| Summary: | Printed Circuit Board (PCB) is a widely used electronic component and plays a critical role in the miniaturization and integration of circuits. However, the detection of PCB defects based on deep learning still encounter difficulties of limited efficiency. In order to address the issues of low speed and accuracy in PCB defect detection process, this paper proposed an innovative PCB defect detection method based on YOLOv7. Firstly, FasterNet was applied as the backbone network structure. With the new partial convolution, the spatial features were extracted more efficiently and the detection speed was improved by reducing redundant computations. Secondly, CBAM attention mechanism was integrated with feature fusion module, which allowed the model to selectively attend to relevant feature channels and spatial locations, thereby enhancing the discriminative ability of the feature representation and improving the accuracy. The experimental results indicated that the proposed model was superior to the traditional network on both PCB defect detection speed and accuracy. (1) The detection speed was increased from 54.3 frames per second to 83.3 frames per second. (2) The mAP0.5 reached 97.5% and mAP0.5:0.95 was increased from 52% to 54.7%. These improvements in speed and accuracy made it a more efficient solution for PCB defect detection.
|
|---|