SSR-GAN: super resolution-based generative adversarial networks model for flood image enhancement

SSR-GAN: super resolution-based generative adversarial networks model for flood image enhancement

---

연구 분야: Artificial Intelligence

---

---

학회: Signal, Image and Video Processing

---

초록

Floods, a common natural disaster, it affects more than half of all natural disasters, primarily due to high floods, high tides, heavy rainfall, and human activity. Distinguishing between the flooded and non-flood areas during the disaster is dependent on the quality of images collected from different sources. Since synthetic aperture radar (SAR) images can see through cloud cover and capture photos of the Earth's surface during bad weather, they are excellent for detecting floods. Image Enhancement is the approach that helps to increase the resolution of the SAR images, which helps to more accurately categorize the flooded and non-flooded areas. The quality of satellite and aerial imagery can be enhanced to recognize flooded and non-flooded areas. Our proposed approach uses a super-resolution technique to enhance the resolution of SAR images. We develop a super resolution-based generative adversarial network, or SSR-GAN (Satellite Super Resolution-based Generative adversarial network). In our approach to estimating generative models through an adversarial process, we concurrently train two different models: a generative model G and a discriminative model D. An image with low resolution is transformed into high pixel density. The peak signal-to-noise ratio (PSNR), the structural similarity index(SSIM), Multiscale Structural Similarity (MSSIM) and mean squared error (MSE) are the performance metrics used to compare the interpolation methods for enhancing resolution. In comparison to the existing approaches with our proposed model SSR-GAN, the values of PSNR, SSIM, and MSSIM are greater, and the MSE error value is lower.

---