Spiking neural networks (SNNs), as an emerging neuromorphic computing paradigm, offer high efficiency, ultra-low power consumption, and robust temporal processing capabilities, making them ideal for intelligent computing at the edge. However, the inherent binary spike representation leads to significant information loss, limiting their practical performance. To overcome this issue, we propose a multi-bit information transmission mechanism, expanding neuron outputs from a single bit to multiple bits. This enhancement significantly enriches spike representation and reduces information loss while preserving low-energy advantages. Furthermore, effective signals from preceding layers are utilized to re-stimulate neurons, promoting complete multi-level spike emissions. Extensive experiments on various benchmark datasets demonstrate consistent performance improvements, highlighting the method’s suitability for resource-constrained intelligent computing applications.