现代电子技术2025，Vol.48Issue(11)：61-68,8.DOI:10.16652/j.issn.1004-373x.2025.11.010
基于音频的多特征融合低慢小目标探测研究
Audio-based multi-feature fusion method for detection of low-slow-small targets
王拓 1张成 1祁万龙 1苏照兵 1齐志强 1隋振雨1
作者信息
- 1. 西北机电工程研究所,陕西 咸阳 712099
- 折叠
摘要
Abstract
UAVs are gradually becoming an important means of reconnaissance and strike in modern warfare,and it is becoming more and more important for the detection and identification of the low-slow-small(LSS)targets.As an important perceptual modality,audio has unique advantages in target detection,especially when radio spectrum resources are limited or image information is unavailable.In view of this,an audio-based multi-feature fusion method is proposed for the detection of the LSS targets.A two-branch structure is adopted in the model.For the first branch,a novel LLSIncepNeXt module is designed by extracting features on the Mel spectrogram of the audio data,and the designed module extracts the information in both time and frequency dimensions by a parallel convolution kernel.For the other branch,the MFCC(Mel-scale frequency cepstral coefficient)features of the audio is input directly into the bi-directional GRU(gated recurrent unit)to extract the temporal features.Subsequently,the extracted features from the two branches are fused.And the focus features are strengthened by the multi-attention mechanism to distinguish the contribution degree of different features.The results on the Drone Detection dataset and the UrbanSound8K dataset show that the proposed multi-feature fusion network has a great improvement over the methods employing a single feature,and achieves better results in classifying UAV audio in comparison with the other methods.关键词
低慢小目标/音频感知/多特征融合/目标探测/深度学习/GRU/InceptionNeXt/MFCCKey words
LSS target/audio perception/multi-feature fusion/target detection/deep learning/GRU/InceptionNeXt/MFCC分类
信息技术与安全科学引用本文复制引用
王拓,张成,祁万龙,苏照兵,齐志强,隋振雨..基于音频的多特征融合低慢小目标探测研究[J].现代电子技术,2025,48(11):61-68,8.
