Session
Advanced Technologies 1
Location
Salt Palace Convention Center, Salt Lake City, UT
Abstract
Hyperspectral imaging in the shortwave-infrared spectrum is valuable for orbital Earth observation, but its use on low-cost, resource-constrained small satellites is limited by thermally dependent dark current, fixed-pattern noise, and intermittent access to matched calibration frames. Traditional mitigation through active cooling, shutters, or frequent dark-frame acquisition can be difficult to accommodate within small-satellite mass, power, volume, and operations budgets. This paper presents a physics-informed independent calibration tool that reconstructs condition-matched dark response from acquisition metadata rather than relying on hardware-heavy calibration at every operating point. The framework combines a stochastic dark-noise simulator, a metadata-conditioned neural dark operator that predicts spectral-spatial dark response from telemetry such as temperature and integration time, and HSDR-ScanFusion, a hybrid transformer and selective-scan denoising backbone for residual hyperspectral restoration. The simulator models Arrhenius temperature response, Poisson dark-electron generation, photo-response non-uniformity, read noise, hot pixels, ADC bias, and smooth detector-scale thermal variation. Real Xenics XEVA-859 dark-frame experiments provide a hardware validation path: a strict 10 degrees Celsius temperature hold-out test achieves 40.17 dB PSNR and 0.982 SSIM for metadata-conditioned dark-frame prediction, while simulator calibration recovers the measured Arrhenius trend with activation energy near 0.256 eV. On synthetic full-cube evaluation, the learned denoising pipeline improves PSNR from 19.43 dB for the uncorrected noisy cube to 38.73 dB after restoration and reduces spectral angle from 12.06 degrees to 3.74 degrees. These results support a scalable software calibration approach for small-form-factor hyperspectral missions where dark-frame acquisition is limited, thermal state changes over time, and spectral fidelity must be preserved without adding complex calibration hardware.
Document Type
Event
Metadata-Conditioned Dark-Frame Modelling and HSDR-ScanFusion Denoising: A Physics-Informed Independent Tool for Low-Cost Hyperspectral Small-Satellite Calibration
Salt Palace Convention Center, Salt Lake City, UT
Hyperspectral imaging in the shortwave-infrared spectrum is valuable for orbital Earth observation, but its use on low-cost, resource-constrained small satellites is limited by thermally dependent dark current, fixed-pattern noise, and intermittent access to matched calibration frames. Traditional mitigation through active cooling, shutters, or frequent dark-frame acquisition can be difficult to accommodate within small-satellite mass, power, volume, and operations budgets. This paper presents a physics-informed independent calibration tool that reconstructs condition-matched dark response from acquisition metadata rather than relying on hardware-heavy calibration at every operating point. The framework combines a stochastic dark-noise simulator, a metadata-conditioned neural dark operator that predicts spectral-spatial dark response from telemetry such as temperature and integration time, and HSDR-ScanFusion, a hybrid transformer and selective-scan denoising backbone for residual hyperspectral restoration. The simulator models Arrhenius temperature response, Poisson dark-electron generation, photo-response non-uniformity, read noise, hot pixels, ADC bias, and smooth detector-scale thermal variation. Real Xenics XEVA-859 dark-frame experiments provide a hardware validation path: a strict 10 degrees Celsius temperature hold-out test achieves 40.17 dB PSNR and 0.982 SSIM for metadata-conditioned dark-frame prediction, while simulator calibration recovers the measured Arrhenius trend with activation energy near 0.256 eV. On synthetic full-cube evaluation, the learned denoising pipeline improves PSNR from 19.43 dB for the uncorrected noisy cube to 38.73 dB after restoration and reduces spectral angle from 12.06 degrees to 3.74 degrees. These results support a scalable software calibration approach for small-form-factor hyperspectral missions where dark-frame acquisition is limited, thermal state changes over time, and spectral fidelity must be preserved without adding complex calibration hardware.
