Gesture recognition based on transient surface electromyographic (sEMG) signals offers a rapid recognition strategy, which is advantageous for hand movement rehabilitation. However, the recognition performance remains constrained due to the limitations of conventional feature extraction and classification methods. To address this, our study proposes a novel approach that combines sEMG decomposition with a residual spiking neural network (Res-SNN) for analyzing transient high-density sEMG signals. Specifically, motor unit spike trains (MUSTs) are extracted through signal decomposition, and transient segments are identified based on gesture onset detection. A Res-SNN model is then designed to accurately classify gestures from these early-stage signal fragments. Experimental results on a self-collected healthy dataset containing 35 gesture classes demonstrate that the proposed method achieves high recognition accuracy (0.915 ± 0.122) using only the initial portion of gesture signals, validating both the effectiveness of the transient recognition strategy and the practical potential of the Res-SNN architecture. The proposed method shows potential to enhance both responsiveness and real-time performance in gesture recognition systems.