Current methods for estimating monocular depth based on global attention mechanisms excel in capturing long-range dependencies, however, they often have drawbacks of high computational complexity and numerous parameters. Additionally, these methods can be susceptible to interference from irrelevant regions, which reduces their ability to accurately estimate fine details. This study proposes a self-supervised monocular depth estimation model based on a local attention mechanism, which further leverages convolution and Shuffle operations for global information interaction. The proposed method first calculates attention within divided local windows and then effectively integrates global information by combining depthwise separable convolutions and Shuffle operations across spatial and channel dimensions. Experimental results on the public KITTI dataset demonstrate that the proposed method significantly reduces computational complexity and parameter count and improves the ability to handle depth details, outperforming mainstream methods based on global attention mechanisms.