Scholarlink

<p>In recent years, unmanned aerial vehicle (UAV) technology has been widely used, such as surveillance and monitoring, target strike, cargo transportation, etc. However, the abuse of drones will pose many severe threats to human society. Aiming at the characteristics of small size, small radar cross-section, and complex use environment of UAV, this paper designs a UAV classification mode based on a convolutional neural network (CNN). Firstly, the one-dimensional RCS sequence is converted into a two-dimensional image representation. In the experiment, four different conversion models are used. The conversion can enrich the amount of information in the original data, and the differences between different categories can be highlighted to obtain better recognition accuracy. Then it is imported into the improved CNN (UAVsNet) for recognition. Adding an attention module and modifying the classification module improves the recognition accuracy of the network. The experimental results show that UAVsNet has the recognition accuracy of 96.32% on the GASF dataset, which is 0.46–4.63% higher than other datasets. Through the ablation experiment, the influence of different changes on recognition accuracy was discussed.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

UAVsNet: A Convolutional Neural Network with Attention Mechanism for UAVs Classification

  • 期刊论文

  • 发表日期：

    2025年4月26日

查看全文
  • Jiangong Ni,
  • Zhigang Zhou

摘要

In recent years, unmanned aerial vehicle (UAV) technology has been widely used, such as surveillance and monitoring, target strike, cargo transportation, etc. However, the abuse of drones will pose many severe threats to human society. Aiming at the characteristics of small size, small radar cross-section, and complex use environment of UAV, this paper designs a UAV classification mode based on a convolutional neural network (CNN). Firstly, the one-dimensional RCS sequence is converted into a two-dimensional image representation. In the experiment, four different conversion models are used. The conversion can enrich the amount of information in the original data, and the differences between different categories can be highlighted to obtain better recognition accuracy. Then it is imported into the improved CNN (UAVsNet) for recognition. Adding an attention module and modifying the classification module improves the recognition accuracy of the network. The experimental results show that UAVsNet has the recognition accuracy of 96.32% on the GASF dataset, which is 0.46–4.63% higher than other datasets. Through the ablation experiment, the influence of different changes on recognition accuracy was discussed.