| Summary: | English: This study describe the use of HySpex hyperspectral images and QuickBird satellite images for the classification of forest and vegetation in Stor-Elvdal, especially forest and vegetation that are a resource for moose browsing. Field data was gathered on several tree species and vegetation types in the study area. The sample points were exactly georeferenced with the use of GPS and Vexcel orthophotos. A HySpex flight strip (25 cm resolution) and a QuickBird satellite image (60 cm resolution) of the exact same area were used for the classification. Both images have been orthorectified. No images were atmospherically corrected. For the QuickBird image, the Normalize Difference Vegetation Index and several texture images were calculated. For the HySpex image two data reduction methods were applied: Principal Component Analysis and Minimum Noise Fraction transformation. Both images were classified by an object oriented approach with use of the software eCognition. After a segmentation procedure, a Nearest Neighbour (NN) classification method was applied. In addition to the Nearest Neighbour classification, Support Vector Machines (SVM) and a Decision Tree (DT) classification was performed on the same classes and sample data. The overall classification accuracy of the QuickBird classification was about 40%. Regrouping the classing into ‘pine, ‘spruce’, ‘deciduous’, and ‘other groundcover’ improved the result up to 78%. There was little difference between the results of the NN, the SVM, and the DT classifiers for the Quickbird images. The overall classification accuracy of the HySpex classification was about 63%. Regrouping the classes improved the result to 76% (for the NN classifier) to 81% (for the SVM and DT classifiers). The HySpex classification discriminated the classes ‘pine’ good, ‘spruce’, and ‘willow’ reasonably well, while the QuickBird classification only discriminated the class ‘pine’ reasonably well (over 70% accuracy). We were not able to map the browsing pressure with the limited field data we ...
|
|---|