The practical use of Spiking Neural Networks (SNNs) in real-world applications is hindered by their limited robustness and efficiency in handling diverse data distributions. This lack of generalization in varying environments stems from the inherent uncertainties present within the model and its input. To address this problem, we propose to integrate prospect certainty into the SNN training process. We first analyze the components of the SNN model to identify the sources of uncertainty and their impact on the output. By incorporating the prospect certainty, we then extend the original loss function into a multi-objective one, allowing the model to balance both the loss and the output certainty during training. In this way, the model's generalization and adaptability are enhanced, thus ensuring its accuracy in variable environments. Furthermore, we design a hybrid architecture that combines a Convolutional Neural Network (CNN) with the SNN to implement the extended loss function. The entire approach is tested on benchmark datasets and the results demonstrate that incorporating the prospect certainty into the training pipeline significantly improves the model's accuracy and adaptability.