人工智慧已成為影像分析任務中的重要技術，然而在實際遙測應用場景中，模型常受限於計算資源、通訊頻寬以及低延遲需求。因此，為提升模型於衛星與航空平台上的部署能力，發展兼具高效率與低計算成本的輕量化模型已成為重要研究方向。此類模型不僅能有效降低雲端運算，亦有助於提升即時遙測分析的可行與應用性。Sentinel-2 多光譜衛星影像具有廣泛的拍攝覆蓋範圍與高重訪頻率，但所含的光譜解析度有限，且不同波段之間存在空間解析度差異，使其在精細光譜分析上的應用受到影響。另一方面，AVIRIS 高光譜影像能提供更豐富的光譜資訊，卻受限於空間覆蓋範圍及資料獲取成本等因素。為了結合兩者優勢，由多光譜影像重建出高光譜影像的技術對遙測應用也因此具有重要實務價值。然而，Sentinel-2 與 AVIRIS 的成對訓練資料不易取得，使得此重建任務面臨相當大的挑戰性。為解決上述問題，本研究提出結合深度學習與凸優化的 COS2A 演算法。其中，深度學習架構採用深度展開模組設計，使所提出的方法具有輕量化特性，並適用於邊緣導向的遙測應用。該模組首先產生初步的 AVIRIS 高光譜影像估計，並將其作為先驗資訊引入至 Q-二次範數正則化項中，以輔助後續的凸優化重建。接著，本研究進一步引入光譜與空間的對偶性，將光譜超解析問題轉換為空間超解析問題，最後透過耦合矩陣分解求得最終重建結果。根據模擬與真實場景下的實驗結果顯示，COS2A 能夠有效重建具有 AVIRIS 等級光譜特性的高光譜影像，並展現其應用於邊緣人工智慧導向高光譜重建之可行性。

Edge artificial intelligence has become increasingly important in remote sensing applications, where image analysis is often expected to be performed under limited computational resources, communication bandwidth, and latency requirements. For satellite and aerial platforms, lightweight models are therefore desirable for reducing the dependence on cloud-based processing. While Sentinel-2 multispectral imagery offers extensive Earth observation coverage and frequent revisit capability, its limited spectral resolution and heterogeneous spatial resolutions constrain detailed spectral analysis. In contrast, hyperspectral imaging provides rich spectral information, but it usually suffers from limited spatial coverage and high acquisition cost. To leverage these complementary advantages, converting from Sentinel-2 to AVIRIS hyperspectral data has significant practical value for remote sensing applications. However, due to the scarcity of paired Sentinel-2/AVIRIS data, this task becomes a challenging spectral super-resolution (SSR) problem under small training data. To address this problem, this thesis adopts a COS2A framework that combines deep learning with convex optimization. The deep learning component is implemented through deep unfolding, resulting in a lightweight and interpretable module suitable for edge-oriented remote sensing applications. The deep unfolding module generates an initial AVIRIS-level hyperspectral estimate, which is incorporated into a Q-quadratic norm regularization term as a data-driven prior for convex optimization. Spectral-spatial duality is further employed to reformulate the SSR problem into a dual spatial super-resolution problem, which can be solved through coupled matrix factorization. Based on experiments conducted on simulated and real Sentinel-2/AVIRIS data, COS2A can reconstruct AVIRIS-level hyperspectral images effectively and provide a feasible direction for edge-AI-oriented hyperspectral reconstruction.

[1] Y. Shi, J. Zhu, C. Jiang, L. Kuang, and K. B. Letaief, “Satellite edge artificial intelligence with large models: Architectures and technologies,” Science China Information Sciences, vol. 68, no. 7, pp. 170302, 2025.
[2] F. Van der Meer, H. Van der Werff, and F. Van Ruitenbeek, “Potential of ESA’s Sentinel-2 for geological applications,” Remote Sensing of Environment, vol. 148, pp. 124–133, 2014.
[3] D. Phiri, M. Simwanda, S. Salekin, V. R. Nyirenda, Y. Murayama, and M. Ranagalage, “Sentinel-2 data for land cover/use mapping: A review,” Remote Sensing, vol. 12, no. 14, pp. 1–35, July 2020.
