期刊文献+

基于机器视觉的景区火灾检测方法

A Machine Vision Based Fire Detection Method for Scenic Areas
下载PDF
导出
摘要 随着人类活动的增加和全球气候变化,火灾事故在自然景区内发生的频率越来越高。为了及时发现和处理火灾,提出了一种基于机器视觉的景区火灾检测方法。通过空中无人机和地面视频监控设施收集景区的实时图像,采用卷积神经网络(Convolutional Neural Network,CNN)的深度学习模型对图像进行分析,用以检测火灾的发生。研究采用了轻量级的神经网络SqueezeNet、ShuffleNet、MobileNet_v2以及ResNet-50进行火灾识别。为了模拟九寨沟景区实地情况,选择复杂度较低的检测算法,还进行了跨数据集识别评估,并将其与复杂度较高的ResNet-50进行了比较,最终得出通过基于ResNet-18进行图像语义分割的方法,识别结果的分类准确率达到96%,验证了方法具有更好的鲁棒性。 With the increase of human activities and global climate change,the frequency of fire accidents in natural scenic areas is higher than before.In order to detect and handle fires in time,a machine vision based fire detection method is proposed for scenic areas.Firstly,the real time images of the scenic areas are collected through aerial drones and ground video monitoring facilities.Then,the deep learning model of the Convolutional Neural Network(CNN)is used to analyze the images to detect the occurrence of fire.The lightweight neural networks,such as SqueezeNet,ShuffleNet,MobileNet_v2 and ResNet-50,are used for fire identification.In order to simulate the actual situation of Jiuzhaigou Valley Scenic and Historic Interest Area and select the detection algorithm with lower complexity,the identification evaluation across the dataset is also conducted,and it is compared with ResNet-50 of higher complexity.Finally,through the method of image semantic segmentation based on ResNet-18,the classification accuracy of the identification results reached 96%,showing that the method has better robustness.
作者 汤云超 周翰 邢秀青 刘占勇 TANG Yunchao;ZHOU Han;XING Xiuqing;LIU Zhanyong(China Communications System Co.,Ltd.,Beijing 100089,China;School of Urban Geology and Engineering,Hebei GEO University,Shijiazhuang 050031,China;CRRC Tangshan Co.,Ltd.,Tangshan 064099,China)
出处 《计算机与网络》 2024年第2期163-170,共8页 Computer & Network
关键词 火灾检测 机器视觉 深度学习 卷积神经网络 fire detection machine vision deep learning CNN
  • 相关文献

参考文献4

二级参考文献24

共引文献34

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部