Spiking neural networks (SNNs) require lower energy consumption in pattern hyperspectral image classification tasks than artificial neural networks due to their biological rationality and information processing methods that execute sparse event-driven information. They are suitable for computing classification tasks on satellite hyperspectral images with limited resources. However, due to the impact of the SNN network structure and spike time steps, there is an urgent need to address the inference speed problem in SNNs for hyperspectral image classification tasks while ensuring classification accuracy. To overcome the aforementioned challenges, this paper introduces a simple and effective training strategy for knowledge distillation-based lightweight low-latency spiking neural networks. This method compresses the network structure and time dimension of the original SNN model. By using knowledge distillation, the original SNN model acts as the teacher network, while the newly constructed SNN serves as the student network. This approach leverages the similarity or consistency between the two structures, transferring high-precision knowledge from the teacher to the light-weight and low-latency student network. This enables the student network to achieve high performance comparable to the teacher network. This study thoroughly evaluated four benchmark spaceborne hyperspectral datasets and two public unmanned aerial vehicle hyperspectral datasets. The experimental results show that the proposed method reduces the number of model parameters by 61% and improves inference speed by 5–7 times with only a slight loss in classification accuracy. This study provides a practical solution for applying SNNs in real-time hyperspectral image analysis.