Towards Deep Learning in Industrial Applications Taking Advantage of Service-Oriented Architectures

Abstract In reverse logistics, identification of products is necessary but due to uninterpretable markers information flow is not always consistent. Recent image-based recognition developments using Convolutional Neural Networks are promising but collecting required labeled data is time- and cost-intensive. To allow a quick deployment and usage of such systems, we present a conceptual service-oriented architecture that enables Deep Learning recognition systems to be used with initially small but growing data sets, as with every usage training data expands on run-time. An identification problem is reduced to digitization and labeling of data and as a side effect digital knowledge retention can be established in companies.