Detection of weak and small defects on the surface of semiconductor lead frame based on SBN-yolov7

The quality of the semiconductor lead frame will directly affect the life of the chip. Due to the complex production process, there will be defects on the surface of the product that are similar to the surface texture. These defects are characterized by low contrast and small area, they are frequently overlooked or incorrectly identified during inspection procedures. In response to the above problems, this paper proposes the SBN_YOLOv7 defect detection algorithm. Firstly, SPD-Conv is introduced in the feature extraction stage to avoid excessive loss of fine-grained information and allow the network to better extract features of weak and small defects. Secondly, in order to prevent the network from being interfered by complex textures and noise, Biformer is added in the feature fusion stage to improve the network's ability to learn deep important features. Finally, in order to reduce the impact of position deviation on the detection network, NWD is embedded in the loss function. Experimental results show that this method has made significant progress in improving detection results. The MAP value reached 89.6%.