Related Experiment Video
Updated: Jul 11, 2025
07:05
Applying Hyperspectral Reflectance Imaging to Investigate the Palettes and the Techniques of Painters
Published on: June 18, 2021
2.4K
Dehazing in hyperspectral images: the GRANHHADA database
Sol Fernández Carvelo1,2, Miguel Ángel Martínez Domingo3, Eva M Valero3
1Andalusian Institute for Earth System Research (IISTA), University of Granada, Granada, Spain.
Scientific Reports
|November 13, 2023
Summary
This study compares hyperspectral image dehazing methods, introducing the GRANHHADA database. Band-per-band dehazing using specific spectral bands offers superior results compared to sRGB dehazing for outdoor scenes.
Area of Science:
- Remote Sensing
- Image Processing
- Computer Vision
Background:
- Haze significantly degrades the quality of hyperspectral images (HSIs) captured in outdoor environments.
- Effective dehazing techniques are crucial for accurate analysis and interpretation of HSIs in various applications.
- Existing dehazing methods often struggle with the unique spectral characteristics of HSIs.
Purpose of the Study:
- To analyze and compare different dehazing techniques specifically for hyperspectral images.
- To introduce a novel hyperspectral image database (GRANHHADA) for evaluating dehazing algorithms.
- To investigate the impact of spectral band selection and processing strategies on dehazing performance.
Main Methods:
- Development and utilization of the GRANada Hyperspectral HAzy Database (GRANHHADA) with 35 diverse hazy scenes.
- Implementation of Multi-Scale Convolutional Neural Network (MS-CNN) for dehazing experiments.
- Three experimental approaches: optimal spectral band selection, sRGB image dehazing, and individual spectral band dehazing.
Main Results:
- Identification of specific near-infrared spectral bands as highly effective for hyperspectral image dehazing.
- Band-per-band dehazing of individual spectral bands outperformed sRGB dehazing, especially in scenes with high atmospheric dust.
- Using a reduced set of optimal spectral bands can decrease processing time and improve dehazing quality over sRGB methods.
Conclusions:
- Individual spectral band dehazing is a more effective strategy than sRGB dehazing for hyperspectral images.
- The selection of appropriate spectral bands significantly impacts dehazing performance and efficiency.
- Findings offer valuable insights for advancing hyperspectral image processing in remote sensing and other fields.
