Effizientes Labeln von Artikeln für das Einlernen Künstlicher Neuronaler Netze

Abstract

Convolutional neural networks (CNN) have been increasingly used in object detection in recent years due to their high detection accuracy and high detection speed. Despite fast and reliable classifications, one of the biggest disadvantages is that the training of such a network is very time consuming. The reason for this is that, depending on the complexity of the object to be detected, several hundred already classified learning images are required. Until now, the creation of these learning images was mainly done manually. For this reason, an assistance system was developed at the Institute for Material Handling and Logistics (IFL), which accelerates the learning of new objects considerably compared to the traditional manual method.