Hyperspectral Image Enhancement Using Enhanced Deep Image Prior

DOI

https://www.doi.org/10.59256/indjcst.20260501027

Published In

Volume 5, Issue 1 (2026)

Published Online

January-April 2026

Pages

186-192

Article Type

Original Article

Check for updates

Enhanced Deep Image Prior (EDIP) is a lightweight framework designed to enhance hyperspectral images without the need for large-scale training datasets. Instead of relying on pretraining, it learns directly from each input image by leveraging the concept of Deep Image Prior (DIP), which allows the network structure itself to act as a prior for reconstruction and enhancement. The method employs a simplified U-Net architecture tailored for hyperspectral data, effectively capturing spatial–spectral correlations while maintaining computational efficiency. To further improve performance, EDIP incorporates scene-aware adaptation, enabling image-specific optimization that adjusts parameters dynamically to the characteristics of each scene. In addition, a basic yet effective spectral band fusion strategy is applied to preserve fine spectral and spatial details across different wavelengths, ensuring consistency in the enhanced output. The framework is complemented by an interactive web-based tool that visualizes the entire enhancement pipeline, providing intuitive before-and-after comparisons and enabling users to explore the effects of EDIP on diverse hyperspectral datasets.