Investigate Higher Accuracy and Speed of Anomaly Detection for Edge AI
-
- KITANO Yuya
- Osaka Institute of Technology
-
- NAKANISHI Chikako
- Osaka Institute of Technology
Bibliographic Information
- Other Title
-
- エッジAI向けの異常検知の高精度,高速化の検討
Abstract
In this paper, we propose a method to improve processing speed while maintaining anomaly detection accuracy. Recently, the use of edge AI for anomaly detection has been attracting attention. Anomaly detection is computationally intensive and difficult to process on edge terminals with low processing performance. Therefore, we propose a method to speed up the processing by reducing the number of features. The results showed that compared to PatchCore, carpet and wood were more accurate, leather and tile were close, and grid was less acculate The processing time was also reduced to 1/3 or less.
Journal
-
- 電子情報通信学会論文誌D 情報・システム
-
電子情報通信学会論文誌D 情報・システム J107-D (4), 188-195, 2024-04-01
The Institute of Electronics, Information and Communication Engineers
- Tweet
Keywords
Details 詳細情報について
-
- CRID
- 1390581070826595840
-
- ISSN
- 18810225
- 18804535
-
- Text Lang
- ja
-
- Data Source
-
- JaLC
-
- Abstract License Flag
- Disallowed