LSTM: Anomaly Activity Type Classification Using Distributed Acoustic Sensing Based on MFCC Features

Abstract

The intergrity and connectivity of the fiber optical network are important in preserving the quality of services and internet realibility between providers and end users. However, these network are vulnerable to disruptions due to unitentional break and damage caused by physical disturbances such as construction activity. An accurate classification of anomaly activty at surounding area plays a crucial role in monitoring the buried fiber optical network from harm which can lead to denial of services. Distributed acoustic sensing (DAS) with combination of deep learning-based technique have potential in targeting this issue, by leveraging the unique pattern of vibration signal measured by the DAS to classify and identify anomaly activities. This work demonstrated utilization of dark fiber buried along the road until the server room, then connected to the DAS interrogator unit(IU). The vibration signals induced by construction hand tools, including hoe, shovel, and sledgehammer, which are used to mimic anamoly activity, are measured by DAS IU and underwent pre processing before exract the mel frequency cepstral coefficient (MFCC) features for long short- term memory (LSTM) model training. The average accuracy score using 25 MFCC resulted up to 87% indicating that the proposed method has
a great potential for anamoly activity monitoring for fiber break prevention.