Two MSc theses to be presented on June 1, 2018.
June 1 is an extra day for coordinated master thesis presentations in Computer Science at Lund University, Faculty of Engineering. Two MSc theses will be presented...
N.B. For presentations taking place on May 31 follow this link.
Note to potential opponents: Register as opponent to the presentation of your choice by sending an email to the examiner for that presentation (email@example.com). Do not forget to specify the presentation you register for! Note that the number of opponents may be limited (often to two), so you might be forced to choose another presentation if you register too late. Registrations are individual, just as the oppositions are! More instructions are found on this page.
13:15 (N.B. Bachelor thesis - not for MSc thesis opponents)
|TITLE||Entity-based Search (A take on the intelligent book)|
|SUPERVISORS||Pierre Nugues (LTH), Marcus Klang (LTH)|
This document describes a system to search entities in text, where we used named entity recognition to complement a traditional full text search. The named entities improve search by enabling a user to formulate queries with concepts and proper nouns and thus increase the precision of the search. Using concept and entity search, we can eliminate more easily name ambiguity and expand the search vocabulary to term variation. To carry this out, the application needs to have unique identifiers of concepts and names that are provides by wikidata. We use these identifiers to annotate the documents, in our case a corpus of textbooks in Swedish. To annotate the documents with entities, the application uses external entity linkers through APIs. Additionally we can combine the search with the information available on the semantic web.
14:15 (no more place for opponents)
|PRESENTERS||Therese Kustvall Larsson, Lisa Silfversten|
|TITLE||Coastline Detection in Satellite Images using Machine Learning Techniques|
|SUPERVISORS||Pierre Nugues (LTH), Jonas Neldeborn (Tactel Ab)|
|ABSTRACT||Today, the growing number of affordable satellite services has provided new opportunities for the use of space data. However, there are still numerous challenges in satellite image processing, and detecting diverse types of terrain can help in improving the data. This thesis investigates three different techniques for segmenting land and water masses in satellite images. The Morphological Active Contours Without Edges (Morphological ACWE) algorithm is compared with a Random Forest and a U-Net Neural Network machine learning model. In this thesis, we evaluate the different methods regarding execution time and accuracy, in addition to identifying difficulties and challenges with each technique. The resulting implementations of the three methods show that a Neural Network is the most favorable technique for this purpose, with a 96.3 % classification accuracy and a faster execution time that the other methods.|