Wildfire Smoke Detection System using Deep Learning

Abstract

This study provides a solution to the problem of detecting wildfire smoke in fire-prone areas. Fire incidents are crucial to the community as they can cause damage to humans and properties. The researcher considers this dilemma to be an example of action detection, and to solve it using deep learning. To create a model in detecting wildfire smokes, this study will utilize the YOLOv3 algorithm which is a great tool in detecting objects in real-time applications. Based on the study’s findings, model 30 yielded the highest mAP of 0.9986 (99.86%) and a training loss score of 3.3692. The generated model, Model 30, was applied to the GUI to test and evaluate the performance of the system. The video testing result yielded a 47.64% accuracy, whereas the live feed or webcam yielded an accuracy of 61.55%. Thus, the system is a viable tool in detecting wildfire smoke in real-time application.