Remote Sensing Image Dehazing through an Unsupervised Generative Adversarial Network.

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• Haze significantly degrades remote sensing image quality, hindering information extraction.
• Existing dehazing methods struggle with the complexities of remote sensing imagery.

Purpose of the Study:

• To develop an unsupervised generative adversarial network (GAN) for effective remote sensing image dehazing.
• To improve the visual quality and interpretability of remote sensing data affected by haze.

Main Methods:

• Proposed an unsupervised GAN with two generators and two discriminators for image dehazing.
• Incorporated multi-scale feature-extraction and attention modules within an encoder-decoder generator architecture.
• Introduced an improved loss function with color-constancy loss and designed a multi-scale discriminator.

Main Results:

• Achieved the highest peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) metrics compared to state-of-the-art methods.
• Demonstrated superior performance in effectively removing haze from remote sensing images.
• The proposed attention modules effectively guide feature extraction for haze and texture emphasis.

Conclusions:

• The unsupervised GAN effectively mitigates haze in remote sensing images.
• The novel architecture and loss function lead to significant improvements in image quality.
• The method offers a promising solution for enhancing remote sensing data analysis.