Distributed optical fiber perimeter security systems have proven to be an effective method for security monitoring of important targets such as power plants, substations, and telecommunications base stations. However, this method can be challenging to distinguish between different types of intrusion behaviors and is prone to false alarms triggered by various environmental interferences. With the increasing actual demand, there are higher requirements for the accuracy of perimeter signal recognition. The perimeter security system in the new era not only needs to perform real-time monitoring, recognition, and response alarms for various types of intrusion behaviors, but also requires features such as remote control and response, high-precision intrusion location, multi-environmental adaptability, resistance to various disturbances, and low energy consumption. Therefore, it is necessary to conduct research on effective extraction and accurate recognition algorithms for intrusion signal characteristics. This article reviews the feature extraction methods combining the time domain, frequency domain, and time-frequency domain of optical fiber perimeter signals, and the classification and recognition methods based on vector machines, neural networks, and deep learning. It specifically discusses the principles and application scenarios of various algorithms, and conducts a comparative analysis of their advantages and disadvantages.