Optical fiber vibration signals are nonlinear.Conventional nonlinear vibration signals recognition methods usually require signal analysis and features selection,both time-consuming and complex.In this paper,we propose a new method for optical fiber vibration signals recognition that can directly extract features,classify original signals and simplify the recognition process.In our method,the one-dimensional convolutional neural network(1DCNN)is im-proved by replacing the Softmax classifier with a support vector machine,so as to improve the stability of 1DCNN res-ults under small sample conditions.Moreover,the bird swarm algorithm(BSA)is applied to optimize the support vector machine(SVM)parameters,improving the recognition accuracy effectively.The performance of the proposed method is compared with that of other four methods,namely 1DCNN,variational mode decomposition(VMD)and SVM optim-ized by genetic algorithm(VMD-GA-SVM),VMD and SVM optimized by particle swarm optimization(VMD-PSO-SVM),VMD and SVM optimized by bird wwarm algorithm(VMD-BSA-SVM).The results show that our BS-1DCNN method performs better in accuracy and timeliness and the recognition accuracy is satisfactory.The algorithm can effect-ively improve the recognition rate of marine cable vibration signals,and can achieve better recognition effect under dif-ferent sample proportions.
维普
万方数据