Abstract

Pan-sharpening combines multidimensional and panchromatic satellite images to generate higher-resolution products. This research compares standard component replacement and index injection approaches to current deep learning methods such as convolutional and adversarial generative networks. A comparative examination utilizing spatial/spectral quality criteria and visual evaluation demonstrates that deep learning algorithms enhance color and feature retention, despite the greater computing cost. Traditional methods successfully brighten features, but they exhibit greater spectrum aberrations. Overall, this trade-off analysis, informed by quantitative and qualitative performance parameters, aids in the appropriate selection of pan-sharpening paradigms adapted to application goals for high fidelity fused Earth observation data. Deep learning has enormous potential for enhancing the state-of-the-art if spectral integrity is prioritized above spatial sharpness alone.

Description