<p>Existing monocular depth estimation methods often struggle to simultaneously achieve high accuracy and high inference efficiency, and some approaches incur substantial inference overhead while improving prediction accuracy. To address this issue, this paper proposes a lightweight self-supervised monocular depth estimation framework that reduces computational cost while maintaining competitive prediction accuracy. First, a lightweight Efficient Model (EMO) encoder based on an inverted residual structure is introduced to achieve unified modeling of local and global features. Second, a lightweight Bidirectional Spatial–Channel attention (BSC) module is designed to enhance feature representation through efficient interaction between spatial and channel information. Finally, an Edge-Aware Dynamic-Weighted Distillation Strategy (EADS) is proposed to guide the student network in learning structured representations from the teacher model, thereby reinforcing structural consistency in key regions such as object boundaries and alleviating artifacts and blurring caused by channel compression. Comprehensive evaluations are conducted on the KITTI and Make3D datasets, and the experimental results demonstrate that the proposed method achieves superior or competitive performance in terms of accuracy, generalization, and efficiency, while exhibiting strong robustness under cross-dataset evaluation. Ablation studies further validate the effectiveness of each component, indicating that the proposed framework enables efficient and reliable depth prediction under lightweight constraints.</p>