基于嵌入式人工智能设备的流星光学监测系统

流星光学监测网是定位陨石和观测火流星的基础科研设施. 流星光学监测系统利用光学相机高速采集天空图像, 使用嵌入式系统实时处理数据, 能够快速识别流星并获取流星位置和陨石落点信息, 是构成流星监测网的关键仪器. 为提高流星光学监测系统获取信息的实时性及准确性, 提出了一种基于嵌入式人工智能设备的流星光学监测系统. 该系统由软件及硬件部分组成: 硬件部分包括观测设备(商用高空抛物摄像头)以及数据处理设备(嵌入式人工智能设备); 软件部分运行于数据处理设备内, 主要包括控制界面模块、流星监测模块、数据管理模块. 实际工作时, 摄像头采集天空视频信息, 流星监测模块从视频流中实时监测流星并存储包含流星视频的数据, 数据管理模块将流星位置信息实时传回数据中心用于预警. 观测结束后, 将原始观测数据同步至数据中心用于后续科学研究. 在整个系统中, 流星监测模块决定了整个监测系统的实时性及准确性. 该系统采用嵌入式人工智能设备与人工智能算法结合的方法构建流星监测模块. 通过使用实测数据对搭载监测模块性能进行测试, 结果表明: 流星监测模块能够达到0.28%的低误检率以及100%的召回率, 且数据处理速度达到了Mobilenetv2的8倍. 进一步将包含监测模块的整个流星光学监测系统部署于太原理工大学-张壁古堡远程天文台, 通过实测表明流星光学监测系统实用中能达到100%的召回率和较低的误检率.

Optical Meteor Monitoring System Based on Embedded Artificial Intelligence Equipment

Optical meteor monitoring networks are composed by several optical meteor monitoring systems placed in different locations, which mainly contain a detector and a data processing modular. The optical meteor monitoring system could locate meteorites, obtain number of meteors, detect and send alerts of fire meteors. Real-time detection of meteors with high efficiency is important both for scientific research and civil affairs. This paper proposes a novel optical meteor monitoring system based on artificial intelligence device. The optical meteor monitoring system includes a hardware part and a software part. The hardware part is composed by a commercial camera and an artificial intelligence device. The software part is running in the artificial intelligence device, which includes the control module, the meteor monitoring module, and the data management module. The camera would capture videos in real time and send observation data to the meteor monitoring module. The meteor monitoring module could process observation data in real time to obtain candidates of meteors and store data of these candidates. At last, the data management module would send all detection data to the data center for further process. This paper uses real observation data to test the performance of the meteor monitoring module and results show that this algorithm can achieve a false positive rate of 0.28% and a recall rate of 100% and the speed of the data processing part is 8 times faster than the Mobilenetv2. This system has been further deployed in Taiyuan University of Technology and the remote observatory of Zhangbi Castle. Results show that the optical meteor monitoring system could achieve a recall rate of 100% and a relatively low false detection rate.