Individual Tree Crown Segmentation Research Articles

The ISPRS International Contest on Individual Tree Crown Segmentation using High-Resolution Images and the Initial Findings

Abstract. Tree canopy plays an essential role in the biophysical activities in forest environment. During the past two decades, individual tree delineation using high-resolution imagery data has become a hot topic in forest sensing research. Individual Tree Crown (ITC) segmentation methods aim to generate masks that delineating the boundary of each ITC, which supports various tree parameter extractions. Thanks to the rapid development of deep learning, the ITC segmentation methods achieved remarkable improvement. However, existing research suffers from the limited availability of the datasets, and the lack of evaluation standards as well as task-orientated neural networks. In 2024, the International Society of Photogrammetry and Remote Sensing (ISPRS) launched the first International Contest of ITC Segmentation. The contest aims to reveal the state-of-the-art of the ITC method development using high-resolution images, to clarify the remaining barriers and challenges, and to guide further explorations in the field. This paper overviews the contest and reports the initial finding regarding the impact factors of the method performance.

Synergizing a Deep Learning and Enhanced Graph-Partitioning Algorithm for Accurate Individual Rubber Tree-Crown Segmentation from Unmanned Aerial Vehicle Light-Detection and Ranging Data

The precise acquisition of phenotypic parameters for individual trees in plantation forests is important for forest management and resource exploration. The use of Light-Detection and Ranging (LiDAR) technology mounted on Unmanned Aerial Vehicles (UAVs) has become a critical method for forest resource monitoring. Achieving the accurate segmentation of individual tree crowns (ITCs) from UAV LiDAR data remains a significant technical challenge, especially in broad-leaved plantations such as rubber plantations. In this study, we designed an individual tree segmentation framework applicable to dense rubber plantations with complex canopy structures. First, the feature extraction module of PointNet++ was enhanced to precisely extract understory branches. Then, a graph-based segmentation algorithm focusing on the extracted branch and trunk points was designed to segment the point cloud of the rubber plantation. During the segmentation process, a directed acyclic graph is constructed using components generated through grey image clustering in the forest. The edge weights in this graph are determined according to scores calculated using the topologies and heights of the components. Subsequently, ITC segmentation is performed by trimming the edges of the graph to obtain multiple subgraphs representing individual trees. Four different plots were selected to validate the effectiveness of our method, and the widths obtained from our segmented ITCs were compared with the field measurement. As results, the improved PointNet++ achieved an average recall of 94.6% for tree trunk detection, along with an average precision of 96.2%. The accuracy of tree-crown segmentation in the four plots achieved maximal and minimal R2 values of 98.2% and 92.5%, respectively. Further comparative analysis revealed that our method outperforms traditional methods in terms of segmentation accuracy, even in rubber plantations characterized by dense canopies with indistinct boundaries. Thus, our algorithm exhibits great potential for the accurate segmentation of rubber trees, facilitating the acquisition of structural information critical to rubber plantation management.

Comparing Deep Learning and MCWST Approaches for Individual Tree Crown Segmentation

Abstract. Accurate segmentation of individual tree crowns (ITC) segmentation is essential for investigating tree-level based growth trends and assessing tree vitality. ITC segmentation using remote sensing data faces challenges due to crown heterogeneity, overlapping crowns and data quality. Currently, both classical and deep learning methods have been employed for crown detection and segmentation. However, the effectiveness of deep learning based approaches is limited by the need for high-quality annotated datasets. Benefiting from the BaKIM project, a high-quality annotated dataset can be provided and tested with a Mask Region-based Convolutional Neural Network (Mask R-CNN). In addition, we have used the deep learning based approach to detect the tree locations thus refining the previous Marker controlled Watershed Transformation (MCWST) segmentation approach. The experimental results show that the Mask R-CNN model exhibits better model performance and less time cost compared to the MCWST algorithm for ITC segmentation. In summary, the proposed framework can achieve robust and fast ITC segmentation, which has the potential to support various forest applications such as tree vitality estimation.

