引用本文: |
杨雷1,郭恩泽2,魏国峰1,杨宁1,郭道省1.基于一维倒残差轻量级网络的无人机个体识别方法[J].陆军工程大学,2022,(6):65-72
[点击复制] |
|
|
|
本文已被:浏览 105次 下载 93次 |
基于一维倒残差轻量级网络的无人机个体识别方法 |
杨雷1,郭恩泽2,魏国峰1,杨宁1,郭道省1 |
(1.陆军工程大学 通信工程学院,江苏 南京 210007;2.陆军工程大学 通信士官学校,重庆 400035) |
摘要:在无人机的个体识别问题中,针对传统的识别方法存在网络模型参数量大和算法时效性差等问题,提出基于一维倒残差轻量级神经网络(1D-IRLNN)的无人机个体识别方法。首先提取反映无人机个体间差异的I/Q包络一维特征作为无人机的浅层特征;其次将深度可分离卷积与倒残差结构等设计思想和一维神经网络模型相结合,设计出跳跃级联的一维倒残差轻量级神经网络;最后利用网络模型提取一维I/Q包络数据中的深层特征,从而实现对无人机个体的快速准确识别。实验结果表明,1D-IRLNN模型的计算量分别是同等体量的深度模型1D-CNN与1D-ResNet的305%和238%,网络模型规模分别是深度模型1D-CNN与1D-ResNet的387%和297%,所提方法相较于其他方法,在保持较高识别准确率的同时具有更快的识别速度且占内存更小。 |
关键词: 一维轻量级网络 倒残差 无人机信号 个体识别 |
DOI:10.12018/j .issn.2097-0730.20220426001 |
投稿时间:2022-04-26 |
基金项目:国家自然科学基金(61971440) |
|
UAV Individual Identification Method Basedon One-Dimensional Inverted Residual Lightweight Neural Network |
YANG Lei1,GUO Enze2,WEI Guofeng1,YANG Ning1,GUO Daoxing1 |
(1.College of Communications Engineering,Army Engineering University of PLA,Nanjing 210007,China;
2.Communications NCO Academy,Army Engineering University of PLA,Chongqing 400035,China) |
Abstract: Regarding the problem of UAV individual recognition, the traditional identification method has some problems, such as large numbers of network model parameters and poor timeliness of algorithm. A method of UAV individual identification based on one-dimensional inverted residual lightweight neural network (1D-IRLNN) is proposed in this paper. Firstly, one-dimensional characteristics of the I/Q envelope reflecting individual differences of UAVs are extracted as the shallow features of UAVs. Then, a one-dimensional inverted residual lightweight neural network based on jumping cascade is designed for the first time by combining the design ideas of depthwise separable convolution and inverted residual structure with the one-dimensional neural network model. Finally, the network model is used to extract the deep features of the one-dimensional I/Q envelope data so as to realize the rapid and accurate identification of UAV individuals. The experimental results show that the computation amount of the designed 1D-IRLNN model is 305% and 238% of the depth models 1D-CNN and 1D-ResNet of the same volume respectively, and the size of the 1D-IRLNN model is 387% and 297% of the depth models 1D-CNN and 1D-ResNet respectively. Compared with other methods, the proposed method has faster identification speed and less memory overhead while maintaining high recognition accuracy. |
Key words: one-dimensional lightweight network inverted residual UAV signals individual identification |
|