University of Twente Student Theses

Login

LULU operators for image segmentation and object detection: Reedland in The Netherlands

Yu, Cheng (2013) LULU operators for image segmentation and object detection: Reedland in The Netherlands.

[img] PDF
13MB
Abstract:Reedlands have important functions for ecosystem and environment. Being able to monitor the condition of the reedlands and to provide correct information on them is necessary for the management on the reedlands. However, the characteristic of the reedlands of being mixed with water and other wetland vegetation makes the task of collecting information on reedlands become difficult. Therefore, this thesis project aims to investigate the possibility of detecting the reedlands from a satellite image by using LULU operators, and then to compare the performance of LULU analysis with that of the traditional method of maximum likelihood classification (MLC). The study site is the Weerribben area in the Netherlands where reedlands flourish. A multispectral image obtained by the Geoeye-1 satellite over this area in 2011 was applied, and was further convert into a 1- band NDVI image for LULU analysis. The validation or reference data is a management types map of 2012 provided by the State Forest Department Staatsbosbeheer (SBB), the owner of the Weerribben. The management types map was further modified into a land cover map to be used as validation/ reference data. Four kinds of uncertainty in the validation/reference data were concluded and discussed. In this research, the ability of segments, which generated by LULU analysis, to represent reedland is first examined. Then reed maps based on the classification by MLC and LULU analysis were produced and evaluated. Validation was performed on the results by overlaying the reference map onto the reed maps from MLC and LULU analysis. For LULU analysis, 2 sets of training samples were used: 3 reedlands samples and 7 reedlands samples. Finally the validation results gave the user accuracy and the producer accuracy of MLC and the LULU analysis based on the two sets of reedlands samples. Keywords: LULU analysis, DPT, MLC, object detection, image segmentation
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:https://purl.utwente.nl/essays/93816
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page