Convolutional Neural Network based Surface Mounted Component Classification using Size Information

Abstract

In Automatic Optical Inspection (AOI), Surface Mounted Components (SMC) classification is a basic step for device recycling or defect detection. In classification, Convolutional Neural Network (CNN) is widely used, but there is limitation to classify SMC with similar appearances. We propose a CNN model with size information for SMC classification. We modify general CNN model by concatenating the size information to the FC layer. Our model is applicable to SMC with various sizes and achieves 91.1% performance for classification. Our model is capable 119 FPS on 128 by 128 RGB image.