Research On Pattern Recognition Method For Fiber Vibration Sensing System Based On Michelson Interferometer

With the advent of the information age, sensing technology is attracting more and more attention. Compared with the traditional electric sensors, optical fiber sensors have advantages of high measuring precision, long sensing distance, strong anti electromagnetic interference ability, which make them have a more wide application prospect. Phase modulation optical fiber sensor is a hot spot in the field of optical fiber sensing, and it is difficult to correctly identify the output sensing signals. In order to improve the recognition accuracy of sensing signals, this paper studies the pattern recognition method for the Michelson interferometric fiber optical vibration sensing system. The main research contents include the following aspects:1. The sensing principle, classification and current research status at home and abroad of the optical fiber sensing technology are introduced briefly, the key problems of information processing technology in phase modulation optical fiber sensors are thoroughly analyzed.2. The sensing principle of the fiber vibration sensing system based on Michelson Interfereometer is thoroughly studied, and the relationship between the output signals of the sensing system and the external vibration signals is derived. Two key factors which affect the stability of the output sensing signal are analyzed and the corresponding solutions are given. On the basis of theoretical analysis, a specific hardware sensing system is built, and vibration sensing experiment is carried out, different kinds of external vibration sensing signals are acquired.3. The wavelet packet transform and hilbert-huang transform are used to extract the wavelet packet energy features and marginal spectrum features of the sensing signals respectively. BP neural network is built by Matlab simulation software, the extracted features are fed into the neural network, and the simulation results are analyzed.4. A two level signal identification method is proposed. First, the threshold-crossing rate algorithm is used as the primary identification to recognize the vibration signals. Second, in order to identify the specific type of sensing external vibration signals, the algorithm combined sparse autoencoder with softmax classifier is applied for high dimensional feature extraction and discrimination. The sparse autoencoder algorithm can realize unsupervised feature learning, it avoids the huge workload of the artificial feature design and the loss of the signal characteristic. The experimental results show that this method can effectively distinguish the role of a variety of transient disturbance events, and the classification accuracy of the proposed algorithm is up to 96.76%, which is consistent with the expected target, and the method improves the practicability of the optical fiber vibration sensing system based on Michelson Interfereometer.