基于卷积神经网络的早期火苗检测研究  

Early Fire Detection Based on Convolutional Neural Networks

作　　者：王丽 沈晓波 束仁义 WANG Li;SHEN Xiaobo;SHU Renyi(School of Electronic Engineering,Huainan Normal University,Huainan 232038,China)

机构地区：[1]淮南师范学院电子工程学院,安徽淮南232038

出　　处：《长春师范大学学报》2024年第6期64-68,89,共6页Journal of Changchun Normal University

基　　金：淮南师范学院校级重点项目“基于卷积神经网络的初期火灾检测算法研究”(2022XJZD020);安徽省省级高校自然科学重点研究项目“基于超宽带技术与智能优化算法的室内定位研究”(2023AH051550)。

摘　　要：为了快速检测出早期火苗,提高火灾的预警功能,本文研究基于YOLOv5s算法的早期火灾视频的火苗识别,并将YOLOv5s算法与YOLOv4-tiny算法、YOLOv3-tiny算法进行了比较。研究结果表明,YOLOv5s的定位损失比YOLOv4-tiny和YOLOv3-tiny分别降低了0.0225和0.0205,YOLOv5s的精确度分别提高了0.1361和0.0507。在视频检测中YOLOv5s最早检测到火苗,且检测的准确率比YOLOv4-tiny和YOLOv3-tiny分别提高了0.22和0.21,在早期小火苗检测中YOLOv5s算法的性能更优。In order to quickly detect the early fire and improve the fire warning function,this paper studies the early fire video flame recognition based on the YOLOv5s algorithm,and compares it with the YOLOv4-tiny and YOLOv3-tiny algorithms.The experimental results show that the box loss of YOLOv5s has reduced 0.0225 and 0.0205 compared with YOLOv4-tiny and YOLOv3-tiny.YOLOv5s’s precision has improved 0.1361 and 0.0507.In the video detection,YOLOv5s is the earliest to detect the flame,and the detection accuracy has improved 0.22 and 0.21 compared with YOLOv4-tiny and YOLOv3-tiny.In the early small flame detection,the performance of YOLOv5s algorithm is better.

关 键 词：早期火苗 YOLOv5s模型 视频检测 准确率 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]