摘要
在工业生产中,利用音频数据来对工厂机器的运行状态进行检测是一种行之有效的解决方法。同种机器的音频由于物理参数不一致或者录制条件的不同导致数据域之间存在差异性。因此,找到数据域之间的跨域信息对解决异常声音检测中数据域偏移问题十分关键。针对上述问题,本文提出一种基于对比学习的异常声音检测方法。该方法首先通过对比学习,让模型学习到源域音频和目标域音频之间的共性。之后,再通过设计合适的辅助分类任务来对模型进行微调,让模型具有域迁移能力的同时也具备良好的异常声音检测能力。另外,引入了频域注意力机制来更好地表征音频频域信息,提升模型的异常声音检测效果。在DCASE2022 Task 2数据集上的实验结果表明,本文的方法在5类机器上得到了良好的实验效果,特别是在目标域上,提升效果更为明显。
In industrial production,using audio data to monitor the operational status of factory machinery is a proven and effective solution.However,due to variations in physical parameters or recording conditions among machines of the same type,discrepancies often arise between data domains.Therefore,identifying cross-domain information is crucial for addressing the domain generalization challenge in anomalous sound detection.To solve this problem,we propose an anomalous sound detection method based on contrastive learning.Initially,the model learns common features between source and target domain audio through contrastive learning.Subsequently,we design appropriate auxiliary classification tasks to fine-tune the model,enabling it to possess both domain adaptation capabilities and strong anomaly detection performance.Additionally,we introduce a frequency-domain attention mechanism to better represent audio frequency information,further enhancing the model's anomalous sound detection ability.Experimental results on the DCASE 2022 Task 2 dataset demonstrate that our method achieves promising performance across five machine types,with particularly notable improvements in the target domain.
作者
肖遥
彭焘
冯时
朱晨阳
李圣辰
邵曦
XIAO Yao;PENG Tao;FENG Shi;ZHU Chenyang;LI Shengchen;SHAO Xi(School of Communications and Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing,Jiangsu 210003,China;School of Advanced Technology,Xi'an Jiaotong-Liverpool University,Suzhou,Jiangsu 215123,China)
出处
《复旦学报(自然科学版)》
北大核心
2025年第5期579-590,共12页
Journal of Fudan University:Natural Science
基金
国家自然科学基金(61936005,62001038)
江苏省科技重大专项(BG2024027)
姑苏领军人才青年人才创新项目(ZXL2022472)。
关键词
异常声音检测
自监督学习
对比学习
域泛化
注意力机制
anomalous sound detection
self-supervised learning
contrastive learning
domain generalization
attention mechanism