Unveiling the Potential of Drone-Borne Optical Imagery in Forest Ecology: A Study on the Recognition and Mapping of Two Evergreen Coniferous Species

The use of drone-borne imagery for tree recognition holds high potential in forestry and ecological studies. Accurate species identification and crown delineation are essential for tasks such as species mapping and ecological assessments. In this study, we compared the results of tree crown recognition across three neural networks using high-resolution optical imagery captured by an affordable drone with an RGB camera. The tasks included the detection of two evergreen coniferous tree species using the YOLOv8 neural network, the semantic segmentation of tree crowns using the U-Net neural network, and the instance segmentation of individual tree crowns using the Mask R-CNN neural network. The evaluation highlighted the strengths and limitations of each method. YOLOv8 demonstrated effective multiple-object detection (F1-score—0.990, overall accuracy (OA)—0.981), enabling detailed analysis of species distribution. U-Net achieved less accurate pixel-level segmentation for both species (F1-score—0.981, OA—0.963). Mask R-CNN provided precise instance-level segmentation, but with lower accuracy (F1-score—0.902, OA—0.822). The choice of a tree crown recognition method should align with the specific research goals. Although YOLOv8 and U-Net are suitable for mapping and species distribution assessments, Mask R-CNN offers more detailed information regarding individual tree crowns. Researchers should carefully consider their objectives and the required level of accuracy when selecting a recognition method. Solving practical problems related to tree recognition requires a multi-step process involving collaboration among experts with diverse skills and experiences, adopting a biology- and landscape-oriented approach when applying remote sensing methods to enhance recognition results. We recommend capturing images in cloudy weather to increase species recognition accuracy. Additionally, it is advisable to consider phenological features when selecting optimal seasons, such as early spring or late autumn, for distinguishing evergreen conifers in boreal or temperate zones.

Instance segmentation of individual tree crowns with YOLOv5: A comparison of approaches using the ForInstance benchmark LiDAR dataset

Fine-grained information on the level of individual trees constitute key components for forest observation enabling forest management practices tackling the effects of climate change and the loss of biodiversity in forest ecosystems. Such information on individual tree crowns (ITC's) can be derived from the application of ITC segmentation approaches, which utilize remotely sensed data. However, many ITC segmentation approaches require prior knowledge about forest characteristics, which is difficult to obtain for parameterization. This can be avoided by the adoption of data-driven, automated workflows based on convolutional neural networks (CNN). To contribute to the advancements of efficient ITC segmentation approaches, we present a novel ITC segmentation approach based on the YOLOv5 CNN. We analyzed the performance of this approach on a comprehensive international unmanned aerial laser scanning (UAV-LS) dataset (ForInstance), which covers a wide range of forest types. The ForInstance dataset consists of 4192 individually annotated trees in high-density point clouds with point densities ranging from 498 to 9529 points m-2 collected across 80 sites. The original dataset was split into 70% for training and validation and 30% for model performance assessment (test data). For the best performing model, we observed a F1-score of 0.74 for ITC segmentation and a tree detection rate (DET %) of 64% in the test data. This model outperformed an ITC segmentation approach, which requires prior knowledge about forest characteristics, by 41% and 33% for F1-score and DET %, respectively. Furthermore, we tested the effects of reduced point densities (498, 50 and 10 points per m-2) on ITC segmentation performance. The YOLO model exhibited promising F1-scores of 0.69 and 0.62 even at point densities of 50 and 10 points m-2, respectively, which were between 27% and 34% better than the ITC approach that requires prior knowledge.Furthermore, the areas of ITC segments resulting from the application of the best performing YOLO model were close to the reference areas (RMSE = 3.19 m-2), suggesting that the YOLO-derived ITC segments can be used to derive information on ITC level.

Study on Individual Tree Segmentation of Different Tree Species Using Different Segmentation Algorithms Based on 3D UAV Data

