Deep learning for monitoring forest fires, Aïn Mimoune forest application
No Thumbnail Available
Date
2022
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
In recent years, major fires with devastating consequences have occurred in various regions
of Algeria. The region of the wilaya of Khenchela was affected by heat waves between
July and August which caused major fires in several forests, precisely in the region of Ain
Mimoune. This thesis addresses the problem of fire and smoke detection in videos, with
the aim of early detection of fires in the forests of Aïn Mimoune.
As we know current methods use traditional image processing methods that rely primarily
on human-made extracted features, which aren’t always appropriate to all forest
environments.
To solve this issue, we have proposed a custom model based on deep learning techniques
that helps in early fire and smoke detectors, the final proposed model integrates
two independent models. The first one uses the features extraction power of VGG16 to
avoid false positives, while the second one uses YOLOV5 that highlight fire or smoke
with a surrounding bounding box.
The proposed solution takes into consideration the time and GPU resources issues, so it
could be applicable in real-world scenarios.
To test our proposed solution on a forest-similar environment, we have created a forestsimulation
environment using the Unity3D game engine, which gave us good results even
with a lack of GPU resources, this step has opened a big door for future perspectives and
suggestions that can be done concerning the fire detection problem.
In the end, to make our solution public and available to everyone, we have used the
Django framework to create a web application interface that uses the final proposed
algorithm.