University of Twente Student Theses

Login
This website will be unavailable due to maintenance December 1st between 8:00 and 12:00 CET.

Multi-Temporal classification and change detection using UAV images

Makuti, Salma A. (2018) Multi-Temporal classification and change detection using UAV images.

Link to full-text:https://ezproxy2.utwente.nl/login?url=https://library.itc.utwente.nl/login/2018/msc/gfm/makuti.pdf
(only accessible for UT students and staff)
Abstract:Change detection is among the important image analysis techniques which help to understand how the area hasbeen changinggiven a specific period.The importance of change detection includes monitoring and controlling land cover and land use changes, city planning and management and updating of the geographic information for a certain area. Change detection requires data to be repeatedly capturedtohave multi-temporal data. The introduction of UAV technology makes easier the capture of aerialand high resolutiondata.UAV is not only the cheapest platform for data acquisition,butit is also the easiest platform to operate and have control on the quality of data needed for a specific task.In this thesis, we explore classification and change detection methods using orthophoto and DSMgenerated by UAV images. Three change detection methods have been evaluated including DSM change detection, post classificationchange detection techniqueand pre classificationchange detection technique. The data used in this thesis was takenin the construction area at Lausanne (Switzerland). A total of eight epochs wasacquiredfrom the beginning of the construction up to the end.Image differencing technique was usedin DSM change detection followed by thresholding which was used to determine the change and unchanged area. Mathematical morphology operator openingwas used to remove the noise in DSM change.By using orthophoto and DSM features as input,post classification and pre classificationchange detection wasconductedto find the change in class between the epochs. For classification purposes,theconditionalrandom field was used wherebyunary potential was defined using random forest,andpairwise potential was defined using fully connected CRF. Experiment results showthat post classification outperformsthe pre classificationchange detection method. Thiswas analysedusing overall accuracy, wherebypost classification have an accuracy of up to 63.9%,andthe pre classificationchange detection hasan accuracy of 46.5%.
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/85870
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page