光纤周界安防系统的高准确度事件识别方法

在双马赫曾德光纤周界安防系统中，如何准确高效地实现模式识别仍是一项有待解决的难题。对此提出了一种基于短时傅里叶变换（STFT）与奇异值分解的模式识别方法，实现了三种不同事件类型的准确识别。该方法包含三个步骤:首先，通过对干涉信号进行短时傅里叶变换得到时频信息，根据时频信息找到事件端点并进行滤波；其次，对时频信息进行奇异值分解，根据奇异值的物理意义，由不同行为的奇异值的特点定义特征向量；最后，用支持向量机的方法进行事件分类。为验证方法有效性，搭建了2 km长的围栏系统进行实验验证。进行了攀爬围栏、敲击光缆、晃动围栏三种不同入侵模式下共360组实验，每种入侵行为各120组得到了良好的识别结果（三种事件识别率均在90%以上），提高了系统的信号处理速度，有较高的实际应用价值。

A high-accuracy event discrimination method in optical fiber perimeter security system

In the dual Mach-Zehnder perimeter security system, it is difficult to achieve pattern recognition accurately and efficiently. To solve this problem, a pattern recognition method based on short-time Fourier transform(STFT) and singular value decomposition was proposed. In order to realize the accurate distinction of three different events, the method consisted of three steps. Firstly, the time-frequency information was obtained by the short-time Fourier transform of the interference signal, finding the event endpoint and filtering the signal according to the information; And then, the singular value of the time-frequency information was got. The characteristic vectors were defined by the characteristic of singular value, according to the physical meaning of the singular value. Finally, SVM was used to classify the events. In order to verify the effectiveness of the method, a fence system of 2 km long for experiment was built. 360 sets of interference signal data of three events (climbing the fence, knocking the cable, and waggling the fence) were collected, each event had 120 sets of data. The result of experiments demonstrates that the proposed scheme can discriminate three common invasive events with a high recognition rate. The recognition rates of the three events were all above 90%. The scheme also improve signal processing speed of the system, which has high application value.