Sea objects detection with COSMO/SkyMed

Transcription

1 Sea objects detection with COSMO/SkyMed Università di Firenze, CNR-IREA, Università di Napoli Federico II, Politecnico di Bari, Università del Piemonte Orientale FINAL WORKSHOP ACHIEVED RESULTS IN RESPONSE TO THE FIRST COSMO-SkyMed ANNOUNCEMENT OF OPPORTUNITY Rome, March 27 th, 28 th and 29 th

2 Aims of the project 1. Investigation about the problem of detection of lost objects and/or small crafts (for instance lost containers). 2. Investigation about the exploitation of the different surface coherence time in oil spill detection. Research units Unit Institution Coordinator 1 Università di Firenze DET Angelo Freni (PI) 2 CNR-IREA Napoli Gianfranco Fornaro 3 Università di Napoli Federico II - DIBET Antonio Iodice 4 Politecnico di Bari Maria Teresa Chiaradia 5 Università del Piemonte Orientale Paolo Trivero

3 Activities: basic subdivision - Development of the electromagnetic model and raw signal simulator; - Development of a SAR processor capable to focalize data in subapertures; - Development of an algorithm to detect small and lost objects on the sea surface; - Development of an algorithm to estimate the surface coherence time;

4 J Iterative PO (UNIFI) n ( p 1) J s ˆs S 0, 0 lit ( p 1) ( p) s ( r ) 2nˆ 2nˆ s s H i PV J s ( r ) G ( r, r) J S m S shadow ( p1) s ( r) ds Computational efficiency obtained by means of use of MLFMA: O(Nlog 2 N) z x step 2 step 1 step 0 Level 1 Level 2 Level 3

5 SAR raw signal simulation (UNINA) s x, r s x, r s x, r sea boat x, rg x x, r r r s sea ( x, r ) dxdr sea m ; sea ( x, r) ( x, r, x' 0) sea g m Impulse response modified according to the sea dispersion relation x, rgx x, r r r s boat ( x, r ) dxdr boat ; boat g ( x, r) Impulse response modified according to the (constant) boat velocity

6 Sea objects detection with COSMO/SkyMed (ID: 1180) wind: 3 m/s Simulated scenes wind: 12 m/s

7 Investigation on surface coherence time for oil slick detection (UNIFI) Usual oil slick detection algorithms expolit dumping of capillary waves. S S 0 slick water 2c w 2c s g g w s w s S s /S w [db] Orbital velocity 2 u r u S( ) d k [m - 1] Coherence time acqua slick 1 Coherence time s Coherence time ratio 0 u 2 r OLA slick SAR resolution: 10 (az) x 7 (rg) m s [sec] s (slick) / s (acqua) u10 [m/s] u10 [m/s]

8 Subaperture processing (IREA) raw data range compression sub/full aperture azimuth focusing focused image Doppler parameter estimation FREQUENCY RENSPONSE 16/16: black 8/16 yellow 1/16 blue TIME RENSPONSE RENSPONSE 16/16: black 8/16 yellow 1/16 blue

9 Subaperture processing (IREA): Mexican Gulf dataset Half aperture focused image resampling Statistical analysis Out

10 Ships and Lost Objects Detection (UNIPO, POLIBA) Coordinated by prof. P. Trivero M. Cavagnero, M. Borasi, W. Biamino D. Loreggia Group directed by prof. M. T. Chiaradia

11 Cosmo Sky-Med Images selected on the E-Geos catalogue Acquisition mode: ENHANCED SPOTLIGHT 2 Polarization: HH e VV Pixel resolution: 1m Pixel spacing: 0,5 m Acquisition mode: HIMAGE Polarization: HH e VV Pixel resolution: 5 m Pixel spacing: 2,5 m Acquisition mode: PINGPONG Polarization: CO (HH e VV) e CH (HV e HH) Pixel resolution: 20 m Pixel spacing: 10 m n. 2 drifting boats (Tuscany) n. 1 Tuscan coast n. 4 Island of Giglio n. 2 Island of Capraia n. 3 Island of Elba n. 1 floating objects( Tuscany) n. 1 Tuscan coast n. 10 New Zealand n. 7 Island of Capraia n. 1 New Zealand Acquisition mode: WIDEREGION Polarization: HH e VV Pixel resolution: 30 m Pixel spacing: 15 m n. 1 New Zealand Total: 33 images

