CB-YOLO: composite dual backbone network for high-frequency transformer coding defect detection

The high-frequency transformer is a crucial component of various electronic devices. The coding on high-frequency transformers indicates the product specification and lot number, which is essential for product usage. The coding may experience defects such as dirt, characters missed, and characters overlapped. Accurately detecting these defects is crucial to ensuring product quality. Conventional defect detection algorithms are unable to achieve satisfactory detection accuracy. To address this issue, we propose a modified detection network called Composite Backbone You Only Look Once (CB-YOLO), based on YOLOv5s. Firstly, a composite dual backbone network that is assembled by the assisting backbone and the lead backbone is established to improve the network’s feature extraction capability. Additionally, an attention mechanism is introduced into the network to enhance its focus on defective regions. The experimental results show that the mAP of CB-YOLO achieves 93.8% on the self-made dataset, increasing by 2.8% over YOLOv5s. CB-YOLO also has a higher accuracy compared to other excellent defect detection algorithms.