| Perimeter security technology plays an important role in areas that are not easy to be monitored manually,such as regional key prevention and security of long-distance oil and gas pipelines.Dual Mach Zehnder interferometer（DMZI）optical fiber sensing system can realize uninterrupted long-distance sensing.It has simple structure and strong signal stability.It is widely used here.At present,the research on perimeter security technology mainly includes disturbance alarm,intrusion location and intrusion pattern recognition.For the identification of intrusion events,with the deepening of the complexity of the system environment and the increase of the types of events that need to be screened,the existing pattern recognition schemes can not meet the needs of practical applications,and there is still a lot of room for improvement in the indicators such as recognition accuracy and real-time recognition.This paper focuses on the event recognition of dmzi optical fiber perimeter security sensing system,focuses on optimizing the pattern recognition algorithm,and aims to improve the recognition efficiency and accuracy.The main research work includes:1.A two-dimensional signal preprocessing scheme based on short-term variance is designed.The solution uses the DMZI optical fiber perimeter security sensor system to collect 0.05 s short signals,performs variance calculation on each segment of the signal,sets thresholds for distinguishing signal intrusion information,and performs fast and effective filtering of event characteristic signal segments to achieve collection"Seamless"screening of signals.The two-dimensional scheme of multi-frame superimposition and interception is adopted.On the basis of retaining the original frequency information,the time scale information is expanded,and the event feature amount in the signal is increased.The deep neural network is used to extract the event feature to reduce the influence of human factors.And maintain a very high processing efficiency.2.Introduce the idea of two-dimensional image recognition and classification,and use the morphological characteristics and texture information of the two-dimensional signal to build a deep neural network model that integrates feature extraction and classification-convolutional long and short-term memory fully connected deep neural network（CLDNN）,optimize the network parameters,directly input the two-dimensional signal into the network for feature extraction and classification,and the network output is the determined event category,which successfully simplifies the traditional perimeter security event classification process.3.The five types of event signals（knocking,smashing,shaking,kicking,no intrusion）are collected on the experimental platform equipped with a low-speed100KS·s-1acquisition card,and the data is used for experiments.The results show that the The average recognition accuracy of the thesis scheme is above 97.9%,and the recognition time is stable within 0.01 s,which proves the superiority of the scheme of the thesis,provides a feasible and stable event recognition scheme for a low-cost and efficient system,and has a good application prospect. |
