Automated method for selecting training areas of sea clutter and detecting ship targets in polarimetric synthetic aperture radar imagery

• US 10,088,555 B2
• Filed: 10/08/2015
• Issued: 10/02/2018
• Est. Priority Date: 12/15/2014
• Status: Active Grant

• Current Assignee: Airbus Singapore Private Limited
• Original Assignee: Airbus Singapore Private Limited
• Inventors: Bretschneider, Timo Rolf, Lee, Ken Yoong
• Primary Examiner(s): Gregory, Bernarr E
• Application Number:
  US15/535,997
• Publication Number:
  US 20180052224A1
• Time in Patent Office: 1,090 Days
• CPC Class Codes: G01S 13/9027 Pattern recognition for feature extraction
• CPC Class Codes: G01S 13/9076 Polarimetric features in SAR
• CPC Class Codes: G01S 7/414 Discriminating targets with respect to background clutter

Abstract

Method for selecting a sea clutter training area in polarimetric synthetic aperture radar input data. A sea clutter reference distribution for a pixel magnitude value is provided. Based on the input data, one or more parameters of the reference distribution and a global covariance matrix are computed. The pixels are grouped into blocks. A block that minimizes a cost function is pre-selected, the cost function being derived from empirical moments of the block and moments of the reference distribution. A goodness-of-fit is computed for the pre-selected block with respect to the reference distribution. If the goodness-of-fit is sufficient, the block is selected as sea clutter training area. Otherwise, the steps of preselecting and computing a goodness of fit are repeated.

Claims

1. A computer implemented method for automatically selecting at least one sea clutter training area in single look polarimetric synthetic aperture radar imagery, the method comprising:

generating, using a polarimetric synthetic aperture radar, input data;

receiving the input data from the polarimetric synthetic aperture radar to a computer, wherein the input data comprises pixels, the pixels comprising amplitude and phase values of electromagnetic signals;

receiving, at the computer, a sea clutter reference distribution for a pixel magnitude;

computing one or more parameters of the sea clutter reference distribution based on the input data;

computing a global covariance matrix based on the input data;

grouping the pixels of the input data into blocks;

computing, for each block, a pixel magnitude using an inverse of the global covariance matrix;

computing, for each block, one or more empirical moments of the pixel magnitude for the pixels of the block;

computing one or more reference moments of the pixel magnitude based on the sea clutter reference distribution;

computing, for each block, a cost function, wherein computing the cost function comprises a computation of respective differences between the empirical moments and the reference moments;

pre-selecting a block with a smallest value of the cost function;

calculating a goodness-of-fit test for the pixel magnitude of the pixels of the pre-selected block compared to the sea clutter reference distribution;

when the pre-selected block passes the goodness-of-fit test, selecting the pre-selected block as a sea clutter training area; and

when the pre-selected block does not pass the goodness-of-fit test, repeating steps of pre-selecting a block and calculating the goodness-of-fit test until one of the blocks is found which passes the goodness-of-fit test.

2. The method of claim 1, wherein the sea clutter reference distribution is a distribution of a magnitude of a first vector valued random variable, the first vector valued random variable representing a pixel and the first vector valued random variable being a product of a square root of a texture variable with a second vector valued variable s, wherein the second vector valued variable s is distributed according to a multivariate Gaussian distribution, and wherein a distribution of the texture variable is selected from a gamma distribution, an inverse gamma distribution, an inverted beta distribution, an inverse Gaussian distribution, and a reciprocal inverse Gaussian distribution.

3. The method of claim 1, wherein the cost function comprises a computation of respective differences between an empirical third moment and a third moment of the sea clutter reference distribution and between an empirical fourth moment and a fourth moment of the sea clutter reference distribution.

4. The method of claim 1, wherein the goodness-of-fit test is a chi-squared test.

5. The method of claim 1, further comprising detecting a target, wherein detecting the target comprises:

estimating a covariance matrix from the pixels of the selected sea clutter training area for every pixel in the input data; and

marking a pixel location as a target pixel when a magnitude value of the pixels corresponding to the pixel location exceeds a pre-determined threshold u, the pre-determined threshold u being determined by a pre-determined probability of false alarm for the sea clutter reference distribution, wherein the magnitude value is the pixel magnitude.

