Although deep learning has made significant progress in remote sensing image change detection (CD), existing methods still struggle to suppress disturbances such as style differences and perspective distortions due to insufficient interaction of deep spatiotemporal features. A wavelet-based feature interaction and modulation network (BFIM-Net) is proposed to address this challenge, Which enhances detection robustness through frequency-domain spatiotemporal interaction. Specifically, we introduce the bitemporal frequency-feature interaction module (BFFI) to align bitemporal features in the frequency domain, thereby mitigating interference caused by style and viewpoint differences. We further propose the cross fusion enhancement module (CFEM) to capture multi-scale semantic information of changes. In the decoder stage, we introduce the feature fusion and enhancement module (FFEM) to integrate upsampled coarse-grained global features with fine-grained local semantics from skip connections, effectively improving boundary localization and detail detection. BFIM-Net achieves F1 scores of 94.45% on WHU-CD and 91.51% on LEVIR-CD, exceeding state-of-the-art (SOTA) approaches by 2.64% and 1.10%, respectively, as evidenced by experiments on these public datasets. Visualization results further validate its robustness under complex imaging conditions.