A Framework for Wildfire Inspection Using Deep Convolutional Neural Networks
Iuliu Novac, Kenneth Richard Geipel, Jacobo Eduardo de Domingo Gil, Lucas Goncalves de Paula, Kristian Hyttel, Dimitrios Chrysostomou
Abstract
This paper presents the details of a holistic framework designed for wildfire inspection and estimation of its geolocation. The system is built around a low-cost, commercial quadcopter, and the main areas of interest we address in this paper are the semi-autonomous navigation of the drone, the training and classification of fire using deep convolutional neural networks, the estimation of the size and location of the wildfire and the real-time feedback and communication with the user. The evaluation of the functionality of the system demonstrates that with the combination of the proposed techniques we can successfully detect and classify fire in video streams at 19.2 FPS while we can calculate the size and location of the fire with an accuracy of 60.76%.