6. The method of claim 5, wherein detecting the target comprises masking out land areas.

7. The method of claim 1, wherein selecting the sea clutter training area comprises masking out land areas.

8. A non-transitory computer readable memory comprising:

a computer instruction set that, when executed by one or more processors of a computation unit in electronic communication with a polarimetric synthetic aperture radar, causes the computation unit to;

receive input data from the polarimetric synthetic aperture radar, wherein the input data comprises pixels, the pixels comprising amplitude and phase values of electromagnetic signals, and wherein the input data is generated by the polarimetric synthetic aperture radar;

receive a sea clutter reference distribution for a pixel magnitude;

compute one or more parameters of the sea clutter reference distribution based on the input data;

compute a global covariance matrix based on the input data;

group the pixels of the input data into blocks;

compute, for each block, a pixel magnitude using an inverse of the global covariance matrix;

compute, for each block, one or more empirical moments of the pixel magnitude for the pixels of the block;

compute one or more reference moments of the pixel magnitude based on the sea clutter reference distribution;

compute, for each block, a cost function, wherein computation of the cost function comprises a computation of respective differences between the empirical moments and the reference moments;

pre-select a block with a smallest value of the cost function;

calculate a goodness-of-fit test for the pixel magnitude of the pixels of the pre-selected block compared to the sea clutter reference distribution;

when the pre-selected block passes the goodness-of-fit test;

select the pre-selected block as a sea clutter training area; and

display at least the sea clutter training area on a display; and

when the block does not pass the goodness-of-fit test, repeating steps of pre-selecting a block and calculating the goodness-of-fit test until one of the blocks is found which passes the goodness-of-fit test.

9. A computation unit, in electronic communication with a polarimetric synthetic aperture radar, comprising:

one or more processors and a non-transitory computer readable memory,

wherein the computation unit is configured to;

receive input data from the polarimetric synthetic aperture radar, wherein the input data comprises pixels, the pixels comprising amplitude and phase values of electromagnetic signals, and wherein the input data is generated by the polarimetric synthetic aperture radar;

receive a sea clutter reference distribution for a pixel magnitude;

compute one or more parameters of the sea clutter reference distribution based on the input data;

compute a global covariance matrix based on the input data;

group the pixels of the input data into blocks;

compute, for each block, a pixel magnitude using an inverse of the global covariance matrix;

compute, for each block, one or more empirical moments of the pixel magnitude for the pixels of the block;

compute one or more reference moments of the pixel magnitude based on the sea clutter reference distribution;

compute, for each block, a cost function, wherein computation of the cost function comprises a computation of respective differences between the empirical moments and the reference moments;

pre-select a block with a smallest value of the cost function;

calculate a goodness-of-fit test for the pixel magnitude of the pixels of the pre-selected block compared to the sea clutter reference distribution;

when the pre-selected block passes the goodness-of-fit test;

select the pre-selected block as a sea clutter training area; and

display at least the sea clutter training area on a display; and

when the block does not pass the goodness-of-fit test, repeating steps of pre-selecting a block and calculating the goodness-of-fit test until one of the blocks is found which passes the goodness-of-fit test.

10. The computation unit of claim 9, wherein the sea clutter reference distribution is a distribution of a magnitude of a first vector valued random variable, the first vector valued random variable representing a pixel and the first vector valued random variable being a product of a square root of a texture variable with a second vector valued variable s, wherein the second vector valued variable s is distributed according to a multivariate Gaussian distribution, and wherein a distribution of the texture variable is selected from a gamma distribution, an inverse gamma distribution, an inverted beta distribution, an inverse Gaussian distribution, and a reciprocal inverse Gaussian distribution.

11. A computer implemented method for automatically selecting at least one sea clutter training area in multi-look polarimetric synthetic aperture radar imagery, the method comprising:

generating, using a polarimetric synthetic aperture radar, input data;

receiving the input data of a digital radar image from the polarimetric synthetic aperture radar to a computer, wherein the input data comprises pixels, the pixels comprising amplitude and phase values of electromagnetic signals;

receiving, at the computer, a sea-clutter reference distribution for a sum magnitude, wherein the sum magnitude is derived from pixels corresponding to a same location and to respective looks of the multi-look polarimetric synthetic aperture radar imagery;

estimating one or more parameters of a sea clutter reference distribution based on the input data;

grouping the pixels into blocks;

computing, for each block, and for each pixel location of the pixels in the block, the sum magnitude;

computing one or more empirical moments of the sum magnitude for the pixels of a block;

computing one or more reference moments of the sea-clutter reference distribution;

computing, for every block of pixels, a cost function, wherein computing the cost function comprises a computation of respective differences between the empirical moments and the reference moments;

pre-selecting a block with a smallest value of the cost function;

calculating a goodness-of-fit test for the sum magnitude of the pixels of the pre-selected block compared to the sea clutter reference distribution;

when the pre-selected block passes the goodness-of-fit test, selecting the pre-selected block as a sea clutter training area; and

when the block does not pass the goodness-of-fit test, repeating steps of pre-selecting a block and calculating the goodness-of-fit test until one of the blocks is found which passes the goodness-of-fit test.

