Rubber trees in southern China are often impacted by natural disturbances that can result in a tilted tree body. Accurate crown segmentation for individual rubber trees from scanned point clouds is an essential prerequisite for accurate tree parameter retrieval. In this paper, three plots of different rubber tree clones, PR107, CATAS 7-20-59, and CATAS 8-7-9, were taken as the study subjects. Through data collection using ground-based mobile light detection and ranging (LiDAR), a voxelisation method based on the scanned tree trunk data was proposed, and deep images (i.e., images normally used for deep learning) were generated through frontal and lateral projection transform of point clouds in each voxel with a length of 8 m and a width of 3 m. These images provided the training and testing samples for the faster region-based convolutional neural network (Faster R-CNN) of deep learning. Consequently, the Faster R-CNN combined with the generated training samples comprising 802 deep images with pre-marked trunk locations was trained to automatically recognize the trunk locations in the testing samples, which comprised 359 deep images. Finally, the point clouds for the lower parts of each trunk were extracted through back-projection transform from the recognized trunk locations in the testing samples and used as the seed points for the region’s growing algorithm to accomplish individual rubber tree crown segmentation. Compared with the visual inspection results, the recognition rate of our method reached 100% for the deep images of the testing samples when the images contained one or two trunks or the trunk information was slightly occluded by leaves. For the complicated cases, i.e., multiple trunks or overlapping trunks in one deep image or a trunk appearing in two adjacent deep images, the recognition accuracy of our method was greater than 90%. Our work represents a new method that combines a deep learning framework with point cloud processing for individual rubber tree crown segmentation based on ground-based mobile LiDAR scanned data.

Leaf area is an important plant canopy structure parameter with important ecological significance. Light detection and ranging technology (LiDAR) with the application of a terrestrial laser scanner (TLS) is an appealing method for accurately estimating leaf area; however, the actual utility of this scanner depends largely on the efficacy of point cloud data (PCD) analysis. In this paper, we present a novel method for quantifying total leaf area within each tree canopy from PCD. Firstly, the shape, normal vector distribution and structure tensor of PCD features were combined with the semi-supervised support vector machine (SVM) method to separate various tree organs, i.e., branches and leaves. In addition, the moving least squares (MLS) method was adopted to remove ghost points caused by the shaking of leaves in the wind during the scanning process. Secondly, each target tree was scanned using two patterns, i.e., one scan and three scans around the canopy, to reduce the occlusion effect. Specific layer subdivision strategies according to the acquisition ranges of the scanners were designed to separate the canopy into several layers. Thirdly, 10% of the PCD was randomly chosen as an analytic dataset (ADS). For the ADS, an innovative triangulation algorithm with an assembly threshold was designed to transform these discrete scanning points into leaf surfaces and estimate the fractions of each foliage surface covered by the laser pulses. Then, a novel ratio of the point number to leaf area in each layer was defined and combined with the total number of scanned points to retrieve the total area of the leaves in the canopy. The quantified total leaf area of each tree was validated using laborious measurements with a LAI-2200 Plant Canopy Analyser and an LI-3000C Portable Area Meter. The results showed that the individual tree leaf area was accurately reproduced using our method from three registered scans, with a relative deviation of less than 10%. Nevertheless, estimations from only one scan resulted in a deviation of >25% in the retrieved individual tree leaf area due to the occlusion effect. Indeed, this study provides a novel connection between leaf area estimates and scanning sensor configuration and supplies an interesting method for estimating leaf area based on PCD.

