Multi-interest modeling has become a crucial approach in sequential recommendation, as it better captures users’ diverse and evolving preferences by analyzing their behavior sequences into interest-specific representations. However, challenges remain in overcoming interference from external information, such as multi-modal data, and superfluous information within the sequence. To address these issues, we propose IAMIM (Interference-Aware Multi-Interest Modeling), a novel sequential recommendation model consisting of the following two main components. (1) The Feature Debiasing Module (FDM) is proposed to reduce irrelevant information from multi-modal data by transforming item features into the frequency domain using Fourier Transform and learnable filters. (2) A Temporal Capsule Network (TCN) is designed, incorporating a temporal relevance coefficient generated via causal convolution, which is integrated into the capsule network to adaptively reduce the influence of outdated items. Experiments conducted on real-world datasets demonstrate that IAMIM outperforms existing methods, achieving superior accuracy in handling both external and internal interference.