Method and apparatus for training a convolutional neural network to detect defects

Abstract

The present application discloses a method of training a convolutional neural network for defect inspection. The method includes collecting a training sample set including multiple solder joint images. A respective one of the multiple solder joint images includes at least one solder joint having one of different types of solder joint defects. The at least one solder joint is located substantially in a pre-defined region of interest (ROI) in a center of the image. The method further includes inputting the training sample set to a convolutional neural network to obtain target feature vectors respectively associated with the multiple solder joint images. Additionally, the method includes adjusting network parameters characterizing the convolutional neural network through a training loss function based on the target feature vectors and pre-labeled defect labels corresponding to different types of solder joint defects. The training loss function includes at least two different loss functions.

Dependencies are inferred from the claim text.