Meteor Head Echo Detection via a Convolutional Neural Network Trained on Synthetic Radar Data
2024 United States National Committee of URSI National Radio Science Meeting (USNC-URSI NRSM)(2024)
摘要
In October 2019, eight hours of concurrent radar observations were performed at three facilities, and thousands of meteor head echoes were observed. This motivates training a convolutional neural network (CNN) to identify them. Since each experiment has unique parameters such as frequency and pulse code, pre-existing training data is unavailable, so a synthetic model is developed to simulate head echo observations at any facility. Real instances of radar clutter and noise are combined with synthetic head echoes to ensure differentiation between these phenomena. The CNNs are tested on observed data, and demonstrate overall accuracy greater than 97% and over 70% sensitivity to head echoes at each facility. Therefore, this technique is capable of identifying a comprehensive population of head echoes at any radar facility that observes them.
更多查看译文
关键词
Convolutional Neural Network,Synthetic Radar Data,Ground Clutter,Convolutional Layers,Fully-connected Layer,Convolutional Neural Network Architecture,Doppler Shift,Strong Echo
AI 理解论文
溯源树
样例
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要