[4] J. Segarra, M. L. Buchaillot, J. L. Araus, and S. C. Kefauver, “Remote sensing for precision agriculture: Sentinel-2 improved features and applications,” Agronomy, vol. 10, no. 5, pp. 1–18, May 2020.
[5] C.-H. Lin and J. M. Bioucas-Dias, “An explicit and scene-adapted definition of convex self-similarity prior with application to unsupervised Sentinel-2 super-resolution,” IEEE Transactions on Geoscience and Remote Sensinging, vol. 58, no. 5, pp. 3352–3365, Dec. 2019.
[6] ——, “Nonnegative blind source separation for ill-conditioned mixtures via John ellipsoid,” IEEE Transactions on Neural Networks and Learning Systems, vol. 32, no. 5, pp. 2209–2223, July 2020.
[7] C.-H. Lin, W.-K. Ma, W.-C. Li, C.-Y. Chi, and A. Ambikapathi, “Identifiability of the simplex volume minimization criterion for blind hyperspectral unmixing: The no-pure-pixel case,” IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 10, pp. 5530–5546, May 2015.
[8] “AVIRIS Data Portal,” accessed: Apr. 8, 2025. [Online]. Available: https://aviris.jpl.nasa.gov/dataportal/.
[9] F. Yasuma, T. Mitsunaga, D. Iso, and S. K. Nayar, “Generalized assorted pixel camera: Postcapture control of resolution, dynamic range, and spectrum,” IEEE Transactions on Image Processing, vol. 19, no. 9, pp. 2241–2253, 2010.
[10] C.-H. Lin, Y.-C. Lin, and P.-W. Tang, “ADMM-ADAM: A new inverse imaging framework blending the advantages of convex optimization and deep learning,” IEEE Transactions on Geoscience and Remote Sensinging, vol. 60, pp. 1–16, Sep. 2021.
[11] C.-H. Lin, T.-H. Lin, T.-H. Lin, and T.-H. Lin, “Fast reconstruction of hyperspectral image from its RGB counterpart using ADMM-Adam theory,” in Proc. Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing, Rome, Italy, 13–16 Sep. 2022, pp. 1–5.
[12] Y. Cai, J. Lin, Z. Lin, H. Wang, Y. Zhang, H. Pfister, R. Timofte, and L. Van-Gool, “MST++: Multi-stage spectral-wise transformer for efficient spectral reconstruction,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition Workshops, New Orleans, LA, USA, 18–24 Jun. 2022, pp. 744–754.
[13] Y. Wu, R. Dian, and S. Li, “Multistage spatial-spectral fusion network for spectral super-resolution,” IEEE Transactions on Neural Networks and Learning Systems, vol. 36, no. 7, pp. 12 736–12 746, 2025.
[14] K. Gregor and Y. LeCun, “Learning fast approximations of sparse coding,” in Proc. 27th International Conference on Machine Learning, Haifa, Israel, 21–24 Jun. 2010, p. 399–406.
[15] C.-Y. Chi, W.-C. Li, and C.-H. Lin, Convex Optimization for Signal Processing and Communications: From Fundamentals to Applications. Boca Raton, FL, USA: CRC Press, 2017.
[16] J. Peng, W. Sun, H.-C. Li, W. Li, X. Meng, C. Ge, and Q. Du, “Low-rank and sparse representation for hyperspectral image processing: A review,” IEEE Geoscience and Remote Sensing Magazine, vol. 10, no. 1, pp. 10–43, Jun. 2021.
[17] C.-H. Lin, C.-Y. Chi, Y.-H. Wang, and T.-H. Chan, “A fast hyperplane-based minimum-volume enclosing simplex algorithm for blind hyperspectral unmixing,” IEEE Transactions on Signal Processing, vol. 64, no. 8, pp. 1946–1961, Dec. 2015.
[18] P. V. Giampouras, K. E. Themelis, A. A. Rontogiannis, and K. D. Koutroumbas, “Simultaneously sparse and low-rank abundance matrix estimation for hyperspectral image unmixing,” IEEE Transactions on Geoscience and Remote Sensing, vol. 54, no. 8, pp. 4775–4789, Apr. 2016.
