Abstract
The article considers the problem of low frequency signal classification, as sound or vibration pattern footprints may describe types of objects very well. In cases of a priori absence of object signal pattern information, the unsupervised learning methods based on clustering looks good enough for classification, and outperform neural net-based methods in case of limited power envelope. We have used big real-world sound and vibration data set to check several clustering methods (K-Means, OPTICS) for classification without any a priori data and have got good enough results. The article considers data set preparations including primary signal processing and the parameters to select appropriate clustering algorithms, which depends on input data shape. There are several examples of data classification, also cascaded methods for data set improvement are considered. Finally, we provide a good and practical guide for exploring low frequency signals using clustering methods, which can be used for real world observations and analysis for open space and inside buildings.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have