Frequency of Interest-based Noise Attenuation Method to Improve Anomaly Detection Performance

2025-05-06 0 0 1.37MB 5 页 10玖币

侵权投诉

Frequency of Interest-based Noise Attenuation

Method to Improve Anomaly Detection

Performance

1st YeongHyeon Park

SK Planet Co., Ltd.

Seongnam, Rep. of Korea

yeonghyeon@sk.com

2nd Myung Jin Kim

SK Planet Co., Ltd.

Seongnam, Rep. of Korea

myungjin@sk.com

3rd Won Seok Park

SK Planet Co., Ltd.

Seongnam, Rep. of Korea

pwswonder@sk.com

Abstract—Accurately extracting driving events is the way

to maximize computational efﬁciency and anomaly detection

performance in the tire frictional nose-based anomaly detection

task. This study proposes a concise and highly useful method

for improving the precision of the event extraction that is

hindered by extra noise such as wind noise, which is difﬁcult

to characterize clearly due to its randomness. The core of the

proposed method is based on the identiﬁcation of the road

friction sound corresponding to the frequency of interest and

removing the opposite characteristics with several frequency

ﬁlters. Our method enables precision maximization of driving

event extraction while improving anomaly detection performance

by an average of 8.506%. Therefore, we conclude our method is

a practical solution suitable for road surface anomaly detection

purposes in outdoor edge computing environments.

Index Terms—Anomaly detection, Noise reduction, Road

safety, Sound event extraction

I. INTRODUCTION

Anomaly detection by utilizing audio has advantages such

as fewer blind spots compared to an image-based anomaly

detection system. However, even in audio, if noise other than

the noise of interest such as driving sound is varied and

excessive, it interferes to recognize the core information. In the

prior study, we performed an audio-based anomaly detection,

which is characterized in an outdoor environment [1]. The

challenging point of the above task is irregular extra noises

such as wind noise that hinder the identiﬁcation of the friction

noise between the tire and the road surface used to determine

the abnormality [2].

An effective way to perform surface anomaly detection

based on driving noise is to extract only the sound corre-

sponding to the driving event and use it as an input. At

the same time, as mentioned above, it will be helpful to

consider information other than driving noise, which is the

disturbance, as noise and attenuate them. The deep learning-

based denoising or noise-robust sound identiﬁcation method

has been recently proposed [3]–[6]. However, these methods

have limitations in that they have an amount of computation

This work has been submitted to the IEEE for possible publication.

no longer be accessible.

Fig. 1. Driving noise extraction minimizes unnecessary computation of the

anomaly detection model and helps to make decisions more accurately. Our

method includes a noise reduction method to extract driving noise based on

the FoI.

that can be overkill in an edge computing environment, or that

the training process takes a lot of cost for dataset preparing [7].

In this paper, we propose a noise reduction method to

improve anomaly detection performance. More speciﬁcally,

our goal is to extract only driving noise accurately and to

attenuate information that interferes with decision-making for

anomaly detection. Since our purpose is to perform road

anomaly detection over a wide area in the outdoor environment

while considering edge computing, it is necessary to develop

a cost-effective solution by avoiding high-cost methods such

as deep learning-based sound event detection [8]. Referring

that our method does not require a deep learning model, it

does not require the cost and effort for preparing the dataset

as prior study [9].

The method presented in this paper is based on the fact

that the preprocessing method in the frequency domain is

still effective [10]. Our band-pass ﬁltering-based noise re-

arXiv:2210.11068v3 [cs.LG] 2 Dec 2022

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

FrequencyofInterest-basedNoiseAttenuationMethodtoImproveAnomalyDetectionPerformance1stYeongHyeonParkSKPlanetCo.,Ltd.Seongnam,Rep.ofKoreayeonghyeon@sk.com2ndMyungJinKimSKPlanetCo.,Ltd.Seongnam,Rep.ofKoreamyungjin@sk.com3rdWonSeokParkSKPlanetCo.,Ltd.Seongnam,Rep.ofKoreapwswonder@sk.comAbstractAccurat...

展开>> 收起<<

Frequency of Interest-based Noise Attenuation Method to Improve Anomaly Detection Performance.pdf

共5页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Frequency of Interest-based Noise Attenuation Method to Improve Anomaly Detection Performance

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: