Fiber Vibration Signal Classification Algorithm Based on Transfer Learning Combined with Feature Combination

Abstract

Optical fiber sensors have important applications in the field of sensing. It is very important to quickly and effectively classify and identify the vibration signals collected by optical fiber sensors. In this paper, a neural network transfer learning method combined with combined feature based on optical fiber sensing signal classification is proposed. The method identifies intrusion signals by traditional manual features combined with deep features. First, time domain and frequency domain manual features extraction are performed on the intrusion signals. At the same time, the signal is decomposed by wavelet packet and the energy of each node is calculated. Then the feature vectors are normalized and fed into the improved CNN to generate deep features under the condition of small samples. After that these deep features are combined with traditional features such as time domain features and frequency domain features to identify the intrusion vibration signal. This paper uses three different intrusion events for experimental verification. The method of combining deep features with manual features is applied to the classification experiments of these three intrusion signals, and the feature extraction methods using only time domain features, frequency domain features, wavelet packet features and single deep features are applied to the same example experiment. The experimental results show that the algorithm can identify the vibration signal more effectively, achieve an accuracy more than 98.2%, and comparative experiment further verify the feasibility and effectiveness of the algorithm.