the offshore platforms.
FQW2 data
20
Application of the Touzi Decomposition

• Dominant eigenvalue (?1)
• Limited oil-ocean discrimination at small
  incidence angles (FQW2), but better at larger
  incidence angle (FQW15)
• Scattering type phase (??S)
• good discrimination between oil and ocean for
  both images, thus suggesting incidence-angle
  invariance.

21
Summary

• Ship Detection
• TCR is larger for
• small incidence angles for HV
• large incidence angles for HH or VV
• TCR (orthogonal) gt TCR (parallel) as ship length
  increases from 50 m to 175 m
• The use of co/cros-pol data provides good ship
  detection across a large range of incidence
  angles
• Oil Slick Detection
• The entropy increased with incidence angle, so
  oil-type discrimination requires validation with
  different oil types
• Scattering dominated by ?1 and ?2 suggesting
  that ?3 can be neglected
• Work is in progress to further understand the use
  of incoherent target decomposition for oil slick
  discrimination (interslick variability and
  oil-type differences)