Acoustic Anomaly Detection Based on Band Energy of Discrete Wavelet Transform

It is important to detect and monitor abnormal conditions of machines in industrial environments. Acoustic signal-based anomaly detection technology has the advantage of performing fault detection at a low cost compared to conventional image/video-based or other sensor-based anomaly detection technologies. However, utilizing a large amount of acoustic signals collected from sensors requires a lot of computing resources for intensive signal processing and learning. Therefore, it is necessary to consider an acoustic signal-based anomaly detection system that utilizes computing resources efficiently. In this paper, we propose a method for extracting the band energy of acoustic signals using discrete wavelet transform and design a lightweight anomaly detection model. Experiments show that the band energy based on discrete wavelet transform can effectively compress the features of acoustic signals, reduce the data preprocessing time, and enable the construction of a lightweight anomaly detection model.