To achieve real-time detection of optical cables and vibration risk assessment，it is necessary to reliably identify fiber optic vibration patterns. Therefore，a performance optimization method for fiber optic vibration deep learning algorithm in pattern recognition was studied. This method utilizes distributed fiber optic sensors to collect vibration signals，and obtains light intensity signals based on the phase information of the Rayleigh scattering light signals from the sensors，converting them into electrical signals；After selecting the windowing and framing method to process the signal，the local feature scale decomposition method is used to obtain multiple feature parameters of the signal，which are then input into a one-dimensional convolutional neural network. The model learns and outputs the fiber vibration pattern recognition results. The test results show that this method can obtain the frequency changes of different vibration signals. After windowing and frame processing，the frequency of the vibration signal fluctuates between 0~400 Hz，and the phase does not exceed 20 rad. It accurately extracts the peak to peak value and spectral characteristics of the signal’s main lobe，and presents its trend with frequency，clearly showing the intensity and distribution of different frequency components. The significance lies in the effective completion of fiber optic vibration pattern recognition，which can determine the depth of fiber optic cable vibration signals and assist in optimizing the effectiveness of fiber optic cable operation fault detection and risk assessment.