Individual tree crowns delineation using local maxima approach and seeded region growing technique

Paper in proceedings (conference paper)

en

Remote sensing applications in forestry can profit from a rapid development of optical sensors. New hyperspectral sensors have very high spatial and spectral resolution and provide continuous spectral cover in visible and infrared spectral region. Applied algorithms should be suited to the new properties of the data to achieve its maximal advantage. Segmentation of the image into objects is a fundamental task in image processing. It is important in forestry applications of optical remote sensing as well. We are looking for a position of individual tree crowns. Such process traditionally involves two parts: 1) detection and 2) delineation phase. Local maxima approach and seeded region growing technique are presented as the key concepts. Furthermore improvements, namely histogram equalization and Voronoi diagram, are incorporated. Two independent datasets were processed and results of the segmentation are presented. Hyperspectral data with spatial resolution of 0.8m were found as a suitable for segmentation process with 84% and 78% accuracy of detection phase and 64% and 52% accuracy in delineation phase respectively. Finally discussion of recommended settings in the algorithm is provided based on the segmentation results.

tree crowns, delineation, seeded region growing technique

2011-01-26

978-80-248-2366-9

GIS Ostrava 2011

49–59

10