Road extraction is an essential task in remote sensing image analysis, supporting applications such as automatic map updating and urban planning. We propose SBA-LinkNet, a lightweight encoder–decoder network that jointly enhances structural continuity and boundary awareness. The model introduces an Atrous-Shift Dual (ASD) module to expand receptive fields and preserve elongated road connectivity, and a Boundary Position Attention Module (BAM) with an auxiliary boundary-aware branch to strengthen edge localization. Experiments on the DeepGlobe and Massachusetts datasets show that SBA-LinkNet achieves IoU scores of 68.80% and 68.34% with only 8.23M parameters, outperforming several state-of-the-art CNN- and Transformer-based methods. These results demon-strate that SBA-LinkNet offers a favorable trade-off between accuracy and efficiency, making it well suited for large-scale remote sensing applications.