Individual structural parameters of trees, such as forest stand tree height and biomass, serve as the foundation for monitoring of dynamic changes in forest resources. Individual tree structural parameters are closely related to individual tree crown segmentation. Although three-dimensional (3D) data have been successfully used to determine individual tree crown segmentation, this phenomenon is influenced by various factors, such as the (i) source of 3D data, (ii) the segmentation algorithm, and (iii) the tree species. To further quantify the effect of various factors on individual tree crown segmentation, light detection and ranging (LiDAR) data and image-derived points were obtained by unmanned aerial vehicles (UAVs). Three different segmentation algorithms (PointNet++, Li2012, and layer-stacking segmentation (LSS)) were used to segment individual tree crowns for four different tree species. The results show that for two 3D data, the crown segmentation accuracy of LiDAR data was generally better than that obtained using image-derived 3D data, with a maximum difference of 0.13 in F values. For the three segmentation algorithms, the individual tree crown segmentation accuracy of the PointNet++ algorithm was the best, with an F value of 0.91, whereas the result of the LSS algorithm yields the worst result, with an F value of 0.86. Among the four tested tree species, the individual tree crown segmentation of Liriodendron chinense was the best, followed by Magnolia grandiflora and Osmanthus fragrans, whereas the individual tree crown segmentation of Ficus microcarpa was the worst. Similar crown segmentation of individual Liriodendron chinense and Magnolia grandiflora trees was observed based on LiDAR data and image-derived 3D data. The crown segmentation of individual Osmanthus fragrans and Ficus microcarpa trees was superior according to LiDAR data to that determined according to image-derived 3D data. These results demonstrate that the source of 3D data, the segmentation algorithm, and the tree species all have an impact on the crown segmentation of individual trees. The effect of the tree species is the greatest, followed by the segmentation algorithm, and the effect of the 3D data source. Consequently, in future research on individual tree crown segmentation, 3D data acquisition methods should be selected based on the tree species, and deep learning segmentation algorithms should be adopted to improve the crown segmentation of individual trees.

Multispecies individual tree crown extraction and classification based on BlendMask and high-resolution UAV images

Accurate detection and segmentation of individual trees from unmanned aerial vehicle images is critical for forestry resource surveys and accurate forest management. Deep learning methods have been used for studies of individual tree crown segmentation, classification, and number of trees in mixed coniferous and broad-leaved forests, but the accuracy needs to be improved. Therefore, this study uses BlendMask, a simpler and more efficient algorithm that combines Mask R-CNN and Yolact algorithms to effectively combine instance-level information with semantic information at a finer granularity level, greatly improving crown segmentation accuracy and classification results. Three coniferous species and five broad-leaved species unmanned aerial vehicle images collected from the Jing Yue multispecies ecological forestry site in Changping District, Beijing, were used as the dataset, and the results were compared with Yolact and Mask R-CNN. The results show that the method described in this work has the highest Kappa coefficient (0.89) and overall accuracy (92.14%) in the test set. For segmentation accuracy, coniferous species’ producer’s accuracy was 0.91 to 0.95, whereas that of broad-leaved species was 0.89 to 0.92. For species classification, the F1-score and mean average precision for coniferous species were greater than 91%, whereas those for broad-leaved species were 77.64% to 85.63%. The accuracy of extracting stand density in low and medium canopy density stands was 0.9909 and 0.9422, respectively, whereas that in high canopy density stands was 0.8913. This study shows that the BlendMask model has a good effect in studying the classification of multiple tree species, the segmentation of individual tree crowns, and the statistics of the number of trees in complex forest areas. Compared with broad-leaved forests and high canopy density stands, this model is more suitable for coniferous forest and medium and low canopy density stand scenarios. This study provides an important tool for obtaining more accurate species classification, canopy segmentation, and resource inventory results in complex forest areas.

Individual Tree Crown Segmentation and Crown Width Extraction From a Heightmap Derived From Aerial Laser Scanning Data Using a Deep Learning Framework.

Deriving individual tree crown (ITC) information from light detection and ranging (LiDAR) data is of great significance to forest resource assessment and smart management. After proof-of-concept studies, advanced deep learning methods have been shown to have high efficiency and accuracy in remote sensing data analysis and geoscience problem solving. This study proposes a novel concept for synergetic use of the YOLO-v4 deep learning network based on heightmaps directly generated from airborne LiDAR data for ITC segmentation and a computer graphics algorithm for refinement of the segmentation results involving overlapping tree crowns. This concept overcomes the limitations experienced by existing ITC segmentation methods that use aerial photographs to obtain texture and crown appearance information and commonly encounter interference due to heterogeneous solar illumination intensities or interlacing branches and leaves. Three generative adversarial networks (WGAN, CycleGAN, and SinGAN) were employed to generate synthetic images. These images were coupled with manually labeled training samples to train the network. Three forest plots, namely, a tree nursery, forest landscape and mixed tree plantation, were used to verify the effectiveness of our approach. The results showed that the overall recall of our method for detecting ITCs in the three forest plot types reached 83.6%, with an overall precision of 81.4%. