Fig 1.
A defect-free PCB undergoes artificial anomaly creation to generate synthetic defects for training. The ViT-Mamba architecture processes these images using a ViT encoder for feature extraction, multi-scale convolutional layers for hierarchical representation, and a Mamba-inspired decoder with attention gates.
Fig 2.
Artificial defect creation pipeline: A defect-free PCB undergoes mathematically defined defect generation and random transformations to produce defected samples.
Fig 3.
Illustration of various masks generated using the artificial defects creation module.
Fig 4.
Mamba attention gate workflow.
Encoder () and decoder (
) features are weighted, combined, and passed through ReLU and sigmoid activations to compute attention weights (ψ). These weights refine
, producing the final feature map (
) for improved segmentation.
Table 1.
Network architecture of the ViT-Mamba framework. The table provides a detailed overview of the components, layer types, output shapes, and additional details.
Fig 5.
Examples from the PCB defect dataset, showcasing one image per defect category with annotated bounding boxes: missing hole, mouse bite, open circuit, short, spur, and spurious copper.
Table 2.
Comparison of different methods for PCB defect detection.
Fig 6.
Performance comparison of top methods across defect types.
Table 3.
Performance comparison from ablation experiments. Each component is removed or replaced to assess its contribution to the model. Results are reported as mean Average Precision (mAP).