A Critical Examination of SAR Colorization Impact on Flood Mapping Accuracy

Synthetic Aperture Radar (SAR) images are commonly utilized in remote sensing applications. Despite their prevalence, SAR images typically lack color information, presenting challenges in interpretation due to inherent speckle noise and monochromatic characteristics. To address this issue, a research focus on SAR colorization has emerged, aiming to infuse color into grayscale SAR images while preserving original spatial and radiometric details. Notably, existing researches focus on SAR colorization without delving into its performance or real-world impact, such as flood mapping. This paper evaluates the practical implications and effectiveness of SAR colorization in improving flood mapping precision. Two recently developed CNNs, namely the Conditional Generative Adversarial Network (cGAN) and the Spatial-spectral Convolutional Neural Network (CNN4ColSAR), exhibiting noteworthy quantitative evaluation metrics in SAR colorization, are employed for the colorization process. Additionally, a proposed 3D adaptation of the UNet model, referred to as 3D-UNet, is utilized for flood mapping and shows enhanced proficiency in extracting information from colorized SAR images compared to its 2D counterpart. Quantitative and qualitative results highlight the effectiveness of the CNN4SARcolor colorized network in conjunction with 3D-UNet for accurately mapping flood-induced changes.