摘要
林业防火监测预警系统必须确保具备极高的报警率与极低的误报率,但当前部分系统因受光照条件变化及山林云雾遮挡的影响,导致图像特征信息变得模糊,严重影响预警结果的准确性。为此,通过无人机航测获取烟雾图像并进行光照自适应调节以及云雾去除,提高图像的清晰度。从烟雾图像中提取颜色特征、纹理特征。使用贝叶斯网络训练分类器,用于识别烟雾图像和非烟雾图像并触发预警机制。结果表明,该方法的误警率、漏警率明显低于2种传统方法基于改进YOLOv5s方法和基于知识图谱的方法,具有更高的准确性,能够更有效地捕捉到火灾早期的烟雾信号。
Forest fire monitoring and early warning systems must ensure a very high alarm rate and a very low false alarm rate.However,some existing systems are affected by changes in lighting conditions and the obstruction of clouds and fog in mountainous forests,resulting in blurred image feature information and seriously reducing the accuracy of early warning results.In this study,unmanned aerial vehicle(UAV)aerial surveys were used to obtain smoke images,followed by adaptive illumination adjustment and cloud–fog removal to enhance image clarity.Colour and texture features were extracted from smoke images,and a Bayesian network was employed to train a classifier for distinguishing smoke images from non-smoke images and to trigger the early warning mechanism.The results showed that the false alarm rate and missed alarm rate of the proposed method were significantly lower than those of the two traditional methods—namely,the improved YOLOv5s-based method and the conventional image-based method—demonstrating that the proposed method achieves higher accuracy and can detect early-stage forest fire smoke more effectively.
作者
郭永奇
GUO Yongqi(Lvliangshan State-owned Forest Management Bureau,Lvliang,Shanxi 041200)
出处
《林业勘查设计》
2025年第6期68-72,共5页
Forest Investigation Design
关键词
无人机航测
林业早期火灾
烟雾监测
特征提取
预警
UAV aerial survey
Early forest fire
Smoke monitoring
Feature extraction
Early warning
