MUMBAI, India, June 22 -- Intellectual Property India has published a patent application (202641048443 A) filed by Nandha Engineering College on April 16, 2026, for An Image Based Hybrid Machine Learning Framework For Disaster Prediction.
Inventors include Vennila V; Elamathi M; Keerthana M N; Narenthiran M; and Sivasangari K.
The application for the patent was published on June 12, 2026, under issue no. 24/2026.
Abstract: The present disclosure relates to an image-based hybrid disa.~tcr prediction system designed to accurately classify disaster types from visual data using a combination of deep learning and classical machine learning techniques. With the increasing availability of large·scalc visual data from satellites, drones, surveillance systems, and social media platforms, automated disaster analysis has become essential for timely decision-making and effective emergency response. The proposed system employs a pre-trained Convolutional Neural Network (CNN), specifically ResNet-18, to extract high-level semantic and spatial features from disaster images, capturing critical patterns such as fire spread, water accumulation, debris, and stmctural damage. These deep feature representations are further processed using Random Forest (RF) and Support Vector Machine (SVM) classifiers to improve classification robustness and effectively model complex non-linear relationships. A weighted ensemble fusion mechanism integrates predictions from CNN, RF, and SVM models, enhancing accuracy, stability, and generalization across diverse disaster scenarios. The system is capable of multi-class classification, including categories such as damaged infrastructure, fire disaster, human damage, land disaster, water disaster, and non-damage. To ensure practical usability, the system is deployed through a Strearnlit·based web interface that supports real-time image upload, instant prediction, adjustable ensemble parameters, and vistmlization of confidence scores. The architecture is modular and scalable, allowing easy integration of new datasets, disaster categories, and advanced models. Experimental evaluation demonstrates that the proposed hybrid approach achieves high accuracy (approximately 93%) with strong precision, recall, and Fl-score, outperforming traditional single-model methods. By addressing challenges such as class imbalance, visual similarity. and environmental variations, the system provides a reliable, efficient, and scalable solution for real-world disaster monitoring, emergency response, and intelligent decision support applications
Disclaimer: Curated by HT Syndication.
