Intelligent event recognition of ϕ-OTDR in pipeline leakage monitoring based on fully connected ANN with signal and features combined input

The event recognition of phase sensitive optical fiber sensor(ϕ-OTDR) is a challenging problem in pipeline leakage monitoring system. Due to the effect of changeable noise and disturbance along the sensing fiber, it is difficult for general signal processing methods including support vector machine(SVM) and general neural network, to classify the complex and vague features in ϕ-OTDR. Therefore, the combined signal and features have been proposed to be together fed into the neural network for pattern recognition in ϕ-OTDR. The combined signal and features sequences are used in fully connected artificial neural network (FC-ANN) for training and testing. Through application in the pipeline leakage monitoring system, it is shown that the proposed method can achieve an average accuracy of 94% for three main classes of events, which is superior to the SVM classifier and FC-ANN with only signal input. And the performance of the proposed method with changing the number of features and the size of signals has also been studied.