Modified pix2pixHD for Enhancing Spatial Resolution of Image for Conversion from SAR Images to Optical Images in Application of Landslide Area Detection
Author:
Open Access
- 21 February 2025
- journal article
- research article
- Published by MDPI AG in Information
- Vol. 16 (3) , 163
Abstract
A method for the conversion of SAR (Synthetic Aperture Radar) images to optical images can be useful for disaster area detection for the following two reasons: (1) it is easier to detect disaster areas with optical images rather than with SAR images; (2) disasters may occur at night and in rainy and cloudy conditions (SAR images can be acquired in daytime and nighttime as well as all weather conditions). Therefore, it becomes easier to detect disaster areas with optical images converted from SAR images. Using GANs (Generative Adversarial Networks), it is possible to convert SAR images to optical images. In particular, pix2pix and pix2pixHD are used for this purpose. The author proposed spatial resolution-maintained pix2pixHD previously. In this paper, a new method of modifying pix2pixHD with a spatial attention mechanism and an edge enhancement mechanism with a Canny filter in the loss function is proposed, and the proposed method is compared to the pix2pixHD with a spatial attention mechanism and pix2pixHD as well as pix2pix. All of these four methods are compared in terms of the spatial resolution (frequency components) of converted optical images. By experiment, the superiority of the modified pix2pixHD with spatial attention and edge enhancement mechanisms is confirmed for disaster area detection (landslide area detection).Keywords
This publication has 9 references indexed in Scilit:
- Method for Landslide Area Detection Based on EfficientNetV2 with Optical Image Converted from SAR Image Using pix2pixHD with Spatial Attention Mechanism in Loss FunctionInformation, 2024
- Image-to-Image Translation: Methods and ApplicationsIEEE Transactions on Multimedia, 2021
- Generative adversarial networksCommunications of the ACM, 2020
- Semantic Image Synthesis With Spatially-Adaptive Normalization2019
- THE SEN1-2 DATASET FOR DEEP LEARNING IN SAR-OPTICAL DATA FUSIONISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2018
- High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs2018
- StackGAN: Text to Photo-Realistic Image Synthesis with Stacked Generative Adversarial Networks2017
- Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks2017
- Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network2017