Superpixel Edge Detection in Speckle-Reduced SAR Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Synthetic aperture radar (SAR) images often contain speckle, which complicates the application of superpixel segmentation algorithms and hinders the accurate identification of object boundaries, as these algorithms require high signal-to-noise-ratio images with low artifacts.
Innovation Solution
The implementation of a method that reduces speckle in SAR images through processes like sub-aperture multilook, mean coherent change detection, and median radar cross section, followed by superpixel segmentation, allows for the identification of internal and external edges by comparing the properties of adjacent superpixels using a ratio-contrast measure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If superpixel segmentation algorithms are applied to SAR images, then object boundary identification is improved, but the presence of speckle complicates the segmentation and reduces accuracy
Solution Approach 1:
The patent applies speckle reduction techniques (such as filtering or transformation methods) to convert the harmful speckle noise into beneficial information. By processing the SAR image through speckle reduction algorithms, the harmful interference is transformed into a cleaned-up image that preserves essential boundary information while eliminating noise that would otherwise complicate superpixel segmentation and reduce boundary detection accuracy.
2Measurement precision
If speckle reduction processes are applied to SAR images, then superpixel segmentation accuracy is improved, but image processing complexity increases
Solution Approach 1:
The patent performs speckle reduction as a preliminary step before applying superpixel segmentation algorithms. By pre-processing the SAR image to reduce speckle interference, the subsequent segmentation process operates on a cleaned image, which simplifies the overall processing flow and improves segmentation accuracy without requiring complex iterative optimization during the segmentation stage itself.
3Measurement precision
If multiple speckle reduction processes are combined, then edge detection accuracy is improved, but processing time increases
Solution Approach 1:
The patent evaluates and selects the most effective speckle reduction techniques rather than blindly applying all possible methods. By using partial action (applying only the necessary reduction steps) or excessive action (applying strong reduction methods when needed), the system achieves high edge detection accuracy while minimizing unnecessary processing time. The methodology involves assessing image characteristics to determine the appropriate level of speckle reduction intensity.
Data Source
AI summary
Various embodiments presented herein relate to identifying one or more edges in a synthetic aperture radar (SAR) image comprising a plurality of superpixels. Superpixels sharing an edge (or boundary) can be identified and one or more properties of the shared superpixels can be compared to determine whether the superpixels form the same or two different features. Where the superpixels form the same feature the edge is identified as an internal edge. Where the superpixels form two different features, the edge is identified as an external edge. Based upon classification of the superpixels, the external edge can be further determined to form part of a roof, wall, etc. The superpixels can be formed from a speckle-reduced SAR image product formed from a registered stack of SAR images, which is further segmented into a plurality of superpixels. The edge identification process is applied to the SAR image comprising the superpixels and edges.