34
Ship Detection From Processing to Instrument Characterisation …and back ! Guillaume HAJDUCH BOOST Technologies [email protected] 115, rue Claude Chappe, 29280 Plouzané

Ship Detection

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

Ship DetectionFrom Processing to Instrument

Characterisation

…and back !

Guillaume HAJDUCH BOOST [email protected], rue Claude Chappe, 29280 Plouzané

Context

SHIPDETECTION

Kerguelen

Context

• SARTool© is a modularsoftware including:– Oil detection– Wind inversion– Wave inversion

– Ship detection

SARTool

Context• Definition of new operational services requires to

assess/monitor the performances of the full ship detection processing chain wrt:

– SAR sensor/processor characteristics

– Local meteo/oceanic conditions

– Characteristics of ships to be detected

Product Cal/Valactivities

Ship detection reports vs meteo-oceanicconditions

Ship detection reports vs ground truth (AIS…)

• Context• Overview of Ship Detection Strategy• Results of Performance Assessment• Conclusion & Perspectives

Overview of Ship Detection Strategy

Preprocessing (Land) Masking Pre-screening Discrimination

SAR

Product

ShipDetection

Report

Calibration Focus on area of interest

First detection of

targetscandidates

(TFAC)

Rejection of False

Alarms

Overview of Ship Detection Strategy• Prescreening (CFAR

Approach)– Detection of bright points

within the SAR image– Requires a statistical

characterization of nominal intensity of sea clutter

• Hypothesis on probability law for homogeneous sea clutter intensity

– Apply an adaptive threshold with respect to a predefined False Alarm Rate

• Within the validity of local approximation of sea clutter characteristics (local discontinuities, instrument artifacts…)

False AlarmRate

To be checked and monitoredwrt SAR sensor, mode, processor, …

Overview of Ship Detection Strategy• Discrimination of False Alarms

– Related to instrument artefacts

Nadir ambiguities (RS1)

Nadir Observedarea

Ambiguousdistance

Adressed by geometric considerations

PRFnchtNadir //2 +=

Overview of Ship Detection Strategy• Discrimination of False Alarms

– Related to instrument artefacts

Nadir ambiguities (RS1)

For ScanSAR modes like

• ENVISAT ASAR Wide Swath

• Radarsat ScanSAR Narrow/Wide

The PRF is changing from one subswath to another

• must take into account PRF variation

• must estimate location of subswathborders and overlapping

PRFnchtNadir //2 +=

Overview of Ship Detection Strategy• Discrimination of False Alarms

– Related to instrument artefacts

Target

Ambiguities

Azimuth ambiguities

Observedarea

Ambiguousarea

Ambiguousdistance

Observedarea

Ambiguousarea

Overview of Ship Detection Strategy• Discrimination of False Alarms

– Related to instrument artefactsRD vV

=

Target

Ambiguities

Azimuth ambiguities

Adressed by geometric/radiometricconsiderations

.2ambv k PRFλ=

2ambR kD PRFV

λ=

Detection of main echo by radiometric considerations.

Main echo must be a candidate ship

Overview of Ship Detection Strategy• Discrimination of False Alarms

– Related to instrument artefactsInterferences ?

Overview of Ship Detection Strategy• Discrimination of False Alarms

– Related to erroneous hypothesis on sea clutterNot homogeneous

Inhomogeneous sea clutter or badprobability density law

Can be addressed by a trainedclassifier using a basic featurevector including radiometricmeasurements on targetcandidate

Adressed by radiometric considerations

Overview of Ship Detection Strategy• Discrimination of False Alarms

– Short life time events

Either:

• spike

• azimuth ambiguities

Azimuth ambiguities off Cherbourg harbour

Overview of Ship Detection Strategy• Discrimination of False Alarms

– Short life time events

[R67] [R67]

• Discrimination by analysis of sublooks of SLC products

• Short life time events vanishfrom one look to another, but are stacked in M-Look image

• Suppressed by studying cross-correlation of sublooks

• Allows to reject azimuthambiguities near shore whennative reflector is not a targetcandidate

LOOK 1 LOOK 2

LOOK 3 M-LOOKS

• Context• Overview of Ship Detection Strategy• Results of Performance Assessment• Conclusion & Perspectives

Performance assessment I• Analysis of detection reports

SAR products

SARTool

Shipdetectionreports

Statisticalanalysis

Performance assessment I• Histogram of target candidates wrt incidence angle

Incidence Angle

Num

bero

f can

dida

tes

•Performed on ASA/WSM products

•From 16 to 45 degrees

High occurrence of missing data atproduct border

Performance assessment I• 2D Histogram of targets wrt incidence angle and RCS

Incidence Angle

Rad

ar C

ross

Sec

tion

PRIOR TO DISCRIMINATION

Sensor NES0

Mean SeaBackscattering

NES0 from instrument calibration

Performance assessment I

• Effectiveness of false alarm discrimination process

• Requires accuratelyradiometricallycalibrated products

Incidence Angle

Rad

ar C

ross

Sec

tion

AFTER DISCRIMINATION

Adressed by radiometric considerations

No data processedin this area

Performance assessment I• Amendment of TSS off Portugal

Performance assessment II• Detection versus meteo/oceanic conditions

SAR products

SARTool

Shipdetectionreports

Statisticalanalysis

Database of meteo/oceanic

conditions

Performance assessment II• Histogram of target candidates wrt wind speed

Wind speed

Num

bero

f can

dida

tes

Computed localyfrom SAR product

using SARTool Wind

Local increase of computed windspeeddue to target sidelobes

SARTool

Database of meteo/oceanic conditions

Performance assessment III• Detection versus ground truth

SAR products

Database of meteo/oceanic

conditionsSARTool

Shipdetectionreports

Statisticalanalysis

AIS reports

Performance assessment III• Collecting information through a collaborative network of

AIS receiver from 2007/10 to now– After rough suppression of erroneous reports

• Information on 2500 ships where collected• Information on 700000 positions where collected (not all collocated with SAR

products)

Dev Interface to SARTool Interface to Google

Performance assessment III• Automatic analysis of AIS reports is under development

– Analysis of ship parameters– Co-localisation AIS / SAR– AIS/SAR Pairing

• Objectives– Variability of RCS– … wrt to ship length– … wrt to incidence angle– … wrt to wind speed (perf.)

Length

Wid

th

• Context• Overview of Ship Detection Strategy• Results of Performance Assessment• Conclusion & Perspectives

Conclusion• From performance assessment to instrument

characterization:– Illustration of impact of NES0 of the sensor in terms of

ship detection• And Back !

– NES0 of the sensor must be taken into account in order to reject a high number of false alarms

– False alarm rejection requires radiometricallycalibrated SAR products & the ability to invert wind information locally

Perspectives• Estimation of ship detection performances by using AIS

data base– To be continued

• Fit of ship RCS from length– Check / enhanced thumb rule

• Characterization of instrumental artefacts for new sensors / missions– TerraSAR X, CSK, ALOS

• Characterization of processor artefacts– …

QUESTIONS ?SARTool