Coresets for PCB Character Recognition based on Deep Learning

The PCB (Printed Circuit Board) character image is acquired in a limited environment. Characters in the form of printed text are considered to be an easier task than handwriting. However, because the different factories use different parts, the shape of the font used on the parts may also change. In addition, characters may be damaged due to the heat treatment occurring during the character printing process. In this study, we analyze the data collected at the production plant and build coresets to create a flexible deep learning model that can be applied to multiple plant sites. Also, This experiment reveals which factors are important for coresets configuration. At this time, the generated coresets are used to analyze performance through various ResNet models.