A new signal recognition method is presented based on the derivative analysis and feature extraction.Firstly,the time-frequency characteristics are obtained thought analyzing the relation of derivative and fr equency.Secondly,according to the time-frequency distribution characteristic,several typical features are extracted to reduce fea ture dimensions and recognition time. Finally,probability neural network (PNN) is used to identify and learn.The algor ithm is verified by four disturbance signals, the average accuracy rate reaches 95.7%,and the recognition time is less than 0. 23s.The experimental results show that the method can identify the type of disturbance quickly and accurately,and provide a scientific and reliable method for the pattern recognition of perimeter system based on Mach-Zehnder (M -Z) interferometer.