Deep Learning Approach For Audio Signal Classification And Its Application In Fiber Optic Sensor Security System

The perimeter intrusion detection system (PIDS), as a new type of security system, is getting more and more people's attention. And the positioning and early warning algorithms for intrusion signals have always been the focus of people's research. Signals collected by fiber optic sensors can be regarded as an audio signals. However, traditional audio signal recognition algorithms have poor classification effects due to the excessive sensitivity of fiber optic sensors. In this paper, a fully residual convolution neural network with long short-term memory (LSTM) is proposed to solve the signal identification problem. Three different audio feature spectrograms are used as parallel inputs to improve the network stability. Experiments and comparisons are carried out among our network and the support vector machines (SVM), back propagation neural networks (BPNN), simple DNN network, which prove that our system has higher recognition accuracy and strong resistance to environmental interference.