To address the problems that fault features of rotating machinery are easily submerged under strong-noise operating conditions and that spiking neural networks exhibit low accuracy under low signal-to-noise-ratio conditions, this paper proposes a rotor fault diagnosis method based on an adaptive soft-threshold residual spiking neural network. First, a data representation method based on Gramian Angular Difference Field transformation is introduced to convert a small amount of one-dimensional vibration signals into two-dimensional images, thereby enhancing the temporal correlation and global structural representation capability of the signals. Second, a two-stage spiking convolutional feature extraction network is constructed, and an adaptive soft-threshold mechanism is incorporated into the high-level feature blocks to suppress noise interference and highlight fault-sensitive information. Meanwhile, the temporal modeling capability of spiking neural networks is utilized to effectively separate noise features from fault features. Finally, experiments are conducted using rotor system data from Wuhan University. The results show that the proposed method demonstrates better anti-noise performance than traditional neural networks under different signal-to-noise-ratio conditions and can effectively improve the accuracy of fault identification for rotor systems in noisy environments.