Transformer-based sequence recommenders have achieved promising results by effectively capturing the dependency relationships between items in sequences. However, the self-attention mechanism focuses on the entire interaction sequence, inevitably assigning certain weights to false-positive items. This generates noisy sequence representations, making it challenging for the model to accurately learn user preferences. In this paper, we propose a novel Filtering-enhanced Denoising Transformer for Sequential Recommendation (FDT4SR). Specifically, we replace the standard self-attention block with a filtering-enhanced self-attention block as the encoder to adaptively attenuate noise in sequence representations. Simultaneously, stochastic substitution operations are employed to introduce random noise into the sequence for adversarial training, thereby enhancing the model's generalization. Extensive experiments conducted on four real-world datasets demonstrate that our model outperforms the state-of-the-art baselines in both performance and robustness. Our code is available at https://github.com/Liminss/FDT4SR .