The written project reports from a Diploma of Technology in Information Technology at the Technical University of Denmark, taken part-time alongside full-time work between 2017 and 2019 and awarded in January 2020. The reports are in Danish.
Six assessed elements make up the 60 ECTS. Five produced a written project report; the sixth was a written examination.
| Term | Module | ECTS | Report |
|---|---|---|---|
| Spring 2014 | 02450 Introduction to Machine Learning and Data Mining | 5 | written exam |
| Spring 2017 | Basic Object Oriented Programming Grundlæggende Objektorienteret Programmering | 10 | RasterLib |
| Autumn 2017 | Advanced Object Oriented Programming Avanceret Objektorienteret Programmering | 10 | C-3PO |
| Autumn 2018 | Web Technologies Webteknologier | 10 | Bouncer |
| Spring 2019 | Big Data | 10 | Remote Sensing |
| Autumn 2019 | Final Project Afgangsprojekt | 15 | NMFk |
| Total | 60 |
The through-line, visible from the first report onwards, is spatial data: a raster library, a soil-contamination planning tool, remote sensing of gravel pits, and finally unsupervised learning on a groundwater catchment.
A new raster data type in C#, built as a reusable .NET assembly: nested classes, properties and methods for import, export and matrix operations over gridded data. The exercise was class design — deciding what a raster is as an object before deciding what it does.
A Windows desktop application in C# and WPF implementing a business process model from the soil contamination domain, as a proof of concept. Layered into client, data access and test projects, with MVVM and a repository over Entity Framework.
An authentication web service in C# and .NET built on OAuth 2.0 and OpenID Connect, providing central authentication and access control for a service-based architecture. The report documents the full authentication flow captured and traced in Wireshark, packet by packet.
Training a Mask R-CNN neural network in Python to delineate the active extraction area inside Danish gravel pits from national orthophotos — using only free public geodata, and aimed at a concrete regulatory task: the supervision of raw material extraction sites.
A test of the NMFk algorithm — non-negative matrix factorisation with automatic model selection — on a Danish dataset assembled for a groundwater abstraction catchment. Can it separate and locate contamination sources nobody identified in advance, from the mixed signal reaching the abstraction wells? Written in Julia.
The reports are published as submitted, including their appendices, which in several cases print the full source code of the project. The code was printed from working machines in 2017–2019, and has been redacted, to remove sensitive information. Nothing else was altered in the reports.
The project code is not published here. It is coursework from 2017–2019, and the appendices already show the parts of it worth reading.
The reports remain my copyright. They are published to be read and cited, not reused. Third-party figures, cover images and quoted material inside them belong to their respective owners.