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
王丽 1沈晓波 1束仁义1
作者信息
- 1. 淮南师范学院电子工程学院,安徽淮南 232038
- 折叠
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 0.022 5和0.020 5,YOLOv5s的精确度分别提高了 0.136 1和0.050 7.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.
Abstract
In order to quickly detect the early fire and improve the fire warning function,this paper studies the early fire video flame rec-ognition based on the YOLOv5s algorithm,and compares it with the YOLOv4-tiny and YOLOv3-tiny algorithms.The experimental re-sults show that the box loss of YOLOv5s has reduced 0.022 5 and 0.020 5 compared with YOLOv4-tiny and YOLOv3-tiny.YOLOv5s's precision has improved 0.136 1 and 0.050 7.In the video detection,YOLOv5s is the earliest to detect the flame,and the detection accu-racy 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Key words
early flame/YOLOv5s model/video detection/accuracy rate引用本文复制引用
基金项目
淮南师范学院校级重点项目(2022XJZD020)
安徽省省级高校自然科学重点研究项目(2023AH051550)
出版年
2024
NETL
NSTL
万方数据