Towards structure‐based classification of tree crowns in high spatial resolution aerial images

Abstract

Some metrics useful for characterizing certain patterns of individual tree crowns are presented. They could be applied in an automated image analysis system for tree species classification in digital high spatial resolution aerial images. The structure‐based measures made use of the internal pattern (also called texture) of individual tree crowns. The infrared layer of colour infrared aerial images was utilized and the different crown patterns were extracted using the second‐order Laplace transform. Skeletonization was used for data compression and to enhance features to be used in the subsequent analysis. The resulting skeleton branches were one pixel thick but they still contained the structural pattern that was identified by the Laplace transform. The skeleton was split into simple segments without branches. The orientation of each individual segment in the tree crown was calculated. A Hough transform technique was developed for analysing whether they collectively had a dominating parallel or radial structural behaviour. It could discriminate clearly between Picea abies (L.) Karst, with a typical radial pattern, and Pinus sylvestris L. and Betula sp., sometimes dominated by parallel structures. In our images, the Scots pines showed a more clearly parallel pattern, which was probably caused by a combination of the inclination angle of the sunlight and a slight camera motion during exposure.

Key words:

• aerial image
• automation
• crown structure
• digital image analysis
• Norway spruce
• pattern recognition
• remote sensing
• texture
• tree species classification