Training spiking neural networks (SNNs) presents significant challenges due to their intrinsic temporal dynamics, controlled by neuron parameters. The initialization and training of these parameters critically affect network performance, but conventional artificial neural network (ANN) methods are unsuitable for SNNs because of their different temporal nature. In this study, we explore a novel initialization and training scheme tailored to the unique temporal characteristics of SNN parameters. We utilize a spiking Gaussian derivative network with sparse event-based vision input signals to improve the performance and efficiency of SNNs by addressing their distinct temporal dynamics.