The accurate implementation of pulmonary nodule detection is crucial for the clinical application of early lung cancer diagnosis. Currently

it faces three core challenges: small target feature loss

low-contrast interference

and insufficient detection accuracy. To address these issues

this study completes the improvement of YOLO series models and the implementation of pulmonary nodule detection based on the concept of cross-domain technology transfer. The key work includes: 1) Transferring the GhostNetV1 lightweight backbone from walnut defect detection to reconstruct the YOLOv11 backbone network

reducing the parameter count by 24.8% while preserving nodule detail features

thus laying a foundation for lightweight model deployment; 2) Adapting the fused module of Spatial Squeeze-and-Excitation (SSE) and Mixed Local Channel Attention (MLCA) from SAR ship detection to enhance the discrimination of low-contrast nodule features

improving detection accuracy by 3.2% on the LUNA16 dataset; 3) Introducing the Adaptive Weighted Normalized Wasserstein Distance (AWNWD) loss from insulator detection to optimize the bounding box regression logic for small nodules

increasing mAP@50 by 4.1%; 4) Incorporating the high-resolution detection layer design from liquor bottle cap defect detection to add a 160×160 P2 layer

enabling effective capture of micro-nodules (≤3mm). Validation on the LUNA16

LungCT

and Node21 datasets shows that the implemented YOLOv11-SSE-AWNWD model achieves 96.1% Precision

96.3% Recall

and 97.8% mAP@50 on LUNA16

outperforming mainstream models such as YOLOv8 and YOLOv10. Meanwhile

the lightweight design with 2.1M parameters balances detection accuracy and clinical application feasibility. This study clarifies the application path of cross-domain technology transfer in YOLO model improvement and pulmonary nodule detection implementation

providing a reference for the engineering application of medical image detection.