Detecting Individual-Tree Crown Regions from Terrestrial Laser Scans with an Anchor-Free Deep Learning Model

摘要

Detecting individual-tree crowns provides a fundamental analysis unit bridging macro ecologicalpatterns and micro physiological functions. This study adapted an anchor-free deeplearning model, CenterNet, to detect individual crown locations and regions from dense 3Dterrestrial laser scans. A total of 1181 crowns from twelve plots were manually delineated asreference, among which eight plots were used for training the CenterNet, and another fourindependent plots for testing model accuracies characterized as the F1-score of locationdetection and Intersection over Union (IoU) of bounding box area. The maximum trainingF1-score and IoU were 0.881 and 0.670 over 40k training iterations, respectively. The resulttesting F1-score and IoU were 0.754 and 0.583, respectively. Five morphological factors werequantified to investigate the causes of accuracy variation among different plots and species,including crown area, tree height, full-width-at-half-maximum, nearest neighbor crown distance,and overlapping ratio of neighboring crowns. Results show that tree height was mostimportant trait for crown detection. A taller, larger, smoother, less crowded, and less overlappedtree was found easier to detect. Among six species, red pine, Scots pine, and silverbirch were successfully detected, and Norway spruce, lodgepole pine, and trembling aspenwere more difficult to detect.