Laser-induced breakdown spectroscopy (LIBS) is promising for rapid contamination analysis of power insulators, but its reliability is limited by laser-energy perturbation during field detection. To improve the robustness of LIBS-based contamination-level classification, this study proposes a LIBS–DVS Hybrid Gated Fusion Spiking Neural Network (LD-HGF-SNN) by integrating LIBS spectra with raw dynamic vision sensor (DVS) event streams. Unlike methods based on reconstructed event images, LD-HGF-SNN constructs a raw-event-derived spatiotemporal DVS event tensor to preserve the time–height–width evolution of laser-induced plasma and encodes it using an LIF-neuron-based spiking branch. A 1D convolutional branch extracts LIBS spectral features, while a sample-adaptive gated fusion module regulates the contributions of spectral and event-stream information. Broad-energy classification and cross-energy evaluation tasks were constructed using paired LIBS–DVS data collected under seven laser-energy conditions. LD-HGF-SNN achieved 97.50% accuracy and 97.49% macro-F1 under the reference-energy condition, and maintained 87.50% accuracy and 87.52% macro-F1 under full-range laser-energy perturbation. In the most challenging task, its accuracy exceeded the strongest ablation model, the ungated fusion model, and the reconstructed-image fusion baseline by 15.50, 21.00, and 22.50 percentage points, respectively. These results indicate that raw-event-derived DVS event streams provide complementary plasma-dynamic information for LIBS spectra, and that preserving the spatiotemporal structure of DVS events can improve LIBS classification robustness under laser-energy perturbation.