Remote sensing image change detection holds significant importance in fields such as urban dynamic monitoring and ecological assessment. However, restricted by parameter capacity, lightweight models struggle to capture subtle changes in complex scenes. Furthermore, factors such as slight registration errors and local textural noise often trigger high-frequency pseudo-change responses, leading to false detections. To this end, this article proposes a lightweight dual-domain directional network (LDDN) for resource-constrained scenarios, achieving efficient and reliable change detection through the synergistic design of frequency-domain noise suppression and spatial-domain directional modeling. Specifically, to suppress unreliable high-frequency responses induced by factors like registration errors, a directional wavelet attention module (DWAM) is designed. It utilizes the directional high-frequency subbands of the discrete wavelet transform (DWT) to perform adaptive screening, thereby suppressing unreliable components and mining structural cues associated with real changes. To enhance the capability of lightweight models in capturing weak change targets, a directional spatial fusion module (DSFM) is proposed, which constructs long-range anisotropic modeling capabilities at a low cost by decoupling horizontal and vertical contextual dependencies. Furthermore, to ensure semantic consistency across multiple scales, a multiscale interaction module (MIM) is introduced to facilitate cross-scale synergy between frequency-domain filtering and spatial-domain directional modeling. The experimental results demonstrate that with only 1.98M parameters and 2.58G FLOPs, LDDN achieves competitive or superior detection performance compared to various state-of-the-art methods across four public benchmark datasets. Particularly on the challenging CLCD dataset, the$F1$-score and intersection over union (IoU) are improved by 0.83% and 1.07%, respectively, compared to the second-best method. The code is available at:https://github.com/Vi-chen/LDDN