SWANet: A Sliding Window Affinity Network With Dual-Branch Structure for Detection of Automotive Chip Packaging Carriers

In automotive chip packaging carrier (ACPC) defect detection, distinguishing defects is particularly challenging due to the high similarity between missing regions and background textures. Additionally, the scarcity of defect samples further exacerbates this issue. To address these problems, a defect detection network based on sliding window affinity (SWA) with dual-branch structure has been proposed. The defect sample pool is effectively expanded through a region transformation augmentation technique, which helps alleviate class imbalance issues. Affinity maps are calculated via local affine transformations to enhance internal pixel relationships, and heterogeneous feature parsing is employed to emphasize subtle proactive layer features. This method achieves a significant improvement in detection accuracy, reaching 74.38% mAP on the ACPC dataset, and demonstrates strong robustness under varying noise conditions. The results indicate that the proposed method meets the requirements of real-world production environments and has the potential for application in other industrial scenarios.