On the application of deep learning methods in the real world: Image-classification of shipping container X-ray scans

Krebbers, Jasper (2020) On the application of deep learning methods in the real world: Image-classification of shipping container X-ray scans. Master's Thesis / Essay, Artificial Intelligence.

Preview	Text MasterThesisJasperKrebbers.pdf Download (7MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (99kB)

Abstract

The port of Rotterdam is the largest port in Europe and millions of containerspass through annually. The herculean task of regulating, securing and inspectingthis flow of goods falls on the shoulders of Dutch customs. Dutch customs is theamong the biggest customs agencies of Europe. In recent decades they adapted tothe increasing globalization by introducing and applying new techniques such asX-ray scanning, however due to manpower shortages only a fraction of shippingcontainers can be inspected. With Brexit around the corner and yearly increasesin shipping volume Dutch customs will need to work both harder and smarter.In this thesis we focus on the application of deep learning algorithms in the cus-toms domain using a real-world dataset of X-ray scanned shipping containers.We examine several aspects of the deep learning design process and evaluatewhich approaches can improve the reliability, extendability and explainability ofdeep learning systems. We created a dataset containing 14 common classes inorder to compare several state-of-the-art pretrained deep learning neural net-works. We found that all of them were able to achieve excellent performancewith recall, precision and F1 scores. In order to verify the importance of pre-training we trained several networks with different types of initialization andsections frozen. Even though the X-ray domain looks very different from theIma

Item Type:	Thesis (Master's Thesis / Essay)
Supervisor name:	Schomaker, L.R.B.
Degree programme:	Artificial Intelligence
Thesis type:	Master's Thesis / Essay
Language:	English
Date Deposited:	02 Apr 2020 11:42
Last Modified:	02 Apr 2020 11:42
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/21725

Actions (login required)

View Item