Terrain detection and classification using single polarization SAR
The various technologies presented herein relate to identifying manmade and/or natural features in a radar image. Two radar images (e.g., single polarization SAR images) can be captured for a common scene. The first image is captured at a first instance and the second image is captured at a second instance, whereby the duration between the captures are of sufficient time such that temporal decorrelation occurs for natural surfaces in the scene, and only manmade surfaces, e.g., a road, produce correlated pixels. A LCCD image comprising the correlated and decorrelated pixels can be generated from the two radar images. A median image can be generated from a plurality of radar images, whereby any features in the median image can be identified. A superpixel operation can be performed on the LCCD image and the median image, thereby enabling a feature(s) in the LCCD image to be classified.
1. An image processing system comprising:
a single polarization synthetic aperture radar (SAR) imager, the single polarization SAR imager configured to capture a plurality of single polarization SAR images of a scene, the SAR images each comprising pixels; and
a processor programmed with instructions that, when executed by the processor, cause the processor to:
generate a long term coherent change detection (LCCD) image from a first SAR image in the plurality of single polarization SAR images and a second SAR image in the plurality of single polarization SAR images; and
analyze the LCCD image and label a first pixel in the LCCD image as corresponding to a paved roadway.
2. The image processing system of claim 1 , wherein the first SAR image is generated at a first time and the second SAR image is generated at a second time, wherein a time between the first time and the second time is of sufficient temporal duration such that pixels in the first SAR image are correlated with pixels in the second SAR image when a feature is manmade, and the pixels in the first SAR image are decorrelated with pixels in the second SAR image when the feature is natural.
3. The image processing system of claim 1 , wherein the processor utilizes coherent change detection (CCD) to determine a value of the first pixel in the LCCD image.
4. The image processing system of claim 3 , wherein the value of the first pixel in the LCCD image is correlated.
5. The image processing system of claim 1 , wherein the processor is further programmed with instructions that, when executed by the processor, cause the processor to generate a median image from the plurality of SAR images.
6. The image processing system of claim 5 , wherein the processor is further programmed with instructions that, when executed by the processor, cause the processor to determine a plurality of SAR products in the plurality of SAR images.
7. The image processing system of claim 6 , wherein the SAR products include different types of features and classes in the scene, wherein the features and classes include the paved roadway.
8. The image processing system of claim 6 , wherein the processor is further programmed with instructions that, when executed by the processor, cause the processor to segment the median image into a first plurality of superpixels, wherein the first plurality of superpixels includes a first superpixel, and the first plurality of superpixels are segmented in accordance with the plurality of SAR products.
9. The image processing system of claim 8 , wherein the processor is further programmed with instructions that, when executed by the processor, cause the processor to segment the LCCD image into a second plurality of superpixels, wherein the second plurality of superpixels includes a second superpixel, wherein the second superpixel is of the same size and homogeneity as the first superpixel.
10. The image processing system of claim 9 , wherein the processor is further programmed with instructions that, when executed by the processor, cause the processor to compute a first probability of the first superpixel being a terrain class.
11. The image processing system of claim 10 , wherein the processor is further programmed with instructions that, when executed by the processor, cause the processor to compute a second probability of the second superpixel being a terrain class.
12. The image processing system of claim 11 , wherein the processor is further programmed with instructions that, when executed by the processor, cause the processor to fuse the first probability with the second probability to generate an in-class terrain probability, the in-class terrain probability indicating that the first pixel corresponds to the paved roadway.
13. A method, comprising:
forming a LCCD image, wherein the LCCD image being formed from a coherent change detection (CCD) operation between a first plurality of pixels in a first single polarization synthetic aperture radar (SAR) image and a second plurality of pixels in a second single polarization SAR image, the first SAR image and the second SAR image are of a same scene and are included in a plurality of SAR images, wherein the first single polarization SAR image is captured at a first instance of time and the second SAR image is captured at a second instance of time, a duration of time between the first instance of time and the second instance of time is of sufficient temporal duration such that pixels in the first SAR image are correlated with pixels in the second SAR image when the feature is manmade, and the pixels in the first SAR image are decorrelated from the pixels in the second SAR image when the feature is natural;
generating a first superpixel in the LCCD image; and
classifying the first superpixel to be a paved road.
14. The method of claim 13 , further comprising generating a median image from the plurality of SAR images.
15. The method of claim 14 , further comprising:
computing a plurality of SAR products from the plurality of SAR images, wherein the plurality of SAR products include the paved roadway and a dirt track.
16. The method of claim 15 , further comprising:
segmenting the median image into a first plurality of superpixels, wherein the first plurality of superpixels includes a second superpixel; and
segmenting the LCCD image into a second plurality of superpixels, wherein the second plurality of superpixels includes the first superpixel, and the segmenting of the median image and the LCCD image being performed based upon homogeneous regions in the median image and the LCCD image being identified based upon the SAR products.
17. The method of claim 16 , further comprising:
computing a first probability of the first superpixel including terrain corresponding to the paved roadway;
computing a second probability of the second superpixel including the terrain corresponding to the paved road; and
fusing the first probability and the second probability to indicate a final probability of the first superpixel includes the paved road terrain.
18. A computer-readable storage medium comprising instructions that, when executed by a processor, cause the processor to perform acts comprising:
receive a first single polarization synthetic aperture radar (SAR) image, the first single polarization SAR image including a first plurality of pixels;
receive a second single polarization SAR image, the second single polarization SAR image including a second plurality of pixels, wherein the first SAR image and the second SAR image are of the same scene and are included in a plurality of SAR images;
form a LCCD image product, wherein the LCCD image being formed from a coherent change detection (CCD) operation between the first plurality of pixels and the second plurality of pixels, wherein the first SAR image is captured at a first instance of time and the second SAR image is captured at a second instance of time, a duration of time between the first instance of time and the second instance of time is of sufficient temporal duration such that pixels in the first SAR image are correlated with pixels in the second SAR image when the feature is manmade, and the pixels in the first SAR image are decorrelated with pixels in the second SAR image when the feature is natural.
19. The computer-readable storage medium of claim 18 , the acts performed by the processor further comprising:
generating a median image from the plurality of SAR images, the plurality of SAR images depict a plurality of SAR products, the plurality of the SAR products have respective radar cross sections (RCSs) for different types of terrain and features encountered in the scene, wherein the features include a paved road and a dirt track;
segmenting the median image into a first plurality of superpixels, wherein the first plurality of superpixels includes a first superpixel; and
segmenting the LCCD image into a second plurality of superpixels, wherein the second plurality of superpixels includes a second superpixel, and the segmenting of the median image and the LCCD image being performed based upon homogeneous regions in the median image and the LCCD image being identified based upon the SAR products.
20. The computer-readable storage medium of claim 19 , the acts performed by the processor further comprising:
computing a first probability of the first superpixel including terrain corresponding to the paved road;
computing a second probability of the second superpixel including the paved road terrain; and
fusing the first probability and the second probability to indicate a final probability of the first superpixel includes the paved road terrain.