Defect detection method of electronic components based on improved YOLOX

It is difficult to detect artificially in view of the small defect size of electronic components in industrial environment, this paper proposed a defect detection method of electronic components based on improved YOLOX algorithm. Firstly, a new Spatial Pyramid Pooling structure is used in the feature extraction part, in which Maxpool is replaced by Softpool to retain more feature information. Secondly, in the feature fusion part, the channel attention is added to enhance effective features and improve the detection ability of small size defects. Finally, the α-Complete Intersection over Union is used to optimize loss function and the cosine annealing algorithm is used to accelerate the convergence of the model to achieve the best detection results. The improved YOLOX method is tested on the self-made electronic component defect dataset, the average detection accuracy is 99.05%, and the detection time of each picture is about 0.18 seconds. It is verified in the PCB defect public dataset with good experimental results.