In recent years, deep neural network (DNN) has shown excellent performance in many applications. However, the huge energy consumption leads to many problems. In order to solve this problem, spiking neural network (SNN) has attracted extensive research attention. SNN is the third-generation neural network that is used to process complex spatiotemporal data, and it has become a hot research topic due to its event-driven and low-power characteristics. However, the propagation function of spiking neurons is usually non-differentiable, which prevents back propagation and makes the training of SNN difficult. This paper proposes an efficient supervised learning algorithm framework based on dual-network-model spiking neural network (DNM-SNN), which is universal to various supervised learning algorithms of spiking neural networks and can effectively improve the prediction accuracy. DNM-SNN includes two key methods. Firstly, a dual model training method in training stage is proposed, which requires an additional auxiliary network same as the network used. Single model training is easy to fall into local optimal problems. By maintaining two networks, the same problem can be viewed from different perspectives, which solves the problem and improves the training effect. Second, we propose a multi-channel mix module inference method in the prediction stage. The prediction accuracy of the model is improved and the performance of the spiking neural network is optimized by multi-channel optimization of mix module. Experimental results show that the DNM-SNN outperforms the single-model algorithm on classification tasks, with a slight improvement on the MNIST dataset and a 3% improvement on the CIFAE-10 dataset.