An Effective Image Denoising Method for UAV Images via Improved Generative Adversarial Networks

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Unmanned aerial vehicles (UAVs) provide cost-effective remote sensing image collection.
• Noise in UAV imagery leads to content loss, hindering analysis.
• Existing denoising methods may not adequately preserve fine details.

Purpose of the Study:

• To propose a novel deep neural network for denoising UAV images.
• To address the content loss problem caused by noise in UAV remote sensing data.
• To improve the quality of UAV images for subsequent applications.

Main Methods:

• A novel deep neural network leveraging generative adversarial learning.
• Utilizing perceptual reconstruction loss within a min-max game theoretic framework.
• Optimizing the mapping between noisy and clean UAV images.

Main Results:

• Generated denoised images exhibit clearer edges of ground objects.
• Enhanced preservation of detailed textures in ground objects.
• Demonstrated superior performance in image matching and classification experiments compared to traditional methods.

Conclusions:

• The proposed generative adversarial learning method effectively denoises UAV images.
• The method preserves crucial image details, enhancing usability for remote sensing tasks.
• This approach offers a significant improvement for UAV-based image analysis.