12 Some algorithms currently available in literature Global threshold fixes a threshold for the intensity and declares each pixel with the intensity greater than the threshold as a potential target; Adaptive threshold calculates the intensity value of each pixel and compares it with neighbourhoods. In this case too, a threshold to discriminate potential target is requires, usually obtained from image background statistics (CFAR - Constant False Alarm Ratio, Cell Average) Colour and Texture Classification compares the intensity of each pixel with neighbour ones lying along the same row and column and combines it with the histogram of each colour when available; Coherence evaluation of image estimates the correlation of the image using a predefined kernel or another part of the same image; Genetic programming compares a reference image with known object population with a template containing all possible objects of interest: calculates the detection ratio and the false alarm ratio and use these gauges to characterize test images; Small object detection in cluttered image using a correlation based active contour model, A. Vard, K. Jamshidi, N. Movahhedinia, Pattern Recognition Letters 33, , 2012; Detection of ship targets in COSMO-SkyMed SAR images, Pastina, D. Fico, F. Lombardo, P., Radar Conference (RADAR), 2011 IEEE; The SAR Ocean Image Correlation Model and Its Validation by MultiBand SAR Ocean Images, X. Wang, Y.Chen, M. Zhu, Y. You, and T. Hu, Progress in Electromagnetics Research Symposium Proceedings, Moscow, Russia, August 2009; A scheme for ship detection in inhomogeneous regions based on segmentation of SAR images, F. Zhang and B. Wu, International Journal of Remote Sensing, Vol. 29, No. 19, , October 2008; Ship Detection in Synthetic Aperture Radar Imagery, Paris W. Vachon, Proceedings OceanSAR 2006 Third Workshop on Coastal and Marine Applications of SAR, St. John s, NL, Canada, October 2006; The State-of-Art in Ship Detection in Synthetic Aperture Radar Imagery, D.J. Crisp, Australian Government, Department of Defense, May 2004; An Automatic Approach to Ship Detection in Spaceborne Synthetic Aperture Radar Imagery: an Assessment of ship detection capability using RADARSAT, Farid Askari, Benoit Zerrr, Saclant Undersea Research Centre, December 2000.

13 Development of the algorithm for ships and lost objects detection (1/2) The procedure for ship and lost objects detection is shown in the following diagram: Study case (input image) Calibration Land masking Initial screening : identification of areas with high brightness (candidate sea object) by Global Threshold Compensation Detection (image processing) Searching for the identified object on successive images Application of the Adaptive Threshold method on the areas previously identified Feature extraction: lost object and ship (tonnage, course e velocity)

14 Development of the algorithm for ships and lost objects detection (2/2) Adaptive Threshold on areas of interest: the approach for searching lost objects is by iteration of the 2D-CA (2 Dimensional Cell Average) method: If necessary, in high noise regime, after a preliminary screening, we apply a correlation filter that scroll over all the image in order to find and to amplify weak signals (A. Vard et al. 2011). 2D-CA: the intensity in the target pixel is compared with the mean intensity in the background window This new image is screened again with 2D-CA method Example of a typical correlation kernel for weak signal enhancement The full procedure could be replicate with different image binning, from FULL FRAME up to a TBC binning value (e.g. 20) to check the persistence of the signal

15 Ship detection: estimation of tonnage, course and speed The ship s speed, in the direction range, is deduced from the shift d between the ship and its wake in the azimuth direction V nave,range V h satellite satellite d tan The speed along the azimuth direction is calculated as: V V nave, azimuth cos where is the direction angle of the ship. The ship s route expressed in degrees in an anticlockwise direction from the geographic North, is calculated as : Rnave r where r is the direction of the ship s wake and is the rotation angle of the image with respect to the North. The ship s tonnage is deduced as a function of the ship dimensions.

