Street-view building structural type recognition based on an enhanced YOLO11 network

The structural type of a building, which reflects its primary load-bearing system, is a key determinant of seismic performance and risk assessment. Accurate and scalable identification of seismic building structural types (SBST) is therefore critical for improving urban and rural seismic resilience. To overcome the limitations of manual surveys, this study proposes YOLO-SBST, a deep learning-based framework for automated SBST recognition using street-view imagery. The model targeted five representative structural types: masonry–timber, single-story masonry–concrete, multi-floor masonry–concrete, reinforced concrete, and earthen–wood structures. YOLO-SBST incorporated a pinwheel-shaped convolution module (PConv) to enhance spatial modeling and receptive field expansion, a multi-dimensional joint attention module with dilated convolution to strengthen feature extraction, and a lightweight upsampling operator to refine spatial details. In addition, an adaptive threshold focal loss function improved detection of underrepresented classes. Experiments demonstrated that YOLO-SBST achieved a precision of 0.817, recall of 0.747, mAP@50 of 0.840, and mAP@50–95 of 0.663, exceeding the YOLO11m baseline by 0.4%, 1.4%, and 2.1% in key metrics. The results demonstrate YOLO-SBST's effectiveness in fine-grained SBST classification, particularly in complex street-view scenes. The framework provides a scalable and reliable solution for seismic vulnerability mapping, supporting disaster risk reduction and resilience planning.