A novel deep unsupervised approach for super-resolution of remote sensing hyperspectral image using gompertz-function...

Area of Science:

• Remote Sensing
• Image Processing
• Artificial Intelligence

Background:

• Hyperspectral remote sensing images suffer from high dimensions and low quality, hindering analytical applications due to poor spectral signature revelation.
• Existing image super-resolution (SR) models, both supervised and unsupervised, often lack the required accuracy for high-fidelity image reconstruction.
• Advancements in visual communication necessitate improved methods for generating high-resolution (HR) images from low-resolution (LR) counterparts.

Purpose of the Study:

• To propose an advanced unsupervised learning framework for generating high-resolution hyperspectral images.
• To enhance the spectral and spatial quality of low-resolution hyperspectral images for improved analytics.
• To develop a novel Generative Adversarial Network (GAN) based approach for accurate image super-resolution.

Main Methods:

• A novel Gompertz Function-based Convergence War Accelerometric Optimization-GAN framework was developed for unsupervised image super-resolution.
• The framework incorporates a multi-stage preprocessing pipeline including Shannon-Gaussian Filter (S-GF) for noise removal, Gradient Domain Approach based Tone-Mapping (TM), and skew correction.
• Boundary and edge enhancement using Inverse Gradient Mapping (IGM), patch extraction, and contrast improvement were performed before feeding into the GAN, which utilizes Krzanowski and Li- Kantorovich Metric-K-Means clustering Algorithm (KL-KM-KMA).

Main Results:

• The proposed framework achieved a high accuracy of 98.05% and precision of 97.98% in generating high-resolution images.
• Quantitative metrics demonstrated strong performance with an inception score of 8.71 and a Fréchet Inception Distance (FID) of 36.4.
• The model exhibited reduced clustering and training times, proving its computational efficiency compared to state-of-the-art methods.

Conclusions:

• The developed Gompertz Function-based Convergence War Accelerometric Optimization-GAN framework effectively generates high-resolution hyperspectral images from low-resolution inputs.
• The proposed unsupervised approach significantly improves image quality and analytical potential for hyperspectral remote sensing data.
• The model demonstrates superior performance and efficiency, offering a promising solution for challenging image super-resolution tasks.