| 摘要: |
| 针对无人机个体识别效率低、计算量大的问题，提出了一种基于多维特征的FPV （First Person View）无人机个体识别方法。该方法构建了“外部特征快速筛选-信号多维特征深度解析”的双层架构，先基于信号外部特征提取、阈值判断，实现模拟图传信号快速检测，筛选出疑似信号，再利用ResNet模型对筛选后的疑似信号进行精细识别与匹配，提高识别的准确性和可靠性。实验结果表明，所提方法快速筛选层拒绝率超过85%，深度解析层对5.8 GHz频段FPV无人机信号识别平均准确率达到94%。 |
| 关键词: FPV无人机 个体识别 多维特征 ResNet模型 |
| DOI：10.20079/j.issn.1001-893x.250912001 |
|
| 基金项目: |
|
| FPV Drone Individual Recognition Based on Multi-dimensional Features |
| WANG Zijian,LI Xinhao,GU Yewei |
| (College of Electronic Engineering，National University of Defense Technology，Hefei 230037，China) |
| Abstract: |
| For the problems of low efficiency and large amount of computation in unmanned aerial vehicle (UAV) individual recognition，a first person view (FPV) drone individual recognition method based on multidimensional features is proposed.This method constructs a two-layer architecture of “rapid screening of external features-deep analysis of signal multidimensional features”.First，based on the external feature extraction and threshold judgment of signals，rapid detection of analog image signals is realized and suspected signals are screened out.Then，the residual network (ResNet) model is used to perform fine-grained identification and matching of the screened suspect signals，so as to improve the accuracy and reliability of identification.The experimental results show that the rejection rate of the fast screening layer of the proposed method is more than 85%，and the average recognition accuracy of the deep parsing layer for FPV drone signals in the 58 GHz band is 94%. |
| Key words: FPV drone individual identification multi-dimensional features ResNet model |
《电讯技术》编辑部 