A Surface Defect Detection method for vacuum gauges based on VAG-YOLO

Vacuum gauges are the key equipment in vacuum inspection equipment, and the surface defects of vacuum gauges will directly affect the inspection performance and service life of vacuum inspection equipment. At present, the surface defect detection of vacuum gauges mainly relies on the visual inspection of workers, which is less efficient and accurate, and the workers are prone to misjudge the products due to subjective factors. To solve the problems of traditional manual inspection, this paper proposes an improved vacuum gauge surface defect detection method based on the YOLOv5s model called VAG-YOLO. we add a multi-scale adaptive fusion structure (MAF) to the YOLOv5s model to make full use of adaptive fusion of features at different scales to improve the detection performance of the network and increase the defect detection accuracy; Meanwhile, the transformer bottleneck structure (BoT) is introduced to combine multi head Self- Attention (MHSA) with convolutional neural network (CNN) to achieve the effect of reducing the number of network parameters and improving the detection speed. The experimental results show that the average detection accuracy of the VGA-YOLO model is 83.4%, which is higher and faster than the detection accuracy of various other algorithms, and can detect vacuum gauge surface defects in real time.