Ultrawideband (UWB) radar is a radar system that leverages an exceedingly broad frequency spectrum for signal transmission, offering distinctive advantages in target recognition. However, the radar’s echo detection may be susceptible to interference from environmental noise and clutter signals, complicating the identification of target information. Addressing this issue, this chapter designs a mixture of experts transformer network based on time-frequency domain fusion (TF-MOEFormer) using data collected from an UWB radar for the task of vehicle classification. The network initially employs a Fast Fourier Transform to obtain frequency domain information from the radar’s time-domain signals, followed by the fusion of time-frequency domain and positional encoding information, which serves as the input features for the feature extraction network. This feature leverages the frequency domain’s interference resistance capabilities in target recognition, thereby enhancing its strength. In this study, an improved transformer network is utilized to extract hidden information from the fused features, which is composed of multiple layers of mixture of experts transformer blocks. Compared to traditional transformer networks, it dynamically selects a subset of expert networks for computation through a routing network, maintaining computational costs while improving the model’s accuracy, and effectively enhancing its generalization capabilities. Extensive experiments on the utilized UWB radar data demonstrate that the proposed model achieves excellent classification accuracy and robust interference resistance.