EEMNet:An End-to-End Efficient Model for Electronic Fabrication Defect Detection

In order to ensure the quality and production efficiency of electronic products, defect detection during electronic fabrication is crucial. In this paper, an end-to-end efficient model (EEMNet) for electronic fabrication defect detection is designed in this paper. In the backbone network, a semi-combined dual attention module(SCDAM) and a feature-enhanced down-sampling module(FEDM) are proposed to enhance the feature representation of tiny defects. In the neck network, a channel-weighted spatial pyramid pooling(CWSPP) is proposed to fuse different receptive field features, and the features are further fused by a simplified feature pyramid network(FPN) and pyramid attention network(PAN). Finally, superior IoU(SIoU) loss is proposed for the problem that CIoU of small defects are susceptible to position deviation and degradation in some cases. The experimental results on printed circuit boards(PCB) and surface mounted devices(SMD) datasets show that our model has superior performance in detecting tiny defects and achieves a good balance between accuracy, speed and model size.