| dc.description.abstract | The small size and light weight of CubeSats, as well as the low cost of the launch vehicle, has driven an increase in progress in CubeSat missions. Due to the high flexibility of CubeSats, which are suitable for various mission objectives, we have developed a 4U space size, 1.4kg hyperspectral scientific payload named HyperSCAN (Hyper SpecCral AnNlyzer). The HyperSCAN is onboard the 12U SCintillation and IONsophere eXtended (SCION-X) CubeSat, using the pushbroom method to scan ground information linearly in sweeping mode. The visible and near-infrared spectra are subdivided into 162 bands, and the spectral features in the ground data are analyzed, which can be used to provide research on the classification of terrain and features as well as the measurement of absolute radiance. With a high-sensitivity CMOS sensor, HyperSCAN offers scientifically valuable and low-cost image data for the scientific community and develops the future hyperspectral imager onboard small satellite missions.
The aims of this thesis consist of three aspects: standard data processing flow, scientific calibration, and data validation. The data flow design is divided into three levels: Level 0 raw packet data, Level 1 hyperspectral cubic dataset (including two spatial and one spectral dimension), and Level 2 absolute radiometric calibration data. The data flow design can help the back-end data processing in real-time and provide ground data standardization.
The calibration process for HyperSCAN is a meticulous one, ensuring the accuracy and reliability of the data it provides. This includes field-of-view (FOV) correction, wavelength measurement, and absolute radiometric calibration. For FOV correction, we use parallel light with infinite focal length to carry out FOV experiment, and use precision six-axis to carry out rotation angle adjustment to measure the FOV of ~ 8.0 degrees, corresponding to the image width on the ground is 70.2 km at the 500 km orbit, and along-track length of 33.7 km. The wavelength correction was carried out by using a monochromator, and the wavelength was measured between 430 - 800 nm in the visible-near-infrared band. The FWHM was 8.72 nm for monochromatic light of 550 nm. We calibrated absolute radiometric calibration using a power meter and an integrating sphere for flat field measurement, and converted the readout 10-bit digital number (DN) value into the absolute spectral radiance (watt/cm2-μm-sr). The HyperSCAN ranges from 10-5 to 10-2 watt/cm2-μm-sr, within the typical range of the top-of-atmosphere radiance (10-3~10-2 watt/cm2-μm-sr). The SNR of HyperSCAN is > 15 dB.
We conducted an outdoor experiment for data validation and calibrated absolute spectral radiance using the above data processing flow. Compared with the standard radiation transmission model (MODTRAN), the measured radiance quantities range within the order of magnitude of 10-3~10-2 watt/cm2-μm-sr, closely matching the model results. Finally, we used the principal component analysis (PCA) to reduce the data dimensions where original data have 162 bands. After sorting the weights of the dataset, the linear Support Vector Machine was used to categorize the hyperspectral data. The classification accuracy was as high as 99% for a total of 162 bands. After PCA, if we downscaled the required spectral features to 10 dominant features, the classification accuracy was as high as 99% and required less training time, further demonstrating the reliability of our results. Therefore, the classification results demonstrate that the advantages of PCA, include data dimensionality reduction and spectral feature extraction.
This thesis develops HyperSCAN’s calibrations and data validation experiment, conforming to in-orbit requirements. We verified that the HyperSCAN performance could be applied to the ground data classification and satisfy the criteria of the future development of a small-satellite hyperspectral imager payload. | en_US |