PCB detection algorithm based on improved YOLO11n

This paper formulates an augmented detection network built upon the YOLO11n model for micro-target defect detection on the surface of PCBs. The refined network introduces the Spatial-channel decoupled downsampling (SCDown) module and substitutes the original layers of the network with the Multi-scale Convolution Block (MSCB). This method amplifies the multi-scale feature extraction ability of the convolution operation. Meanwhile, while guaranteeing accuracy, it reduces the number of model parameters and optimizes the execution efficiency of the model.The Multi-Scale Dilated Spatial Attention (MSDA) mechanism has been integrated into both the medium and small object detection layers. By doing so, it can aggregate semantic information across diverse scales, thereby boosting the proficiency of the model in detecting small targets within an extensive field of view. This approach has been validated through experiments carried out on public datasets. The results show that the improved network achieves 94.4% mAP50 in PCB surface defect detection. This marks 1.4% improvement compared to the baseline model. Simultaneously, the quantity of model parameters has been diminished by 16.9%, which improves the lightweight of the model.