基于目标检测的紫外光火情预警系统设计  

Design of UV fire warning system based on target detection

作　　者：董钢 李成勇 卢瑛 DONG Gang;LI Chengyong;LU Ying(School of Electronic Information,Chongqing Institute of Engineering,Chongqing 400056,China)

机构地区：[1]重庆工程学院电子信息学院,重庆400056

出　　处：《激光杂志》2025年第3期216-219,共4页Laser Journal

基　　金：重庆市自然科学基金(No.cstc2021jcyj-msxmX0921)资助;重庆市教委科学技术研究计划项目基金(No.KJQN202201901)资助。

摘　　要：近年来,频频发生的火灾,造成大量的人员伤亡及财产损失,因此,火情预警越来越关键。为了有效解决火情预警系统反应相对缓慢及检测范围有限,造成预警能力受限的问题。结合火情特点,使用Jetson Nano(人工智能边缘计算套件)为主控中心,利用LabelImg(标签)将两千多张图像数据进行标记,通过YOLOv5(目标检测算法)训练,得到火灾图像的训练权重;Jetson Nano读取紫外摄像头采集到的图像数据经训练权重推测是否发生火灾,若发生火灾则通过Jetson Nano的串口向微控制器发送数据,触发报警。本设计经过测试对火灾以及烟雾检测的正确率达97%以上,识别烟雾火灾的准确度高,火情预警迅速。相比于传统火灾检测技术,检测范围更广,精度更高,不易受环境影响。In recent years,frequent fire has caused a large number of casualties and property losses,so fire warning is becoming more and more critical.In order to effectively solve the fire early warning system response is relatively slow and the detection range is limited,resulting in limited early warning ability.Combined with the characteristics of fire,Jetson Nano(artificial intelligence edge computing suite) was used as the main control center,and LabelImg(label) was used to label more than two thousand image data.The training weight of fire image was obtained by YOLOv5(target detection algorithm) training.Jetson Nano reads the image data collected by the UV camera and calculates whether there is a fire by training the weight.If there is a fire,it sends data to the microcontroller through the serial port of Jetson Nano to trigger the alarm.This design has been tested for fire and smoke detection accuracy of more than 80%,smoke and fire identification accuracy is high,fire early warning is rapid.Compared with the traditional fire detection technology,the detection range is wider,the accuracy is higher,and it is not easy to be affected by the environment.

关 键 词：目标检测 图像识别 火灾检测 特征提取 

分 类 号：TN249[电子电信—物理电子学]