16 Validation: warning 1 from the Harbour Master's office floating objects A) WARNING TO MARINERS OF FLOATING/ HALF- FLOATING OBJECTS ADRIFT POSITION φ N - λ E DATE ADDITIONAL INFORMATION ABOUT THE POINT: _(A, B o C)_ WEATHER IN THE AREA (direction and intensity of the wind and sea condition); wind E 2 Moderate Sea Overcast Sky Discrete Visibility - DIRECTION AND INTENSITY OF THE SEA CURRENT N NW2 - HEIGHT OF THE WAVES 1.5 mt HOUR 09.00/A Validation: warning 2 from the Harbour Master's office Boat adrift HMO PORTOFERRAIO DISCRIMINATION OF COSMOSKYMED IMAGES OF 10/01/ LOCAL TIME There are no Cosmo Sky-Med images of the area available in the hours immediately after the warning SHIP FLAG CALL SIGN LENGTH (MT) POSITION Object adrift sighting: SMALL BOAT OF ABOUT 5 MT, WHITE, WITH FLAT BOTTOM, WHITOUT AN ENGINE AND WITH NO PEOPLE ON BOARD Position: N E Sighting time: A Weather in the area: WIND N/W 3 MODERATE SEA. Warning received from M/N STRADA CORSA Flag : United Kingdom of Great Britain and Northern Ireland Call sign : GZOL Length : 150 MT Position: N E Enhanced Spotlight 2 image acquired 3 days and 7hs after the sighting Detected a white spot 3.6 km North from the sighting point; width of the spot 2,5 m

17 Validation: case study New Zealand As the satellite sees an object (container ISO 220 x 600 cm) pixel resolution 5 m pixel spacing 2,5 m pixel resolution 5 m pixel spacing 2,5 m On 5th October 2011 the merchant ship Rena Liberianflagged ran aground in the cliff called Astrolabe, 22km offshore of Tauranga, the largest city in the Bay of Plenty, in the North Island of New Zealand. 14 image pixels containing the signal of the object 12 image pixels containing the signal of the object

18 Quicklook Full resolution 2,5 m pixel spacing CSK HI VV 13 october :09 CSK HI VV 14 october :09 CSK HI VV 15 october :09 Quicklook Full resolution 2,5 m pixel spacing Quicklook Full resolution 2,5 m pixel spacing CSK HI VV 16 october :15 CSK HI VV 16 october :03

19 Experiments: measurements in the sea at the Livorno Harbour It was realized a floating aluminium coated platform dimensioned as an ISO container 220 x 600 cm. The platform will be anchored from the beginning of April at least an hundred meter from the shore in a sea area not sheltered by jetties or sea cliffs so that it can freely float. If possible, it will be left on-site for long enough to gather a considerable number of satellite acquisitions with different weather conditions. Otherwise it will be scheduled to go out in sea to coincide with the satellite transits. After the first observations, if positive, the platform will be ballasted, with appropriate weights, so that it will be semi-submerged and we will gather some new satellite acquisitions. Leaf aluminum 0,1 cm marine plywood structure of fir wooden sect.10x10 hook anchor ~15 cm 220 cm 600 cm

20 Conclusions - Development of the electromagnetic model and the raw signal simulator: A fast high-order PO model has been developed and included in SAR raw signal simulator for marine scenes. - Development of a SAR processor capable to focalize data in subapertures: A SAR processor capable of subaperture focalization has been developed. - Development of an algorithm to measure the surface coherence time: The difference in terms of surface coherence time in case of a spill and clean water has been estimated and we are completing an estimation algorithm. - Development of an algorithm to detect small and lost objects on the sea surface: The problem of detection of small objects on sea surface has been addressed and an appropriate algorithm has been developed.