Dehaze-EEGAN: Remote Sensing Image Dehazing Using a Generative Adversarial Network with Edge Enhancement

• Published in: Human-centric computing and information sciences pp. 55 - 82
• Main Authors: Yu Wenshuo, Zhao Liquan, Yanfei Jia
• Format: Journal Article
• Language: English
• Published: 한국컴퓨터산업협회 01.06.2025
• Subjects: 컴퓨터학
• ISSN: 2192-1962
• DOI: 10.22967/HCIS.2025.15.033

Haze frequently deteriorates remote sensing images during the acquisition process, having a substantial impact on later studies. To handle this issue, a novel generative adversarial network with edge enhancement strategy is proposed, named Dehaze-EEGAN. The network is proposed to remove haze while preserving the edge and texture information in remote sensing images. A multi-scale feature extraction module is introduced to improve the capture of feature details from remote sensing images. This module enables the extraction of features across varying receptive field scales. Meanwhile, a novel attention module is designed. This module enhances the extraction of useful feature information by allocating larger weights to significant features. In addition, the discriminative ability of the adversarial network is enhanced by a designed multi-scale discriminative network. By introducing color loss and Charbonnier loss, the generative adversarial network’s standard loss function is improved, which excels at retaining color details within remote sensing images while simultaneously mitigating the risk of gradient vanishing. By the use of simulated and real hazy remote sensing images as research subjects, the dehazing performance of the proposed method and four other methods are evaluated. According to the experimental results, the Dehaze-EEGAN method that had been suggested showed the greatest structural similarity (SSIM) and peak signal-to-noise ratio (PSNR), along with the smallest mean square error, learned perceptual image patch similarity, and natural image quality evaluator. Moreover, images dehazed using our proposed Dehaze-EEGAN method are more distinct than the outcomes from alternative methods. KCI Citation Count: 0