TL;DR
This paper introduces a novel geometric self-calibration method for YaoGan-13 satellite images that eliminates the need for ground control points or high-precision reference images. By leveraging the spatial intersection residuals of conjugate points across multiple overlapping images, the method accurately corrects systematic errors, achieving high absolute positioning accuracy comparable to traditional methods but with significantly reduced operational overhead.
What Problem Does This Paper Address?
Accurate geolocation is crucial for the effective use of satellite imagery, especially for applications like resource monitoring. However, synthetic aperture radar (SAR) images, including those from the Chinese YaoGan-13 satellite, often suffer from poor absolute positioning accuracy due to systematic errors like timing offsets (azimuth and range delays) within the SAR system. Traditional geometric calibration methods rely on ground control points (GCPs) or calibration fields, which are labor-intensive, time-consuming, and can be expensive to establish and maintain. While some newer methods reduce reliance on GCPs, they still often depend on high-precision reference images, which may not always be available or up-to-date. This paper addresses the need for a fast, accurate, and independent geometric calibration method for SAR imagery that bypasses the limitations of existing approaches.
Methods
The proposed method is a geometric self-calibration technique that relies on the consistency of conjugate image points across multiple overlapping SAR images. The core idea is to exploit the spatial intersection residual that arises when geometric positioning errors are present.
- Geometric Model: The method models the SAR imaging process using slant range and Doppler equations, incorporating potential errors like slant range correction ($R_s$) and atmospheric path delay ($R_{atmo}$).
- Error Equations: Error equations are derived from the geometric model to represent the relationship between measurement errors (in slant range and Doppler frequency) and unknown parameters (corrections to sensor position $\Delta X, \Delta Y, \Delta Z$ and slant range error $\Delta R_s$).
- Spatial Intersection Residual: When multiple images of the same ground area are acquired from slightly different viewpoints, any errors in the satellite's positioning parameters will cause the calculated intersection points of conjugate rays to deviate. The proposed method uses the spatial intersection residual between conjugate points in at least three overlapping images as a constraint. This residual is minimized to solve for the geometric errors.
- Iterative Solution: The method employs an iterative least-squares adjustment approach. It involves:
- Selecting conjugate points across overlapping images.
- Obtaining satellite positioning and imaging parameters.
- Estimating initial values for unknown parameters (typically setting initial errors to zero).
- Calculating approximate slant range and Doppler frequencies.
- Formulating and solving a system of normal equations to estimate corrections ($\Delta X, \Delta Y, \Delta Z, \Delta R_s$).
- Updating the ground coordinates and slant range correction.
- Repeating the process until convergence criteria (e.g., small error limits for slant range correction and systematic azimuth shifts) are met.
- Systematic Azimuth Shifts: The method also accounts for systematic azimuth shifts, which are corrected in a subsequent iterative step.
The novelty lies in its independence from external GCPs and reference images, using only the inherent geometric relationships within overlapping SAR data.
Datasets
The experimental data used for validation consisted of YaoGan-13 Stripmap images acquired between December 18, 2015, and March 30, 2016. These images covered several regions in China, including Songshan, Taiyuan, Anping, Xianning, and Tianjin.
The data was divided into two groups based on bandwidth and pulse width:
- Calibration Group A: 200 MHz bandwidth, 24.4 µs pulse width.
- Calibration Group B: 150 MHz bandwidth, 24.4 µs pulse width.
These groups were further processed into various combinations of images (e.g., A1-B1-C1, A1-B1-C1-D1, etc.) to test the method's performance with increasing redundancy.
For validation of the self-calibration accuracy, independent check points (ICPs) were manually extracted from control data. The sources of control data included:
- Six Corner reflectors.
- Global Positioning System (GPS) control points.
- Digital Orthophoto Map (DOM) / Digital Elevation Model (DEM).
Key Insights and Ideas
- Geometric Consistency as a Constraint: The fundamental insight is that the geometric consistency of conjugate points across multiple overlapping views provides sufficient information to correct for systematic positioning errors, even without external ground truth.
- Self-Sufficiency: The method demonstrates that SAR systems can "self-calibrate" by analyzing their own internal geometric relationships, reducing reliance on costly and time-consuming external calibration procedures.
- Handling Multiple Error Sources: The approach effectively addresses both range delays (internal electronic delay) and azimuth shifts, which are common systematic errors in SAR systems.
- Robustness with Redundancy: The results show that the calibration parameters remain stable and consistent even as more overlapping images are added to the calibration set, indicating the robustness of the method.
- Practical Implications: This self-calibration technique offers a pathway to rapidly improve the geometric accuracy of SAR data, making it more readily usable for applications that require precise geolocation.
Key Results
- Improved Absolute Positioning Accuracy: Without calibration, the absolute positioning accuracy of YaoGan-13 Stripmap images was poor (e.g., 29.47 m plane error for Group A, 22.76 m for Group B).
- High Accuracy Achieved: After self-calibration, the plane absolute positioning accuracy was significantly improved, achieving better than 3.83 m for Validation Group A and 3.41 m for Validation Group B.
- Comparison to Conventional Methods: The accuracy achieved by the proposed self-calibration method was very close to that obtained by conventional field calibration using corner reflectors. The difference in plane positioning accuracy was approximately 2.53 m for Group A and 1.74 m for Group B, both smaller than the 3-m image resolution.
- Stable Calibration Parameters: The differences between the maximum and minimum values of slant range correction (1.83 m) and systematic azimuth shifts (0.153 ms) across different image combinations demonstrated the stability of the calibration results.
Limitations and Future Work
- Limitations:
- The method requires at least three overlapping images. The quality and distribution of these overlaps are crucial for accurate calibration.
- The accuracy of the initial conjugate point measurements directly impacts the final calibration result.
- While the method is independent of GCPs, it still relies on the availability of overlapping imagery.
- Future Work:
- The paper suggests that further research could explore applying this method to other SAR satellites and different imaging modes.
- Investigating the impact of different types and densities of overlapping imagery on calibration accuracy would be beneficial.
- Extending the method to incorporate more complex error models or to address other types of geometric distortions could enhance its applicability.
Summary updated