Inspection method and inspection device

Abstract

This application discloses an inspection method and an inspection device, which relates to a field of intelligent manufacturing technology and aims to solve a problem of how to improve inspection accuracy of circuit boards. The inspection method includes: constructing and training an inspection model, a network architecture of the inspection model is YOLOv8 network, and a loss function of the inspection model includes a Gaussian Distribution Similarity Measure (GDSM) function; obtaining an image of a circuit board; inputting the image into the inspection model and outputting an inspection result. The loss function of the YOLOv8 network usually includes an Intersection Over Union (IOU) function, the IOU function has lower error sensitivity to local features and is difficult to accurately capture image details. Replacing the IOU function with the GDSM function, the GDSM function has higher error sensitivity to both global features and the local features, and can accurately identify the image details, such that improving an accuracy of the inspection model.

Images