Optimizing PCB Defect Detection through the Integration of Multi-strategy Small-Target Data Enhancement

To address core challenges in small-target printed circuit board (PCB) defect detection—including inadequate feature description, semantic overlap, and size mismatch—this study proposes a novel closed-loop multi-strategy data augmentation framework. A key contribution lies in moving beyond traditional linear feature extraction toward a holistic, multi-scale analytical framework, replacing fine-grained point-by-point analysis with broader regional assessment to boost detection precision while enforcing stricter matching criteria. To mitigate geometric distortions from viewing angle variations, a geometric transformation-based transfer matrix is employed for spatial alignment, effectively correcting image geometric deviations. Building on this, a convolutional feature pyramid module integrated with DenseNet-style dense connectivity significantly enhances small-target representation without increasing the number of parameters. Experimental results demonstrate that the proposed framework achieves mAP @0.5 scores of 92.6%, 87.8%, and 93.9%, yielding statistically significant improvements of 1.0–2.3% over state-of-the-art methods (p < 0.01). These results indicate enhanced preservation of small defects under complex background conditions, confirming the robustness and practical effectiveness of the proposed approach.