Abstract:In addition to destroying the natural environment and causing loss of private and public property, human-caused and natural fires also greatly endanger public safety. As a result, accurately assessing the fire situation and keeping an eye on it in real time are highly valuable both theoretically and practically. In order to address the shortcomings of earlier methods and raise the model's detection accuracy even more, we proposed a smoke and fire target identification algorithm in this thesis that is based on YOLOv8. Enhancing the C2f module by expanding the algorithm's capacity for feature extraction without adding more model parameters by utilizing deformable convolution in the backbone network; To increase the feature fusion capabilities, the Dynamic Head, a target detection head based on the attention mechanism, replaces the original model's detection head. Furthermore, numerous inanimate items that share similarities with fire and smoke characteristics are wrongly identified in the scene, and there are false alarms and missed alarms in the current studies on fire and smoke recognition. On the other hand, smoke and fire in a fire situation have dynamic characteristics like diffusion and spreadability. Static targets can be recognized as false alarm scenarios, while dynamic targets can be recognized as locations where smoke and fire actually occur. In order to differentiate between dynamic and static targets, the SSIM metric is introduced into the DeepSORT algorithm, which is enhanced based on the target identification algorithm. The enhanced DeepSORT algorithm essentially addresses the issues of false alarms and missed alarms that arise during the detection phase, while the enhanced target identification method raised the average accuracy by 1.5% over the initial YOLOv8 algorithm.

Embedded smoke recognition system based on object detection

Deep Learning Based Fire Detection System For Surveillance Videos

A Fire Monitoring and Alarm System Based on YOLOv3 with OHEM

Forest Fire Alarm System Based on Object Detection

An Integrated Smoke Detection Method Based on Convolutional Neural Network and Image Processing

An Embedded Portable Lightweight Platform for Real-Time Early Smoke Detection

Research on Fire Smoke Detection Algorithm Based on Improved YOLOv8

Real Time Fire and Smoke Detection System

Early Smoke Recognition Algorithm for Forest Fires

An Improved YOLOv5s Fire Detection Model

Real-Time Fire Smoke Detection Method Combining a Self-Attention Mechanism and Radial Multi-Scale Feature Connection

Multi-sourced sensing and support vector machine classification for effective detection of fire hazard in early stage

FS-YOLO: Real-time Fire and Smoke Detection Based on Improved Object Detection Algorithms

YOLOv7-FIRE: A tiny-fire identification and detection method applied on UAV

Research on Smoke and Fire Target Detection with Tracking Algorithm Based on YOLOv8 and DeepSORT

Early Smoke Detection Based on Improved YOLO-PCA Network

Research on Detection Techniques of Early Forest Fire Based on Dynamic Characteristics of Smoke

YOLOv8 for Fire and Smoke Recognition Algorithm Integrated with the Convolutional Block Attention Module

Object Detection through Fires Using Violet Illumination Coupled with Deep Learning

Fire smoke detection based on target-awareness and depthwise convolutions

Video Smoke Detection Base on Dense Optical Flow and Convolutional Neural Network.