Due to differences in the size, shape, and density of lung nodules, traditional single-scale methods struggle to effectively handle nodules of varying sizes and often face issues of low accuracy and detail loss. This paper proposes a lung nodule detection method based on an improved YOLO11n model. Firstly, in the C2PSA module, the self-attention mechanism is replaced with a multi-frequency multi-scale attention module (MFMSA), which combines frequency-domain and multi-scale features to enhance the detection capability for small targets. Secondly, a modulated fusion module (MFM) replaces the concatenation method in the bottleneck structure, dynamically adjusting weights to improve the representation of key features and the adaptability of the model. Finally, a convolutional additive self-attention mechanism (CASA) is combined with C3K2 to alleviate the loss of detailed features. Experimental results show that the improved YOLO11n-sms model achieves increases of 1.29, 0.71, 0.63, and 1.77 percentage points in precision, recall, mAP50, and mAP50-95, respectively, reaching 92.04, 92.8, 95.5, and 73.0%.