<![CDATA[Modern deep space exploration generates massive volumes of high-resolution imagery, creating a significant bottleneck for data transmission over limited bandwidths. This research addresses this challenge by evaluating Singular Value Decomposition (SVD) as a numerical Low-Rank Approximation technique for compressing astronomical data. Unlike data-hungry deep learning models, this study applies matrix factorization directly to raw 16-bit .TIF images from the Hubble Space Telescope, varying the rank (k) to analyze the trade-off between storage efficiency and reconstruction fidelity. Experimental results demonstrate that SVD offers a stable compression mechanism, effectively filtering sensor noise while preserving coherent celestial structures. The analysis identifies rank k=100 as the optimal threshold, achieving an average Peak Signal-to-Noise Ratio (PSNR) of 30.81 dB with a Compression Ratio of 12.87 times. These findings suggest that SVD provides a computationally efficient and mathematically deterministic alternative to complex neural networks for onboard satellite data processing, successfully balancing scientific accuracy with transmission constraints.]]>
Copyrights © 2026
