Paper Title (use style: paper title)

Keywords—YOLOv3, Faster R-CNN, Siamese Fire detection, wildfire detection, YOLOv5, YOLOv6, SSD, forest fire, fire emergencies.

1. Introduction (Heading 1)

Fire emergencies are a growing threat around the world. There were over 66,000 wildfires worldwide in 2022, burning an estimated 7.5 million acres. These fires also claimed the lives of more than 3000 people worldwide [1]. The adverse effects of fires are not limited to the immediate area where the fire occurs. Smoke of wildfires can travel long distances, polluting air in other regions. The economic losses from wildfires can also have a ripple effect, impacting businesses and communities that are not directly affected by the fire.
We can divide fire detection systems into two categories, traditional and modern fire detection systems. Traditional fire detection systems are fire lookout towers, Aerial surveillance, Ground patrols, Remote sensors. These traditional fire detection systems have already been considered as an outdated method because of some disadvantages such as relatively inexpensive cost, labor-intensive as well as they cannot detect fires early on. That is why a lot of organizations, businesses, governments that are responsible for managing fire risks using modern wildfire detection systems which are based on artificial intelligence technologies. They can help us to spot fires more accurately and much faster. Moreover, modern fire detection systems can detect fires from a long distance as well as they do not rely on human labor. According to above advantages of modern wildfire detection systems, we are planning these technologies in Uzbekistan for detecting fires in residential areas, forests as well as industrial areas early on. In terms of fire detection systems based on Artificial Intelligence, we use deep learning algorithms which can detect objects (fire or smoke) in a real time. We have a number of options in this case and our goal is to compare these algorithms and find and apply most reliable, the fastest as well as accurate one in Uzbekistan.
Despite of a lot of advancements in technology, current fire detection systems are often not able to detect fires early enough to prevent them from spreading. This is a major challenge, as early detection is essential for preventing loss of life and property. Moreover, due to lack of accuracy of deep learning models which are being used for building modern fire detection systems, fire detection systems can sometimes miss fires, or they can incorrectly identify non-fire events as fires. This can lead to false alarms, which can waste resources and divert attention from real fires.
These days, there are a lot of real time object detection algorithms which we will apply in our fire detection systems such as SSD, Faster R-CNN, YOLO. However, in this article we will be looking at different versions of YOLO (You Only Look Once) algorithm, including YOLOv8, YOLOv7, YOLOv6, YOLOv5, YOLO by training these models on our custom dataset of smoke and fire images.

2.