Leaf attribute estimation is crucial for understanding photosynthesis, respiration, transpiration, and carbon and nutrient cycling in vegetation and evaluating the biological parameters of plants or forests. Terrestrial laser scanning (TLS) has the capability to provide detailed characterisations of individual trees at both the branch and leaf scales and to extract accurate structural parameters of stems and crowns. In this paper, we developed a computer graphic-based 3D point cloud segmentation approach for accurately and efficiently detecting tree leaves and their morphological features (i.e., leaf area and leaf angle distributions (leaf azimuthal angle and leaf inclination angle)) from single leaves. To this end, we adopted a sphere neighbourhood model with an adaptive radius to extract the central area points of individual leaves with different morphological structures and complex spatial distributions; meanwhile, four auxiliary criteria were defined to ensure the accuracy of the extracted central area points of individual leaf surfaces. Then, the density-based spatial clustering of applications with noise (DBSCAN) algorithm was used to cluster the central area points of leaves and to obtain the centre point corresponding to each leaf surface. We also achieved segmentation of individual leaf blades using an advanced 3D watershed algorithm based on the extracted centre point of each leaf surface and two morphology-related parameters. Finally, the leaf attributes (leaf area and leaf angle distributions) were calculated and assessed by analysing the segmented single-leaf point cloud. To validate the final results, the actual leaf area, leaf inclination and azimuthal angle data of designated leaves on the experimental trees were manually measured during field activities. In addition, a sensitivity analysis investigated the effect of the parameters in our segmentation algorithm. The results demonstrated that the segmentation accuracy of Ehretia macrophylla (94.0%) was higher than that of crape myrtle (90.6%) and Fatsia japonica (88.8%). The segmentation accuracy of Fatsia japonica was the lowest of the three experimental trees. In addition, the single-leaf area estimation accuracy for Ehretia macrophylla (95.39%) was still the highest among the three experimental trees, and the single-leaf area estimation accuracy for crape myrtle (91.92%) was lower than that for Ehretia macrophylla (95.39%) and Fatsia japonica (92.48%). Third, the method proposed in this paper provided accurate leaf inclination and azimuthal angles for the three experimental trees (Ehretia macrophylla: leaf inclination angle: ${\mathrm{R}}^2$ = 0.908, RMSE = 6.806° and leaf azimuth angle: ${\mathrm{R}}^2$ = 0.981, RMSE = 7.680°; crape myrtle: leaf inclination angle: ${\mathrm{R}}^2$ = 0.901, RMSE = 8.365° and leaf azimuth angle: ${\mathrm{R}}^2$ = 0.938, RMSE = 7.573°; Fatsia japonica: leaf inclination angle: ${\mathrm{R}}^2$ = 0.849, RMSE = 6.158° and leaf azimuth angle: ${\mathrm{R}}^2$ = 0.947, RMSE = 3.946°). The results indicate that the proposed method is effective and operational for providing accurate, detailed information on single leaves and vegetation structure from scanned data. This capability facilitates improvements in applications such as the estimation of leaf area, leaf angle distribution and biomass.

Rubber trees in southern China are often impacted by natural disturbances, and accurate rubber tree crown segmentation and property retrieval are of great significance for forest cultivation treatments and silvicultural risk management. Here, three plots of different rubber tree clones, PR107, CATAS 7-20-59 and CATAS 8-7-9, that were recently impacted by hurricanes and chilling injury were taken as the study targets. Through data collection using ground-based mobile light detection and ranging (LiDAR) technology, a weighted Rayleigh entropy method based on the scanned branch data obtained from the region growing algorithm was proposed to calculate the trunk inclination angle and crown centre of each tree. A watershed algorithm based on the extracted crown centres was then adopted for tree crown segmentation, and a variety of tree properties were successfully extracted to evaluate the susceptibility of different rubber tree clones facing natural disturbances. The results show that the angles between the first-order branches and trunk ranged from 35.1–67.7° for rubber tree clone PR107, which is larger than the angles for clone CATAS 7-20-59, which ranged from 20.2–43.2°. Clone PR107 had the maximum number of scanned leaf points, lowest tree height and a crown volume that was larger than that of CATAS 7-20-59, which generates more frontal leaf area to oppose wind flow and reduces the gaps among tree crowns, inducing strong wind loading on the tree body. These factors result in more severe hurricane damage, resulting in trunk inclination angles that are larger for PR107 than CATAS 7-20-59. In addition, the rubber tree clone CATAS 8-7-9 had the minimal number of scanned leaf points and the smallest tree crown volume, reflecting its vulnerability to both hurricanes and chilling injury. The results are verified by field measurements. The work quantitatively assesses the susceptibility of different rubber tree clones under the impacts of natural disturbances using ground-based mobile LiDAR.