University of Twente Student Theses

Login

Effectiveness of Markov random field based method for super-resolution mapping in identifying small landscape elements and forest encroachment from satellite images

Tiwari, Laxmi Kant (2011) Effectiveness of Markov random field based method for super-resolution mapping in identifying small landscape elements and forest encroachment from satellite images.

[img] PDF
2MB
Abstract:Small Landscape Elements (SLE) and Forests are major determinants of a landscape’s identity and help to maintain ecological and environmental stability, play a key role in subsistence economy. However, little is known about SLE and their existence outside forests and few attempts has been done to study forest encroachment (FE) in the Indian scenario. The present research was carried out to check the effectiveness of Markov Random Field (MRF) based Super Resolution Mapping (SRM) in the identification of SLE and FE. In this study ASTER image with spatial resolution of 15 m is used for all experimental tests. SLE identification is done at Buurserzand in the Netherlands, whereas, the study for identification of FE is performed at Rutland Island in India. This research work dealt with the real world problem by using real remotely sensed image. Quality of SRM is compared with MLC classified map. The results of this study were validated using Google Earth image data. Simulated Annealing (SA) parameters were tuned on real data and result is compared with study done on simulated data before. Method parameter of MRF based SRM was evaluated on ASTER data. Accuracy was assessed at fine and coarse resolution using kappa statistics and error measures. It is observed that SRM outperformed MLC in case of fine and coarse resolution. Experimental test was done on the ASTER data to study the optimal neighborhood system size with respect to various Scale factor (S). It is found that resultant map of SRM was smoother than MLC due to resolution issue. This study dealt with Low resolution case in both study sites wherein object was bigger than resolution of pixel cell. Hence low accuracy was reported. Moreover, MRF based SRM is successfully identified SLE and FE at S = 2 by using ASTER image with spatial resolution of 15 m. Similarly, SRM result was successfully validated using Google Earth image. This study provides guidelines for conducting similar research work by using MRF based SRM for identification of Tree resources outside forests (TROF) and other similar studies related with vegetation. Key words: Markov Random Field, Simulated Annealing, Super Resolution Mapping, ASTER, Small Landscape Elements, Forest Encroachment, Root Mean Square Error, Google, Correlation Coefficient, Area Error Proportion, Scale factor (S), Fine resolution, Coarse resolution, Sub-pixel mapping, Maximum Likelihood Classification, Parameter Optimization.
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/92791
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page