PCB surface defect detection algorithm based on improved GM-DEIM

In order to solve the problem of low detection accuracy and poor detection effect caused by the small defect size and the similarity of the background, an improved Geometric-adaptive Multi-resolution DEIM (GM-DEIM) algorithm was proposed to solve the six types of defects of printed circuit board (PCB), such as rat bite, burr, leakage, short circuit, open circuit and residual copper. The feature extraction ability of the model for irregular defects such as mouse bites and burrs was improved by embedding DCNv2 and Dynamic Snake Conv. Wavelet Pooling is introduced to replace traditional down-sampling, and Deformable Large Kernel Attention (DLKA) mechanism is combined to strengthen the effect of multi-scale feature fusion. Experimental results show that compared with the baseline model, the mAP50 of the improved model for PCB surface defect detection reaches 94.9%, P reaches 90.7%, and R reaches 91.4%, which are increased by 2.1%, 1.7%, and 1.2% respectively. It provides a new solution for micron-scale industrial defect detection.