MUMBAI, India, May 29 -- Intellectual Property India has published a patent application (202641061836 A) filed by Vignan's Institute Of Information Technology, Visakhapatnam, Andhra Pradesh, on May 15, for 'satellite image enhancement using gan with explainable ai for flood detection.'

Inventor(s) include Dr. Y. Salini; Pachigulla Deepthi; Vechalapu Swamy Hyma Kumar; Donthemsetti Nandini; and Samineni Nikesh.

The application for the patent was published on May 29, under issue no. 22/2026.

According to the abstract released by the Intellectual Property India: "Floods are among the most frequent and devastating natural disasters, causing severe loss of human life, large-scale damage to infrastructure, and long-term environmental impacts. Timely and accurate flood detection is critical for effective disaster preparedness and response. Satellite imagery plays a central role in flood monitoring; however, its practical utility is often hindered by low spatial resolution, atmospheric noise, cloud cover, and sensor limitations, which reduce the reliability of flood identification and extent assessment. To address these challenges, this project presents a deep learning pipeline that combines Generative Adversarial Networks (GANs) for satellite image super-resolution with semantic segmentation for pixel-level flood detection, supported by Explainable AI (XAI) techniques for decision transparency. [0018] Specifically, an Enhanced Super-Resolution GAN (ESRGAN) with Residual-in-Residual Dense Block (RRDB) architecture is trained to perform 4 super-resolution on Sentinel-2 satellite images from the Sen1Floods11 dataset. The super-resolved images are then passed to a U-Net segmentation model with an EfficientNet-B4 encoder to classify each pixel as flooded or non-flooded. Explainability is incorporated through both Grad-CAM (Gradient-weighted Class Activation Mapping) and SHAP (SHapley Additive Explanations), which generate visual attribution maps that highlight the image regions most responsible for flood predictions. The complete pipeline is deployed as an interactive web application using Gradio, trained on Google Colab with GPU acceleration. Experimental results demonstrate strong performance with validation PSNR reaching 17.9 dB and SSIM 0.63 for the GAN module, and Intersection-over-Union (IoU) of 0.63 for the segmentation module, confirming the effectiveness of the proposed system for real-world satellite-based flood monitoring."

Disclaimer: Curated by HT Syndication.