Related Experiment Video
Updated: Jun 23, 2026
07:05
Applying Hyperspectral Reflectance Imaging to Investigate the Palettes and the Techniques of Painters
Published on: June 18, 2021
2.4K
Remote sensing image dehazing using generative adversarial network with texture and color space enhancement
Helin Shen1, Tie Zhong2, Yanfei Jia3
1Key Laboratory of Modern Power System Simulation and Control and Renewable Energy Technology (Ministry of Education), Department of Communication Engineering, College of Electric Engineering, Northeast Electric Power University, Jilin, 132012, China.
Scientific Reports
|May 29, 2024
Summary
This study introduces a novel generative adversarial network (GAN) for remote sensing image dehazing. The new method significantly enhances image quality and color recovery in hazy conditions.
Area of Science:
- Earth Observation
- Computer Vision
- Artificial Intelligence
Background:
- Remote sensing is crucial for ground information detection, but image quality is often compromised by atmospheric conditions like haze.
- Classical Convolutional Neural Networks (CNNs) show promise for dehazing but have limitations in feature extraction.
- Generative Adversarial Networks (GANs) offer a promising approach for addressing image dehazing challenges.
Purpose of the Study:
- To propose a novel generative adversarial network (GAN) for reconstructing clean images from hazy remote sensing data.
- To enhance feature extraction capabilities beyond classical CNNs for improved dehazing performance.
- To improve color recovery and overall image quality in hazy remote sensing images.
Main Methods:
- Developed a novel dehazed generative adversarial network (GAN) comprising a generator and a discriminator.
- The generator incorporates color/luminance and high-frequency feature extraction modules for multi-scale information acquisition.
- Introduced a color loss function based on Hue Saturation Value (HSV) for superior color recovery.
- Designed a parallel structure for the discriminator to improve texture and background information extraction.
Main Results:
- The proposed GAN significantly improves the quality of dehazed remote sensing images.
- Experimental results show a notable increase in Peak-Signal-to-Noise Ratio (PSNR) compared to existing methods.
- Dehazed images generated by the proposed method closely resemble original haze-free images in terms of visual quality.
Conclusions:
- The novel GAN architecture effectively addresses the limitations of classical CNNs in remote sensing image dehazing.
- The integrated feature extraction modules and HSV color loss function contribute to superior performance in texture, color, and luminance reconstruction.
- This method offers a significant advancement in obtaining clear and accurate ground information from hazy remote sensing imagery.
