|
PresentationsMachine learning methods for segmenting three-dimensional images of forest vegetationM.V. Lomonosov Moscow State University Laser scanning is increasingly being used in forestry. The use of LiDAR and similar technologies makes it possible to abandon the use of hand tools in favor of computer data processing. The result of laser scanning of an object is a three-dimensional point image called a point cloud.
To measure some characteristics (trunk diameter at breast height, green mass volume), it is necessary to separate the crown from the trunk. From the point of view of image processing, this task is equivalent to image segmentation, i.e. splitting an image into non-overlapping semantically complete parts.
When processing tree images, segmentation is usually performed by classifying points according to a set of features using a random forest. The objective of this work is to investigate the possibility of using a simpler naive Bayesian classifier to solve the segmentation problem.
The research was carried out on images from the SYSSIFOS collection [2]. To train the classifier, a set of 8 scanned trees stored as point clouds and marked up manually was used [3]. The proportions of correctly classified crown and trunk points were used as a metric for evaluating the quality of the machine learning model.
Literature 1. Lin, W., Fan, W., Liu, H., Xu, Y., Wu, J. Classification of Handheld Laser Scanning Tree Point Cloud Based on Different KNN Algorithms and Random Forest Algorithm // Forests , 2021, Vol 12(3), 292. https://doi.org/10.3390/f12030292 2. Weiser, H., Schäfer, J., Winiwarter, L., Krašovec, N., Fassnacht, F.E., Höfle, B. Individual tree point clouds and tree measurements from multi-platform laser scanning in German forests // Earth System Science Data, Vol. 14 (7), 2022, pp. 2989-3012. https://doi.org/10.5194/essd-14-2989-2022 3. Weiser, H., Veit, U., Winiwarter, L., Esmorís, A.M., Höfle, B. Manually labeled terrestrial laser scanning point clouds of individual trees for leaf-wood separation// 2024. https://doi:10.11588/data/UUMEDI
Presentation |
