Disaster Detection on the Fly: Optimized Transformers for UAV

Project Code :TCMAPY2416

Objective

The objective of this project is to develop an optimized deep learning solution for the classification of disaster-related scenarios using UAVs (Unmanned Aerial Vehicles). Specifically, the project focuses on classifying five key categories: collapsed buildings, fire, flooded areas, traffic incidents, and normal conditions. By leveraging the EfficientViT and DaViT (Dual Attention Vision Transformer) models, the primary goal is to achieve high accuracy in classifying disaster scenarios from aerial images. This system aims to provide faster decision-making capabilities, enabling timely response to natural and man-made disasters, ultimately reducing damage and enhancing disaster management efforts.

Abstract

This project aims to develop an optimized solution for the classification of disaster-related scenarios using advanced deep learning techniques. The dataset comprises multiple disaster categories, including collapsed buildings, fire, flooded areas, traffic incidents, and normal conditions. The classification task is tackled using two transformer-based models, EfficientViT and DaViT, which are known for their high accuracy and efficiency in handling complex image classification tasks. The EfficientViT_B1 model achieves an accuracy of 98.03%, while the DaViT_Tiny model shows a slight improvement with an accuracy of 98.14%. The models are evaluated based on metrics such as accuracy, precision, recall, and F1-score to ensure robust performance. This system, aimed at UAVs, assists in classifying disaster scenarios from aerial imagery, enabling faster response times and effective disaster management strategies. The user interface is built as a web application, offering an easy-to-use platform for UAV operators to quickly assess disaster areas from real-time images.

Keywords: Disaster Detection, UAV, Image Classification, EfficientViT, DaViT, Deep Learning, Transformer Models, Disaster Management, Aerial Imagery, Classification Accuracy.

NOTE: Without the concern of our team, please don't submit to the college. This Abstract varies based on student requirements.