A Random Forest Algorithm for PCB SMD Defect Detection

A printed circuit board (PCB) surface-mounted devices (SMDs) defect detection algorithm based on random forest (RF) was proposed to detect pin defects, resistor defects, and capacitor defects. It can solve the problem of low detection accuracy and low efficiency caused by dense layout and small size of SMD. First, the PCB image is preprocessed, which includes image graying, image stitching, geometric correction, component positioning, and image denoising. Second, the SMD to be tested is divided into subregions. The shape features, gray features, and texture features of the subregions are extracted. Then the RF model is established, and the classification and regression tees (CART) decision tree is selected as the basic classifier. Each decision tree in the RF makes a prediction according to the sample characteristics. Finally, all the decision tree predictions are voted to get the final prediction. The experimental results show that the average accuracy of this method for pin defect, resistor defect, and capacitor defect detection is 97.7%, 97.0%, and 96.7%. The detection time for a PCB board containing 200 SMDs is only 3.78 s.