Abstract: In an optical fiber perimeter alarm system, when analyzing and identifying the fiber vibration signal, there are a series of limitations such as network broadband, storage capacity and computing speed in the process of sampling, storage, transmission and signal processing of high-frequency large-scale signal. In order to solve this problem, an adaptive compression sensing method of optical fiber perimeter alarm signal based on wavelet packet was proposed. Firstly, multi-scale wavelet packet decomposition was used to decompose the optical fiber vibration signal. By calculating the mathematical expectation of the high frequency part of the wavelet packet coefficients at different scales as the threshold value, wavelet packet coefficients were set to zero. The wavelet packet decomposition scale was adaptively selected so that the signal sparse in frequency domain. Then, wavelet packet coefficients were classified according to the mathematical expectation and information entropy of the wavelet packet coefficients. According to different types of coefficient blocks, the corresponding processing methods were designed to improve the speed of signal transmission and processing. The results show that this method can effectively reduce the observation data of optical fiber vibration signal. At the same sampling rate, it can improve the accuracy and speed of signal reconstruction.