University of Twente Student Theses


Automatic container code identification using Machine Learning

Licu, D.V. (2020) Automatic container code identification using Machine Learning.

[img] PDF
Abstract:With the accelerated development of the global industries and the increasing number of containers that pass through a single terminal every day, repetitive and slow parts of the shipping process need to be automatized in order to keep up with the market. Possible automations are numerous, but all start and make use of an automatic container code identification system, needed in order to replace the manual identification that takes place nowadays at most of the terminals. The goal of this research is to design and evaluate such a system and also investigate the effect of damage on the performance of the automatic identification. The proposed method is composed by two parts, a detection step and a recognition step and uses a Convolutional Neural Network and Gaussian heatmaps for text detection and a kNN classifier for character recognition. The results show a 100% detection rate for containers in good shape and 80% for damaged containers. The overall identification results show 40.4% of complete container codes identified for containers in good shape and 12.0% for damaged containers. When the proposed algorithm outputs a complete code, that is in 100% of the times the real container code, with 0% false positive rate. The proposed pipeline proved to be successful for the container identification task, facing some challenges due to the chosen recognition method, challenges that can be overcome increasing the recognition dataset. Regarding the damage, the proposed pipeline has some issues detecting heavily damaged characters at the extremities of a container code.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Interaction Technology MSc (60030)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page