Study of Visual Inspection for Liquid Pouches Using Deep Learning

Visual inspection for liquid pouches using deep learning is under discussion. Pictures of the produced liquid pouches were taken using soft X-ray transmission to generate deep learning datasets. Our deep learning system was designed with an autoencoder and a neural network for two-class classification. Our autoencoder was trained with defect-free datasets and performed for an image pre-processing to extract abnormal defect features. Our neural network for two-class classification was trained with pre-processed defect-free and defective datasets. Our proposed method detected defective products using our test datasets at running time.