Abstract: To address the issues of unstable learning and uniform weight distribution in spiking neural networks during complex feature learning and classification tasks, an adaptive LIF model is proposed, which is combined with a newly designed adjustable multiplicative STDP rule. By introducing exponential mapping of synaptic traces and multiplicative modulation mechanisms, the LIF neuron's responsiveness to input spikes is effectively improved, and the network's adaptability to complex signals is optimized. The proposed STDP rule, which integrates normalized presynaptic traces and a Sigmoid function, balances synaptic weight adjustments between adaptability and stability, significantly improving learning efficiency and network stability. Experimental results show that the proposed method accurately identifies features of different directions and polarities in real-world roadmap texture and rotating disk datasets collected by a dynamic vision sensor. On the MNIST handwritten digit dataset, the improved model achieves a classification accuracy of 98.7%, validating the method's effectiveness and robustness.