Doorenbos, Lars (2020) Classification of remotely sensed imagery for assessing machine learning algorithms. Master's Thesis / Essay, Computing Science.
|
Text
mCS_2020_DoorenbosLJ.pdf Download (5MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
Toestemming.pdf Restricted to Registered users only Download (95kB) |
Abstract
The information present in remotely sensed hyperspectral imagery allows for the subtle discrimination between land covers, with applications in domains such as agriculture and surveillance. However, the inherent complexities and high volume of the data create the need for automation. For this purpose, we investigate the performance on three distinct datasets of seven supervised classifiers: Bagging, Random Forest, Extremely Randomized Trees, Stochastic Gradient Boosting, Extreme Gradient Boosting, the Support Vector Machine and 1-dimensional Convolutional Neural Networks. To allow for a fair comparison between these methods, they have to be optimized in order to determine whether the differences in performance are due to their hyperparameter configuration or the inherent qualities of the algorithms. For this purpose we use Bayesian optimization, and compare its effectiveness with grid and random search. We find that even though Bayesian optimization gives a slight improvement in two cases, it has no impact on the ordering in quality of the models. We find that Extremely Randomized Trees provides a good baseline. The Support Vector Machine is an excellent choice, as it achieves high performance while being both fast and easy to optimize. Tuning Convolutional Neural Networks is a time-consuming and unintuitive process, but once optimized they can provide good results.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Biehl, M. and Bunte, K. |
| Degree programme: | Computing Science |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 03 Aug 2020 07:13 |
| Last Modified: | 03 Aug 2020 07:13 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/22970 |
Actions (login required)
 |
View Item |