Using Terrestrial Laser Scanning Data to Unsupervised Segment Broadleaved Tree and Estimate Leaf Area

Leaves are the main organs of trees and are closely related to the biological and physical processes involved in tree growth. Segmenting leaves and estimating leaf area (LA) aids in the assessment and understanding of biological parameters, photosynthesis, and respiration in trees and forests. However, the robustness of previous segmentation methods for large and structurally complex tropical broadleaved trees is not clear. Moreover, LA is mostly estimated via indirect methods such as inversion, which is a complicated process with many influencing factors. Therefore, we propose a novel method for unsupervised segmentation of individual leaves and direct estimation of LA through biological features. The method proposes two different segmentation strategies, neighboring leaf segmentation based on point cloud linkage and individual leaves segmentation based on adaptive Euclidean clustering. Finally, LA is estimated by reconstructing the surface from the completed point cloud. The method is evaluated from two aspects: leaf segmentation and LA estimation at the tree level and leaf level using broadleaved trees in Africa. The results show that the average recall, precision, and F1-score of leaf segmentation are 96.12%, 91.09%, and 93.54%, respectively. Compared with those of common algorithms (DGG), the abilities of our method in terms of leaf segmentation are better. For LA estimation, our method demonstrated a significant correlation with the allometric model through linear regression analysis (R² = 0.55, slope = 0.84, intercept=18.86, Pearson correlation coefficient = 0.74). The average accuracy for LA estimation at the leaf level is 88.28%. Compared with those of DGG, the abilities of our method to estimate the LA at the leaf level are better. In summary, this method can be used to segment leaves effectively and estimate directly LA in broadleaved forests, indicating considerable potential for application and promotion.