Improving YOLOv5s PCB defect detection based on ECA and BiFPN

The proposed method addresses the issues of low precision and inaccurate defect location in conventional PCB defect detection algorithms by incorporating ECA and BiFPN improved YOLOv5s. Firstly, the ECA attention mechanism is introduced into C3 of the backbone network to enhance the model's focus on small target feature information, ensuring effective detection. Secondly, the weighted bidirectional feature pyramid network (BiFPN) is employed for efficient multi-scale feature fusion. Lastly, SIoU Loss replaces CIoU Loss to further enhance model stability. Experimental results using the same PCB dataset demonstrate that the improved model achieves a mAP of 98.1%, an increase in FPS by 4.68 compared to the original model, thereby improving both detection accuracy and speed to meet practical requirements for PCB defect detection.