An improved YOLOv8-SRA-based method was developed for apparent defect detection in 3D-printed concrete to improve the quality control level of 3D-printed concrete. Real-time detection challenges and low detection accuracy were addressed for defects including holes, pocked surfaces, fractures, and collapses. Initially, the Small head module was inserted into the Head layer with a view to enhancing the detection of minor defects. Concurrently, the Restormer attention mechanism module was incorporated into the Backbone layer, with the objective of enhancing the extraction of multi-scale defect features in complex backgrounds. Furthermore, the AKConv module was incorporated into the Neck and Backbone layers with a view to enhancing the extraction and detection of irregular defects. Ultimately, the synergistic effectiveness of the modules and the model’s superiority were confirmed by means of ablation experiments and comparative tests. The modified model achieved a 94.7% mAP@0.5, a 92.6% recall rate, and a 91.3% precision rate, which represented enhancements of 6.2, 9.7, and 7.4 percentage points, respectively, when compared to the original model. These enhancements addressed the limitations of the original model, such as its tendency to overlook minor defects and its proclivity to erroneously detect irregular defects. This study offers a novel approach to the quality control of 3D-printed concrete buildings and structures in practical engineering applications.