A Full-Spectrum Registration Method for Zhuhai-1 Satellite Hyperspectral Imagery

Jinjun Meng, Jiaqi Wu, Linlin Lu, Qingting Li, Qiang Zhang, Suyun Feng, Jun YanSensors 2020 · DOI

TL;DR

This paper introduces a novel "full-spectrum registration" method for hyperspectral satellite imagery, specifically addressing the challenges posed by the Zhuhai-1 constellation. By combining affine transformation transfer between adjacent bands with Delaunay triangulation for differential correction, the method achieves accurate registration across all spectral bands, even for non-adjacent ones, which is crucial for detailed remote sensing analysis.

What Problem Does This Paper Address?

Accurate registration, or aligning different images of the same area, is fundamental for analyzing remote sensing data. Hyperspectral imagery, which captures light across many narrow spectral bands, presents a unique challenge: features can look very different in different bands due to varying spectral responses. This makes it difficult to find matching points, especially between bands that are far apart in the spectrum (non-adjacent bands). Furthermore, variations in terrain (topography) introduce geometric distortions that a single, simple transformation (like an affine transformation) can't fully correct across the entire image. The Zhuhai-1 satellite constellation, with its hyperspectral sensors, suffers from these registration issues, hindering its application in fields like land mapping and environmental monitoring.

Methods

The proposed method tackles this by breaking down the registration process into two main stages: relative and absolute registration.

  • Relative Registration (Adjacent Bands): This stage focuses on aligning adjacent spectral bands. It employs the Scale-Invariant Feature Transform (SIFT) to detect and match distinctive keypoints. To refine these matches and remove outliers, the Random Sample Consensus (RANSAC) algorithm and a least-squares method are used, resulting in precise matching point pairs.
  • Absolute Registration (Full Spectrum): This is the core innovation.
  • Affine Transformation Transfer: The method leverages the accurate registration between adjacent bands to establish an affine transformation relationship. This relationship is then "transferred" to bridge the gap between non-adjacent bands.
  • Dense Delaunay Triangulation: A Delaunay triangulation is constructed using the fine matching points. This creates a network of triangles across the image.
  • Differential Correction: For each small triangle within the triangulation, an affine transformation is calculated using the transferred relationship. This allows for localized geometric correction, effectively handling the varying distortions caused by terrain. The method iteratively applies this correction across all triangles to achieve full-spectrum registration.

Datasets

The researchers evaluated their method using three datasets from the Zhuhai-1 hyperspectral satellite constellation, covering diverse terrains:

  • OHS-Arizona, USA: A large mountain basin with plain areas.
  • OHS-Guangxi, China: An area with a river, tributaries, alluvial plains, and hilly regions.
  • OHS-Xinjiang, China: Characterized by mountains, deserts, and low-lying oasis plains.

Each dataset consists of 5056 x 5056 pixel images.

Key Insights and Ideas

  • Bridging the Spectral Gap: The most significant insight is the successful transfer of registration information from adjacent to non-adjacent spectral bands. This circumvents the difficulty of directly matching features across spectrally dissimilar bands.
  • Terrain Adaptation: Using Delaunay triangulation for differential correction is key to handling local geometric distortions caused by varying topography, which is a major limitation for global transformation models.
  • Robust Feature Matching: The combination of SIFT for feature extraction and RANSAC for outlier rejection provides a robust foundation for obtaining accurate initial matches.

Key Results

The proposed full-spectrum registration method demonstrated strong performance:

  • Adjacent Band Registration: Achieved an accuracy ranging from 0.2 to 0.6 pixels.
  • Non-Adjacent Band Registration: Structural Similarity Index (SSIM) and Cosine Similarity (CS) measures between non-adjacent bands were consistently above 0.80, indicating high similarity and good registration.
  • Overall Full-Spectrum Accuracy: The final registration accuracy for the entire spectrum was less than 1 pixel.
  • Qualitative Improvement: Visual inspection of true and false-color composite images showed sharper edges and clearer details compared to the existing IOEM method, with the absence of ghosting artifacts.

Limitations and Future Work

  • Computational Cost: While not explicitly detailed as a limitation, the multi-step process involving SIFT, RANSAC, and Delaunay triangulation suggests a potentially higher computational cost compared to simpler methods.
  • Reference Band Selection: The paper mentions selecting the middle or red band as a reference due to better texture and contrast. However, the optimal selection criteria for diverse scenarios might warrant further investigation.
  • Comparison with Deep Learning: The authors suggest that their method could be compared with deep-learning-based approaches for further improvement, hinting at a potential future research direction. They also propose that the method could be adapted for other satellite sensors and for multi-temporal or multi-sensor image registration.

Summary updated

Abstract

Accurate registration is an essential prerequisite for analysis and applications involving remote sensing imagery. It is usually difficult to extract enough matching points for inter-band registration in hyperspectral imagery due to the different spectral responses for land features in different image bands. This is especially true for non-adjacent bands. The inconsistency in geometric distortion caused by topographic relief also makes it inappropriate to use a single affine transformation relationship for the geometric transformation of the entire image. Currently, accurate registration between spectral bands of Zhuhai-1 satellite hyperspectral imagery remains challenging. In this paper, a full-spectrum registration method was proposed to address this problem. The method combines the transfer strategy based on the affine transformation relationship between adjacent spectrums with the differential correction from dense Delaunay triangulation. Firstly, the scale-invariant feature transform (SIFT) extraction method was used to extract and match feature points of adjacent bands. The RANdom SAmple Consensus (RANSAC) algorithm and the least square method is then used to eliminate mismatching point pairs to obtain fine matching point pairs. Secondly, a dense Delaunay triangulation was constructed based on fine matching point pairs. The affine transformation relation for non-adjacent bands was established for each triangle using the affine transformation relation transfer strategy. Finally, the affine transformation relation was used to perform differential correction for each triangle. Three Zhuhai-1 satellite hyperspectral images covering different terrains were used as experiment data. The evaluation results showed that the adjacent band registration accuracy ranged from 0.2 to 0.6 pixels. The structural similarity measure and cosine similarity measure between non-adjacent bands were both greater than 0.80. Moreover, the full-spectrum registration accuracy was less than 1 pixel. These registration results can meet the needs of Zhuhai-1 hyperspectral imagery applications in various fields.
BibTeX
@inproceedings{meng2020full,
  author = {Jinjun Meng and Jiaqi Wu and Linlin Lu and Qingting Li and Qiang Zhang and Suyun Feng and Jun Yan},
  title = {{A Full-Spectrum Registration Method for Zhuhai-1 Satellite Hyperspectral Imagery}},
  booktitle = {Sensors},
  year = {2020},
  doi = {10.3390/s20216298},
}
Figure