Deep Learning based method for Fire Detection

No Thumbnail Available

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis

Date

2023-10-09

Department

Major/Subject

Machine Learning

Mcode

SCI3113

Degree programme

Master’s Programme in Security and Cloud Computing (SECCLO)

Language

en

Pages

59

Series

Abstract

Fire accidents have become increasingly frequent and have profound effects on today’s society, leading to injuries, fatalities, and significant economic losses. It is crucial to develop effective and early fire detection systems that can promptly detect and prevent fire disasters. Machine learning and computer vision provide a promising solution for the early detection of fires, mitigating potential risks and enhancing safety measures. In this study, we present an extensive and comprehensive fire dataset, surpassing existing datasets in terms of both scale and diversity. This dataset enables robust and thorough training of fire detection models and serves as a benchmark for evaluating future fire detection systems. The core of our fire detection system is the state-of-the-art Yolov5 model, known for its simplicity, speed, and efficiency in object detection tasks. We demonstrate the effectiveness of our proposed model with promising results, achieving an average F1 score of 0.77 and an mAP@0.5 score of approximately 0.77. These metrics reflect the model’s capability to accurately detect fires across various scenarios. Moreover, we take our research further by focusing on the deployment of the trained model to the cloud. The cloud deployment aspect enhances the practicality and accessibility of our fire detection system, making it more scalable and efficient. Furthermore, it opens up avenues for future advancements and integration with other smart technologies, contributing to the development of smarter and safer environments. Overall, this work contributes to the advancement of fire detection systems, offering a robust dataset, a powerful detection model, and an efficient cloud deployment approach. With this research, we aim to foster a safer and more secure environment by reducing the risks posed by fire accidents and enabling timely and effective fire prevention measures.

Description

Supervisor

Jung, Alexander

Thesis advisor

Khajavi, Siavash

Keywords

fire detection, yolo, cloud, machine learning

Other note

Citation