Dynamic Visual Sensors (DVS) offer significant advantages over traditional cameras by capturing event-based data. However, they face challenges in managing time-dimensional data redundancy. In this paper, we propose a novel approach for constructing Spiking Neural Networks (SNNs) utilizing a Recurrent and Stateful Self-Connected Leak-Integrate-and-Fire (RS-LIF) model to enhance the efficiency and effectiveness of processing DVS event streams. The RS-LIF model integrates recursion and self-connection mechanisms, facilitating continuous neuron firing and improving processing efficiency. This model also includes an event-to-frame integration technique to ensure compatibility with traditional computer vision frameworks. Additionally, we introduce a hybrid pooling mechanism that combines the advantages of max and average pooling to further enhance SNN performance. The proposed framework was evaluated on the DVS128 Gesture and Fashion-MNIST datasets. Experimental results demonstrate that our framework achieves higher accuracy in temporal information processing with fewer time steps, particularly excelling in action recognition accuracy compared to the standard Leaky Integrate-and-Fire (LIF) model.