主管：中华人民共和国应急管理部
主办：应急管理部天津消防研究所
ISSN 1009-0029  CN 12-1311/TU
消防科学与技术

消防科学与技术 ›› 2020, Vol. 39 ›› Issue (12): 1713-1717.

• • 上一篇    下一篇

基于O-YOLOv3 的无人机森林火灾识别研究

尹建平1,3，曾美琳2,3，徐文磊3，熊强强3，李柯3   

  1. 1. 豫章师范学院，江西南昌330103；2. 江西工业贸易职业技术学院，江西南昌330038；3. 南昌蛋讯电子科技有限公司，江西南昌330029
  • 出版日期:2020-12-15 发布日期:2020-12-15
  • 作者简介:尹建平（1979-），女，豫章师范学院高级工程师，主要从事计算机视觉及多目标数据融合技术研究，江西省南昌市红谷滩新区嘉言路699 号，330103。
  • 基金资助:
    2018 年江西省高等学校教学改革研究项目（JXJG-18-25-3）；2019 年江西省教育科学规划“十三五”课题（19YB251）

Research on UAV forest fire recognition based on O-YOLOv3

YIN Jian-ping1,3, ZENG Mei-lin2,3, XU Wen-lei3, XIONG Qiang-qiang1,3, LI Ke1,3   

  1. 1. Yuzhang Normal University, Jiangxi Nanchang 330103，China; 2. Jiangxi Vocational College of Industry and Trade, Jiangxi Nanchang 330038，China; 3. Nanchang Eggun Electronic Technology Co., Ltd., Jiangxi Nanchang 330029，China
  • Online:2020-12-15 Published:2020-12-15

摘要：

为解决传统森林火灾检测误报率高、响应速度慢等问题，提出了以无人机作为探测平台，地面站作为火灾识别系统，实现森林火灾的自动探测、识别和定位。开发了六旋翼无人机平台，通过所搭载的红外摄像机和机载计算机获取森林火灾现场图像并实时传回地面。利用地面站对所接收到的火灾图像进行处理，实现对森林火场的在线监测。在森林火灾识别算法方面，提出了O_YOLOv3 算法，采用Darknet 框架进行网络训练，使用K_means 方法自动生成锚点，有效提高火灾识别精度与响应速度。将O_YOLOv3 算法与其他几种算法进行对比实验验证本文算法的有效性。实验结果表明：O_YOLOv3 火灾识别算法能够快速、精准识别森林火灾；所研制的基于O_YOLOv3 的无人机森林火灾探测系统能够用于实际森林火灾探测。

关键词: 无人机, 森林火灾, O-YOLOv3, 火灾识别

Abstract:

To solve the problems of high false alarm rate and slow response speed in traditional forest fire detection, it is proposed to use UAV as the detection platform and ground station as the fire identification system to realize the automatic detection, identification and location of forest fires. First, a six- rotor UAV platform was developed to obtain forest fire scene images through the mounted infrared camera and onboard computer and transmit them back to the ground in real time. Secondly, the ground station was used to process the received fire images to achieve forest fire online monitoring of fire sites. In terms of forest fire recognition algorithm, the O- YOLOv3 algorithm is proposed, the Darknet framework is used for network training, and the K-means method is used to automatically generate anchor points, which effectively improves the accuracy and response speed of fire recognition. Finally, the O-YOLOv3 algorithm is compared with several other algorithms to verify the effectiveness of this algorithm. Experimental results show that: OYOLOv3 fire identification algorithm can quickly and accurately identify forest fires.The developed UAV forest fire detection system based on O-YOLOv3 can be used for actual forest fire detection.  

Key words: UAV, forest fire, O-YOLOv3, fire recognition