Microseismic monitoring technology can give the spatial location of rock body rupture or instability in real time and accurately, and has become one of the important means of early warning for disasters such as coal and gas herniation and tunnel rock explosion. Aiming at the problem of complex environment and weak signals difficult to be recognized effectively in underground engineering, a microseismic signal recognition method combining convolutional neural network and Transformer (T_CNN) is proposed. Six kinds of signals in tunnel engineering in a western region are collected by using fiber-optic acceleration sensors, and the signals are input into the model for training and verification after band-pass filtering for noise reduction and Fourier transform. Convolutional neural network in the model is utilized for feature extraction, focusing on the key information based on Transformer, and the final multi-classification results are derived by multilayer perceptron. The results show that the classification accuracy of the T_CNN-based model reaches 98.09% and converges faster. Compared with the current state-of-the-art residual neural network, the accuracy is improved by 6.2%, and the precision, recall, and F1 score are improved by 0.036, 0.023, and 0.033, respectively, which confirms the superiority of the algorithm in practical engineering applications. In addition, the energy of the fiber microseismic signal can also be estimated more accurately after the fiber microseismic signal is input into the model after the feature transformation, which further verifies that the model has good application prospects.