[19] M. Simões, J. Bioucas‐Dias, L. B. Almeida, and J. Chanussot, “A convex formulation for hyperspectral image superresolution via subspace-based regularization,” IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 6, pp. 3373–3388, 2015.
[20] D. D. Lee and H. S. Seung, “Learning the parts of objects by non-negative matrix factorization,” Nature, vol. 401, no. 6755, pp. 788–791, Oct. 1999.
[21] N. Yokoya, T. Yairi, and A. Iwasaki, “Coupled nonnegative matrix factorization unmixing for hyperspectral and multispectral data fusion,” IEEE Transactions on Geoscience and Remote Sensing, vol. 50, no. 2, pp. 528–537, Aug. 2011.
[22] C.-H. Lin, F. Ma, C.-Y. Chi, and C.-H. Hsieh, “A convex optimization-based coupled nonnegative matrix factorization algorithm for hyperspectral and multispectral data fusion,” IEEE Transactions on Geoscience and Remote Sensing, vol. 56, no. 3, pp. 1652–1667, Nov. 2017.
[23] D. Ulyanov, A. Vedaldi, and V. Lempitsky, “Deep image prior,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, 18–23 Jun. 2018, pp. 9446–9454.
[24] N. Parikh and S. Boyd, “Proximal algorithms,” Foundations and Trends® in Optimization, vol. 1, no. 3, pp. 127–239, Jan. 2014.
[25] K. Zhang, Y. Li, W. Zuo, L. Zhang, L. Van-Gool, and R. Timofte, “Plug-and-play image restoration with deep denoiser prior,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 44, no. 10, pp. 6360–6376, Jun. 2021.
[26] K. Zhang, W. Zuo, Y. Chen, D. Meng, and L. Zhang, “Beyond a Gaussian denoiser: Residual learning of deep CNN for image denoising,” IEEE Transactions on Image Processing, vol. 26, no. 7, pp. 3142–3155, Feb. 2017.
[27] M. D. Craig, “Minimum-volume transforms for remotely sensed data,” IEEE Transactions on Geoscience and Remote Sensing, vol. 32, no. 3, pp. 542–552, Aug. 2002.
[28] M.-D. Iordache, J. M. Bioucas-Dias, and A. Plaza, “Sparse unmixing of hyperspectral data,” IEEE Transactions on Geoscience and Remote Sensing, vol. 49, no. 6, pp. 2014–2039, Jan. 2011.
[29] G. C. McDonald, “Ridge regression,” Wiley Interdisciplinary Reviews: Computational Statistics, vol. 1, no. 1, pp. 93–100, July 2009.
[30] C.-C. Hsu, C.-H. Lin, C.-H. Kao, and Y.-C. Lin, “DCSN: Deep compressed sensing network for efficient hyperspectral data transmission of miniaturized satellite,” IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 9, pp. 7773–7789, Sep. 2021.
[31] C.-H. Lin and Y.-Y. Chen, “HyperQUEEN: Hyperspectral quantum deep network for image restoration,” IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–20, May 2023.
[32] “Sentinel-2 Google Earth Engine,” Accessed: Apr. 08, 2025. [Online]. Available: https://developers.google.com/earth-engine/datasets/catalog/COPERNICUS_S2_SR_HARMONIZED.
[33] “ENVI 5.2,” [Online]. Available: https://www.nv5geospatialsoftware.com/.
[34] D. B. Huberman, B. J. Reich, and H. D. Bondell, “Nonparametric conditional density estimation in a deep learning framework for short-term forecasting,” Environmental and Ecological Statistics, vol. 29, no. 4, pp. 677–704, Aug. 2022.
[35] R. Sedgewick and K. Wayne, Algorithms. Boston, MA, USA: Addison-Wesley Professional, 2011.
[36] S. Arora, R. Ge, Y. Halpern, D. Mimno, A. Moitra, D. Sontag, Y. Wu, and M. Zhu, “A practical algorithm for topic modeling with provable guarantees,” arXiv preprint arXiv:1212.4777, Dec. 2012.
[37] R. Achanta, A. Shaji, K. Smith, A. Lucchi, P. Fua, and S. Süsstrunk, “SLIC superpixels compared to state-of-the-art superpixel methods,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 34, no. 11, pp. 2274–2282, Nov. 2012.