Individual Tree Crown Detection and Delineation from High Spatial Resolution Imagery Using Active Contour and Hill-climbing Methods

Efficient forest management requires detailed, timely information on forests. The increasing availability and affordability of high spatial resolution remotely sensed imagery provides viable opportunities for developing automatic forest inventories at fine scale. Individual tree crown detection and delineation has become increasingly important for forest management and ecosystem monitoring. Existing methods aiming at individual tree crown detection and delineation were mainly developed for coniferous tree crown in vertical aerial imagery, and had limited capabilities in the off-nadir imagery that is commonly acquired by satellite-based sensors. In this paper, we presented a new approach applicable under various imaging conditions. This approach took into account both spectral and shape characteristics of individual tree crowns represented in different imaging conditions, as well as the expert knowledge of the stand to improve tree crown detection. The approach consists of first extracting crown area using a region-based active contour model followed by spectral-shape-based tree top detection within the crown area. The tree tops determine individual tree crown location allowing subsequent clustering of crown pixels using a new hill-climbing algorithm. We tested the approach on a Norway spruce stand using three types of high spatial resolution images: an Emerge natural color vertical aerial image, a QuickBird panchromatic image with an off-nadir view angle, and a natural color Orthoimagery. Evaluation showed our algorithm produced less than 10% tree count estimation error in the three images. Although the algorithm provided lower tree detection accuracy in off-nadir QuickBird imagery than in vertical Emerge image and orthoimagery, the accuracies for the three images were higher than the existing methods. Tree crown diameter estimation errors were less than 0.5m. Future research includes evaluating tree crown detection and delineation results by comparing with ground-based reference data.