Remote sensing image dehazing using a wavelet-based generative adversarial networks.

Area of Science:

• Remote Sensing
• Computer Vision
• Image Processing

Background:

• Atmospheric haze in remote sensing images degrades data quality and utility.
• Existing dehazing methods may struggle with preserving details and color fidelity.

Purpose of the Study:

• To develop a novel generative adversarial network (GAN) based dehazing method for remote sensing images.
• To enhance image clarity, detail, and color accuracy while removing atmospheric haze.

Main Methods:

• A generator network incorporating dense residual blocks, wavelet transform, and global/local attention mechanisms.
• A PixelShuffle upsampling operation for fine control of image details.
• An improved discriminator network with a noise module for enhanced robustness.
• A novel loss function combining color and SSIM losses with traditional losses.

Main Results:

• The proposed method achieved the highest Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) scores compared to leading methods.
• The dehazing technique effectively maintained color fidelity and image details.
• Generated images exhibited significantly improved clarity.

Conclusions:

• The novel GAN-based dehazing approach is highly effective for remote sensing applications.
• The method offers superior performance in terms of image quality, color accuracy, and detail preservation.
• This technique addresses the challenge of atmospheric haze degradation in remote sensing data.