Enhancing the Quality of Satellite Images in Disaster Management: A Comparative Analysis of Zero-DCE, CIDNet, and MIRNet Models

Abstract

Satellite images play a critical role in disaster management and rescue operations in natural disasters. However, these images create difficulties in analysis due to noise and loss of detail. This study evaluates the performance of deep learning models—Zero-DCE, CIDNet, and MIRNet—in enhancing low-light satellite images. The performances of the models were analyzed with metrics such as PSNR, SSIM and LPIPS using post-earthquake satellite images of the Hatay region. In addition, the performance of the models on a benchmark dataset was analyzed. The results showed that CIDNet was superior in detail and structural accuracy, while MIRNet was successful in color and brightness enhancement. Although Zero-DCE was effective in brightness enhancement, it lagged behind other models in structural accuracy. In this study, the potential of deep learning-based image enhancement models in disaster management and the image features on which they are effective were revealed. © 2025 Afet ve Acil Durum Yonetimi Baskanligi (AFAD). All rights reserved.