To solve the problems of small target missed detection and false detection, difficulty in capturing multi-scale information, and low detection accuracy in complex scenes in existing target detection models for safety helmet wearing detection tasks, a safety helmet wearing detection model, SPCC-YOLOv11n, based on an improved YOLOv11n was proposed. Based on YOLOv11n, the Slim Neck lightweight network structure was adopted, with grouped shuffle convolution (GSConv) replacing standard convolution Conv and variety of view-grouped shuffle convolution cross stage partial (VoV-GSCSP) model replacing C3K2. A new P2 small target detection layer was added. Dynamic tanh (DyT) function was integrated into position-sensitive attention block (PSABlock) to form C2PSA_DyT. Channel prior convolutional attention (CPCA) was introduced at the end of the backbone network. Ablation experiments and comparative experiments with the mainstream detection models were conducted on the public dataset SHWD, and the model′s performance was verified. The results show that compared with YOLOv11n, SPCC-YOLOv11n improves precision by 22 percentage points, recall by 2.8 percentage points, and mAP@50 by 2.2 percentage points, reaching 94.5%, while mAP@50-95 improves by 17 percentage points, reaching 62.5%. The proposed model can effectively improve the detection accuracy of safety helmet wearing in complex scenarios and provide a feasible technical solution for the detection task of safety helmet wearing on construction sites.