全光纤安防系统模式识别混合编程的实现

Implementation of pattern recognition in all fiber security monitoring system using mixed programming

介绍了宽域全光纤监控预警系统及光纤扰动信号模式识别算法.利用小波包分解对信号进行分解与重构,并提取各频段内信号分量能量谱作为特征向量;应用综合考虑观测数据和确定性先验知识的BVC-RBF神经网络实现信号分类,提高了网络的泛化能力.利用Matlab引擎、动态链接库(DLL)、组件对象模型(COM)3种混合编程方法将Matlab开发的算法程序集成到VC编写的用户操作平台中.使用文本文件存取网络数据,克服了Matlab无法编译神经网络工具箱函数的缺陷,并从操作复杂度、执行时间、内存消耗等方面对3种方式进行对比,最终选定动态链接库作为系统应用方案.

An all fiber security monitoring system and pattern recognition algorithm on fiber disturbance signal is presented.As a method of analyzing and reconstructing signal wavelet packet is adopted to extract feature vector.Classification is realized by BVC-RBF（boundary value constraints-radial basis function） neural network,which improves generalization by considering deterministic prior knowledge besides observational data.Pattern recognition algorithm is developed with Matlab and integrated into the application developed with VC through three mixed programming methods： Matlab engine,dynamic link library（DLL）,component objective model（COM）.Network data are read and written through text file to overcome the limitation that Matlab can not compile functions in neural network toolbox.Operation complexity,running time and memory consumption of these methods are compared and dynamic link library is the final choice.