Master thesis presentation by Lars Kristian Topphem Kinn

Examinator: Erik Agrell

Abstract:
Because of extensive downtime caused by damaged cardboard boxes at an automated warehouse in Vänersborg, Sweden. In this study we examine the possibility of using computer vision and artificial intelligence when identifying irregularities among the handled boxes. As a motive for doing this study it is assumed that a fault detection application based on computer vision and artificial intelligence would be cost effective. A dataset containing images of both damaged and unharmed boxes were gathered and used to train the convolutional neural networks. Due to the small size of the dataset, transfer learning was used to speed up the learning process. The hyperparameter values were altered throughout the process to maximize the validation accuracy. To decide upon suitable hyperparameter values, hyperparameter values are optimized through Bayesian optimization techniques. Our results showed that the outcome of DarkNet-19 as the best network fitting our problem.

Welcome!

Lars Kristian and Erik