TL;DR
This paper introduces a novel method for generating highly accurate orthorectified SAR images from China's GF-3 satellite without relying on Ground Control Points (GCPs). By first performing geometric calibration and then employing a DEM-assisted planar block adjustment, the method achieves sub-pixel accuracy, significantly reducing the cost and effort associated with traditional GCP-dependent approaches.
What Problem Does This Paper Address?
Accurate geometric positioning and mapping are crucial for the effective use of spaceborne Synthetic Aperture Radar (SAR) imagery. While many SAR systems boast high precision, achieving reliable absolute positioning for mapping often necessitates the use of Ground Control Points (GCPs). However, acquiring GCPs can be challenging, time-consuming, and expensive, especially in remote or inaccessible areas. This paper addresses the critical need for a GCP-free method to achieve high-accuracy geometric processing and orthorectification for GF-3 SAR imagery, a powerful new tool for high-resolution mapping in China.
Methods
The proposed method integrates geometric calibration with a GCP-free block adjustment strategy. The core steps are:
- Geometric Calibration: This initial step focuses on identifying and compensating for systematic error sources within individual GF-3 SAR images. This includes addressing fixed system errors (e.g., time delay) and time-varying system errors (e.g., atmospheric propagation delay).
- RPC Model Generation: After geometric calibration, the Rational Polynomial Coefficient (RPC) model is generated. This model provides a generalized geometric representation that can simplify subsequent processing and allow for the use of existing optical imagery processing software. The RPC model is derived by compensating the original RPCs with the parameters obtained from the geometric calibration.
- DEM-Assisted Planar Block Adjustment: This is the key GCP-free component. Instead of using GCPs, the method leverages a Digital Elevation Model (DEM) to assist in the planar block adjustment. Tie points (TPs) between overlapping SAR images are identified, and their elevations are interpolated from the DEM. The orientation parameters of the GF-3 images are then solved through a least-squares adjustment process that incorporates the RPC model and the DEM-derived elevations. This process iteratively refines the image orientation parameters until convergence.
- Ortho-rectification: Finally, using the solved orientation parameters, the SAR images are ortho-rectified to produce accurate orthophotos.
Datasets
The study utilizes GF-3 SAR imagery from two distinct test areas in China:
- Wuhan City, Hubei Province: This area was used with GF-3 images acquired in the QPSI imaging mode. Data included 11 images from 4 orbits.
- Hubei Province: This larger test area was covered by 31 GF-3 images acquired in the FS2 imaging mode, spanning 9 orbits.
The study also mentions the use of a global DEM provided by the United States Geological Survey for DEM interpolation. The DOM and DEM for the Hubei province were generated using ZY-3 satellite stereoscopic surveying and mapping images.
Key Insights and Ideas
- Geometric Calibration is Crucial: The paper strongly emphasizes that geometric calibration of individual SAR images is a vital prerequisite for achieving high absolute positioning accuracy, even before block adjustment. It effectively reduces systematic errors that block adjustment alone cannot fully resolve.
- RPCs as a Bridge: Converting the SAR-specific models (like Range Doppler) to RPCs is a smart move. It allows leveraging established, widely available software and processing pipelines designed for optical imagery, simplifying the workflow and broadening accessibility.
- DEM as a GCP Substitute: The innovative use of a DEM to provide elevation constraints for tie points in the block adjustment process is the core of the GCP-free approach. This effectively replaces the need for ground-based elevation information.
- Sub-pixel Accuracy Achievable Without GCPs: The most significant insight is that high geometric accuracy (better than one pixel) can be achieved for both relative (mosaic) and absolute positioning without relying on GCPs, which was a major bottleneck previously.
Key Results
The experimental results demonstrate the effectiveness of the proposed GCP-free block adjustment method:
- Wuhan Test Area (QPSI mode):
- After geometric calibration and block adjustment, the Independent Check Points' (ICPs) accuracy improved from 29.93 m to 7.31 m. This translates to better than 1 pixel accuracy relative to the 8 m resolution of the QPSI model.
- The RMSE of Tie Points (TPs) was better than 1 pixel, with no significant change before and after calibration, indicating that geometric calibration primarily addresses systematic errors affecting absolute positioning.
- Hubei Test Area (FS2 mode):
- The ICPs' accuracy improved from 38.97 m to 8.97 m after geometric calibration and block adjustment, achieving better than 1 pixel accuracy relative to the 10 m resolution of the FS2 model.
- The study also showed that adding a limited number of GCPs (13) did not significantly improve the ICP accuracy beyond what was achieved with the GCP-free method after calibration.
- Overall: The method consistently achieved geometric mosaic accuracy and absolute positioning accuracy of the orthophoto better than one pixel across different imaging modes and test areas.
Limitations and Future Work
- DEM Dependency: The accuracy of the DEM is a critical factor. While a global DEM was used, higher resolution and more accurate local DEMs could further improve results.
- Weak Convergence: The paper acknowledges that SAR images can sometimes suffer from weak convergence geometry, which can pose challenges for traditional block adjustment. The proposed planar block adjustment with DEM assistance is designed to address this, but further investigation into extreme cases might be warranted.
- Future Work: The authors suggest that this method lays a good geometric foundation for global mapping with GF-3 images. Future work could focus on extending this methodology to other SAR systems and exploring its application in more challenging terrains or for specific applications requiring even higher precision. The paper also implies that further refinement of the geometric calibration process could yield even better results.
Summary updated