- 36 views
- 28 downloads
E-SSRAN: Resolution Enhancement by Merging Multispectral and Hyperspectral Satellite Data
-
- Author / Creator
- Padhiar, Karansinh A
-
Remote sensing is an effective tool to monitor and assess the dynamics across the Earth’s surface. Despite significant technological advancements, there remains a constant demand for high-resolution remote sensing data in spatial, spectral, and temporal contexts. Such high-resolution data is crucial for accurate analysis in diverse fields including agriculture, environmental monitoring, disaster mapping, and many more. However, acquiring such high-resolution remote sensing data is often expensive, with limited geographic coverage and restricted accessibility for general researchers. To address these challenges, our study leverages the power of convolutional neural networks (CNNs) to enhance the resolution of open-source satellite images. We introduce E-SSRAN (Extended Spatial-Spectral Residual Attention Network), an algorithm that integrates the complementary spatial and spectral characteristics of multispectral images (MSIs) and hyperspectral images (HSIs). Our approach consists of two key modules: Spectral-Enhancement and Spatial-Enhancement. The Spectral-Enhancement module uses E-SSRAN to enhance the spectral resolution of MSI by learning the mapping between MSI and HSI, while the Spatial-Enhancement module uses E-SSRAN to improve the spatial resolution of MSI by mapping low-resolution MSI to high-resolution MSI. Additionally, we proposed a comprehensive pipeline to merge the outputs of the Spatial- and Spectral-Enhancement modules, resulting in satellite images with enhanced spatial and spectral resolution. We evaluated the performance of E-SSRAN using MSI and HSI data acquired from Sentinel-2B and DESIS (DLR Earth Sensing Imaging Spectrometer), respectively, focusing on agricultural areas in Central Alberta. Both qualitative and quantitative assessments demonstrate the effectiveness of E-SSRAN in enhancing the spatial and spectral resolution of low-resolution satellite images. The Spectral-Enhancement module achieved a mean absolute error (MAE) of 0.014, a root mean squared error (RMSE) of 0.018, a spectral angle mapper (SAM) score of 0.144, a universal image quality index (UIQI) score of 0.96, a peak signal-to-noise ratio (PSNR) of 37 dB, and a structural similarity index measure (SSIM) of 0.95. The Spatial-Enhancement module successfully improved the spatial resolution of the 20 m and 60 m ground sampling distance (GSD) Sentinel-2B bands to 10 m GSD. Furthermore, the 10 m GSD MSI produced by the Spatial-Enhancement module was subsequently processed through the Spectral-Enhancement module, generating 10 m GSD HSI. This integrated approach produces satellite images with high spatial and spectral resolution, demonstrating that E-SSRAN significantly improves the quality of satellite images by optimizing the trade-off between spatial and spectral resolutions. Our method has the potential to add substantial value to various applications in agriculture and beyond, by providing more detailed and accurate insights for satellite data analysis.
-
- Graduation date
- Fall 2024
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- This thesis is made available by the University of Alberta Library with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.