一种基于轻量型神经网络的火情早期预警方法  

Fire Fast Warning Method Based on Lightweight Neural Network

作　　者：宋世淼 顾非凡 葛家尚 杨杰[1] 宋述歆 SONG Shimiao;GU Feifan;GE Jiashang;YANG Jie;SONG Shuxin(College of Mechanical and Electrical Engineering,Qingdao University,Qingdao 266071,China;Jinan Zhangqiu District Agricultural Development Service Center,Jinan 250200,China)

机构地区：[1]青岛大学机电工程学院,山东青岛266071 [2]济南市章丘区农业发展服务中心,山东济南250200

出　　处：《青岛大学学报（工程技术版）》2024年第3期7-12,共6页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：山东省自然科学基金资助项目(ZR2021MF025)。

摘　　要：为了提高火灾检测效率,基于模型压缩思想提出了一种火情早期实时检测模型FRDnet (Fire rapid detection network),利用低值滤波器修剪策略优化ShuffleNetV2网络,优化后的网络参数量比原网络减少了50%,提高了运算效率。针对检测结果假性的问题,提出了基于阈值判定的预警逻辑,提高了预警的鲁棒性。在公开数据集上的实验结果表明,模型的检测精度达到了95%,检测效率达到了44 fps;预警逻辑使模型能够在火情发生4.5 s内发出报警信号,表明模型在火灾发生早期能够快速准确预警。In order to improve the efficiency of fire detection,a fire rapid detection network(FRDnet)was proposed based on the model compression idea.The low-value filter pruning strategy was used to optimize the ShuffleNetV2 network.The optimized network parameters were reduced by 50%compared with the original network,improving computing efficiency.Aiming at the problem of false positive detection results,a warning logic based on threshold judgment was proposed to improve the robustness of early warning.Experimental results on public data sets show that the detection accuracy reaches 95%and the detection efficiency reaches 44 fps.The early warning logic enables the model to issue an alarm signal within 4.5 seconds once a fire occurs,indicating that the model can provide rapid and accurate warning in the early stage of fire.

关 键 词：ShuffleNetV2 滤波器修剪 检测效率 火情早期预警 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]