基于深度学习的林火烟雾识别系统设计  

Design of a Deep Learning-Based Smoking Recognition System for Forest Fires

作　　者：李梓铭 石振威 徐海文 龙骏 朱勇兵 周国雄[2] LI Ziming;SHI Zhenwei;XU Haiwen;LONG Jun;ZHU Yongbing;ZHOU Guoxiong(Hunan Forest and Grassland Fire Monitoring,Dispatchand Evaluation Center,Changsha 431007,Hunan,China;Central South Forestry University of Science and Technology,Changsha 431018,Hunan,China)

机构地区：[1]湖南森林草原防火监测调度评估中心,湖南长沙431007 [2]中南林业科技大学,湖南长沙431018

出　　处：《中南林业调查规划》2023年第3期36-40,共5页Central South Forest Inventory and Planning

摘　　要：通过构建林火烟雾数据集,选取Inception V3模型,基于参数的迁移学习方法构建林火烟雾图像训练模型,经过训练测试后得到模型在测试数据集识别率达到92%,最后运用Python语言编程将训练后保存下来的模型文件应用到林火烟雾视频上,对视频进行逐帧预测,并将预测结果可视化注释在视频上,达到视频监控的目的。将网络训练和视频监控分开,通过软件编程实现应用模型文件监测林火烟雾视频,较以往直接通过视频训练和预测的方法更方便快捷,实际可操作性更强,具有广泛的应用前景。By constructing a forest fire smoke dataset,selecting the Inception V3 model,a forest fire smoke image training model based on the parameter migration learning method was obtained.A model that achieved a recognition rate of 92%in the test dataset.Finally,Python language programming is used to implement the application of the saved model file after training to the forest fire smoke video,and to predict the video frame by frame,and to annotate the prediction results visually on the video in order to achieve the purpose of video monitoring.It is more convenient and quicker than the previous method of training and prediction directly through video to separate the network training and video monitoring,and realize the application of model files to monitor the forest fire smoke video through software programming,which is more practicable and has a wide application prospect.

关 键 词：林火烟雾识别系统 深度学习 卷积神经网络 Inception V3 PYTHON 

分 类 号：S757[农业科学—森林经理学]