12. The method of claim 11, wherein the sea clutter reference distribution of a sum magnitude is equal to a distribution of a random variable which is a product of a texture variable t and a variable z, wherein the variable z is distributed according to a gamma distribution, and wherein a distribution of the texture variable t is selected from a gamma distribution, an inverse gamma distribution, an inverted beta distribution, an inverse Gaussian distribution, and a reciprocal inverse Gaussian distribution.

13. The method of claim 11, wherein the cost function comprises a computation of respective differences between an empirical third moment and a third moment of the sea clutter reference distribution and between an empirical fourth moment and a fourth moment of the sea clutter reference distribution.

14. The method of claim 11, wherein the goodness-of-fit test is a chi-squared test.

15. The method of claim 11, further comprising detecting a target, wherein detecting the target comprises:

estimating a covariance matrix from the pixels of the selected sea clutter training area for every pixel in the input data; and

marking a pixel location as a target pixel when a magnitude value of the pixels corresponding to the pixel location exceeds a pre-determined threshold u, the pre-determined threshold u being determined by a pre-determined probability of false alarm for the sea clutter reference distribution, wherein the magnitude value is the sum magnitude.

16. The method of claim 15, wherein detecting the target further comprises masking out land areas.

17. The method of claim 11, wherein selecting the sea clutter training area comprises masking out land areas.

18. A non-transitory computer readable memory comprising:

a computer instruction set that, when executed by one or more processors of a computation unit in electronic communication with a polarimetric synthetic aperture radar, causes the computation unit to;

receive input data of a digital radar image from the polarimetric synthetic aperture radar, wherein the input data comprises pixels, the pixels comprising amplitude and phase values of electromagnetic signals, and wherein the input data is generated by the polarimetric synthetic aperture radar;

receive a sea-clutter reference distribution for a sum magnitude, wherein the sum magnitude is derived from pixels corresponding to a same location and to respective looks of multi-look polarimetric synthetic aperture radar imagery;

estimate one or more parameters of a sea clutter reference distribution based on the input data;

group the pixels into blocks;

compute, for each block, and for each pixel location of the pixels in the block, the sum magnitude;

compute one or more empirical moments of the sum magnitude for the pixels of a block;

compute one or more reference moments of the sea-clutter reference distribution;

compute, for every block of pixels, a cost function, wherein computation of the cost function comprises a computation of respective differences between the empirical moments and the reference moments;

pre-select a block with a smallest value of the cost function;

calculate a goodness-of-fit test for the sum magnitude of the pixels of the pre-selected block compared to the sea clutter reference distribution;

when the pre-selected block passes the goodness-of-fit test;

select the pre-selected block as a sea clutter training area; and

displaying at least the sea clutter training area on a display; and

when the block does not pass the goodness-of-fit test, repeating steps of pre-selecting a block and calculating the goodness-of-fit test until one of the blocks is found which passes the goodness-of-fit test.

19. The non-transitory computer readable memory of claim 18, wherein the sea clutter reference distribution of a sum magnitude is equal to a distribution of a random variable which is a product of a texture variable t and a variable z, wherein the variable z is distributed according to a gamma distribution, and wherein a distribution of the texture variable t is selected from a gamma distribution, an inverse gamma distribution, an inverted beta distribution, an inverse Gaussian distribution, and a reciprocal inverse Gaussian distribution.

20. A computation unit, in electronic communication with a polarimetric synthetic aperture radar, comprising:

one or more processors and a non-transitory computer readable memory, wherein the computation unit is configured to;

receive input data of a digital radar image of the polarimetric synthetic aperture radar, the input data comprising pixels, the pixels comprising amplitude and phase values of electromagnetic signals, and wherein the input data is generated by the polarimetric synthetic aperture radar;

receive a sea-clutter reference distribution for a sum magnitude, the sum magnitude being derived from pixels corresponding to a same location and to respective looks of multi-look polarimetric synthetic aperture radar imagery;

estimate one or more parameters of a sea clutter reference distribution based on the input data;

group the pixels into blocks;

compute, for each block, and for each pixel location of the pixels in the block, the sum magnitude;

compute one or more empirical moments of the sum magnitude for the pixels of a block;

compute one or more reference moments of the sea-clutter reference distribution;

compute, for every block of pixels, a cost function, wherein computation of the cost function comprises a computation of respective differences between the empirical moments and the reference moments;

pre-select a block with a smallest value of the cost function;

calculate a goodness-of-fit test for the sum magnitude of the pixels of the pre-selected block compared to the sea clutter reference distribution;

when the block passes the goodness-of-fit test;

select the pre-selected block as a sea clutter training area; and

display at least the sea clutter training area on a display; and

when the block passes the goodness-of-fit test, repeating steps of pre-selecting a block and calculating the goodness-of-fit test until one of the blocks is found which passes the goodness-of-fit test.