摘要
针对复杂背景噪声下航空噪声识别困难的问题,提出一种基于动态风车卷积和残差注意力的航空噪声识别方法。该方法以Log-Mel频谱图为输入,通过动态风车卷积-残差注意力分支与Transformer分支协同分别提取局部时频特征与全局时序依赖关系,经自适应融合机制实现特征高效融合,完成对航空噪声的识别和分类。基于机场周边实地采集的航空噪声及城市环境噪声构建数据集,将所提方法与8种主流识别方法及3种代表性双分支网络进行对比实验,并通过消融实验验证各核心模块有效性。实验结果表明,该方法在准确率(99.52%)、精确率(99.78%)及F1分数(99.84%)上均优于对比方法,能有效感知噪声时变特性、抑制背景干扰,可为航空噪声实时监测与精准溯源提供可靠技术支撑。
To address the difficulty in aircraft noise recognition under complex background noise,a model method based on dynamic pinwheel convolution and residual attention is proposed.The method takes Log-Mel spectrograms as input,extracts local time-frequency features and global sequence features through the dynamic pinwheel convolution-residual attention mechanism branch and the Transformer branch respectively,and achieves efficient feature fusion via an adaptive fusion mechanism to accomplish the recognition and classification of aircraft noise.A dataset is constructed based on field-collected aircraft noise and other urban environmental noises around airports.Comparative experiments are conducted between the proposed method,eight mainstream methods,and three representative dual-branch networks,with ablation experiments performed to validate the effectiveness of the core modules.The results show that the proposed method outperforms the comparative models methods in accuracy(99.52%),precision(99.78%),and F1-score(99.84%).It can effectively perceive the time-varying characteristics of noise,suppress background interference,and provide reliable technical support for real-time monitoring and precise traceability of aircraft noise.
作者
郭二崇
原霞
王玉帅
管鲁阳
Guo Erchong;Yuan Xia;Wang Yushuai;Guan Luyang(School of Mechanical Engineering,North University of China,Shanxi Taiyuan,030051,China;School of Mechanical and Electrical Engineering,North University of China,Shanxi Taiyuan,030051,China;The Institute of Acoustics of the Chinese Academy of Sciences,Beijing,100190,China)
出处
《机械设计与制造工程》
2026年第4期79-85,共7页
Machine Design and Manufacturing Engineering
