Skip to main content
Login | Suomeksi | På svenska | In English

Browsing by Author "Huusari, Riikka"

Sort by: Order: Results:

  • Huusari, Riikka (2016)
    This study is part of the TEKES funded Electric Brain -project of VTT and University of Helsinki where the goal is to develop novel techniques for automatic big data analysis. In this study we focus on studying potential methods for automated land cover type classification from time series satellite data. Developing techniques to identify different environments would be beneficial in monitoring the effects of natural phenomena, forest fires, development of urbanization or climate change. We tackle the arising classification problem with two approaches; with supervised and unsupervised machine learning methods. From the former category we use a technique called support vector machine (SVM), while from the latter we consider Gaussian mixture model clustering technique and its simpler variant, k-means. We introduce the techniques used in the study in chapter 1 as well as give motivation for the work. The detailed discussion of the data available for this study and the methods used for analysis is presented in chapter 2. In that chapter we also present the simulated data that is created to be a proof of concept for the methods. The obtained results for both the simulated data and the satellite data are presented in chapter 3 and discussed in chapter 4, along with the considerations for possible future works. The obtained results suggest that the support vector machines could be suitable for the task of automated land cover type identification. While clustering methods were not as successful, we were able to obtain as high as 93 % accuracy with the data available for this study with the supervised implementation.