Distributed optical fiber acoustic sensing (DAS) signal has problems with strong noise and difficult recognition. To solve these problems, a deep residual shrinkage network based on new threshold function (DRSN-NTF) is proposed. DRSN-NTF uses new threshold function instead of soft threshold function on the basis of deep residual shrinkage network (DRSN), which makes it more capable in signal noise processing and classification recognition. DAS system is used to collect the experimental data of perimeter intrusion events, and six groups of experiment with different signal-to-noise ratios (0 dB?5 dB) are designed by adding Gaussian white noise. The experimental results of the four models are compared to investigate the recognition effect of DRSN-NTF. The results show that the average test accuracy of DRSN-NTF is 1.05% higher than that of DRSN in the case of strong noise. With the reduction of the signal-to-noise ratio, the difference between the test accuracy of DRSN-NTF and that of DRSN increases, indicating that DRSN-NTF is more capable in signal noise processing and classification recognition, which can lead to relatively higher recognition accuracy. Therefore, DRSN-NTF is more suitable for recognition of DAS signal.