Basic Application of Deep Convolutional Neural Network to Visual Inspection

Abstract

Artificial neural network (ANN) which has four or more layers structure is called deep NN (DNN) and is recognized as a promising machine learning technique. In this paper, a deep convolutional neural network (DCNN) is designed to inspect defects such as crack and burr phenomena occurred in the manufacturing process of resin molding, then the trained DCNN is evaluated through inspection experiments. A training image generator is first developed to systematically generate a lot of similar images for training. Similar images are efficiently produced by rotating, translating, scaling and transforming an original image. Then, the designed DCNN is trained using the images for the input layer and their categories for the output layer. The developed approach is assessed using the trained DCNN and shows the ability to classify sample images in a training test set into “OK” or “NG” category with high accuracy.