Title:
APPARATUS AND METHOD FOR SOUND ANOMALY DETECTION BASED ON NON-COMPRESSION CONVOLUTIONAL NEURAL NETWORK
Document Type and Number:
WIPO Patent Application WO/2022/191361
Kind Code:
A1
Abstract:
A method for sound anomaly detection, according to the present invention, comprises steps in which: an audio unit acquires noise from an object to be inspected; a data processing unit generates, from the noise, an input value, which is a feature vector matrix including a plurality of feature vectors; a detection unit generates, through a detection network, which is a deep neural network trained with respect to the input value, a restoration value simulating the input value; the detection unit determines whether a restoration error indicating the difference between the input value and the restoration value is greater than a pre-calculated reference value; and the detection unit determines that the object being inspected is anomalous, if the determination result shows that the restoration error is greater than or equal to the reference value.
Inventors:
PARK YEONG HYEON (KR)
JUNG JONG HEE (KR)
JUNG JONG HEE (KR)
Application Number:
PCT/KR2021/007661
Publication Date:
September 15, 2022
Filing Date:
June 18, 2021
Export Citation:
Assignee:
SK PLANET CO LTD (KR)
International Classes:
G01H15/00; G01D1/14; G01H17/00; G06F15/16; G06F17/10; G06N3/02; G06N3/08
Foreign References:
| KR20170119353A | 2017-10-27 | |||
| US8484022B1 | 2013-07-09 | |||
| KR20200109904A | 2020-09-23 | |||
| KR102027389B1 | 2019-10-01 |
Other References:
SON JUNE-HYOUNG, JANG JUN-GUN, CHOI BONGWAN, KIM KYEONGTAEK: "Detection of Abnormal Vessel Trajectories with Convolutional Autoencoder", JOURNAL OF SOCIETY OF KOREA INDUSTRIAL AND SYSTEMS ENGINEERING, vol. 43, no. 4, 30 December 2020 (2020-12-30), pages 190 - 197, XP055966827, ISSN: 2005-0461, DOI: 10.11627/jkise.2020.43.4.190
Attorney, Agent or Firm:
PARK, Chonghan (KR)
Download PDF:
Previous Patent: BATTERY RACK ACCOMMODATING BATTERY PACKS, AND BATTERY RACK MODULE
Next Patent: PHOTOCATALYTIC FILTER MODULE
Next Patent: PHOTOCATALYTIC FILTER MODULE