Semi-supervised semantic segmentation leverages a substantial amount of unlabeled data to significantly enhance the performance of semantic segmentation networks when annotated data is limited. However, this approach still faces challenges such as multi-scale target segmentation difficulties and ambiguous category boundaries. To address these issues, this paper proposes a semi-supervised segmentation network combining spatial enhancement and feature alignment. Firstly, the study merges two shallow features in the encoder using a Spatial Dual-attention Enhancement Module (SDEM) to capture both global and local information simultaneously. This process enhances the corresponding spatial detail features while filtering out irrelevant interference information to accommodate variations in the target scale. Secondly, a Cross-layer Feature Alignment Module (CFAM) is employed to align high-level and low-level features, generating fine features with higher resolution. This alignment compensates for information lost during the downsampling of deep features and further enhances segmentation performance. Finally, the Boundary Loss (BL) is introduced to increase attention to the boundary pixels, thereby significantly improving the segmentation accuracy of boundaries and small-scale targets. In various partition protocols, the proposed method demonstrates performance improvements of 1.44%, 1.36%, and 1.21% compared to the baseline, respectively.