Human intrusion detection with distributed fiber optic data based on Squeeze-Excitation and hierarchical connection enhancement network

Abstract

Intrusion event detection technology plays an important role in home security, however, traditional intrusion detection methods have problems such as having a blind field of vision and not being able to make good use of distributed spatial information, which affects the accuracy of detection. This paper suggests a method for tackling this issue through employing a distributed fiber optic acoustic sensing technique, which makes full use of distributed spatial information to identify important features in fiber optic data. Then, on this basis, a Squeeze-Excitation and Hierarchical Connection Enhancement Network is proposed, whose key idea is to utilize different branches in the network to deal with feature maps of different resolutions or scales, to enhance the important features, and attenuate the unimportant ones, to make the features more directional and achieve higher accuracy. The method’s efficacy is ultimately assessed through analysis of intrusion events in the gathered fiber optic dataset, which achieves an intrusion detection accuracy of 99.2%.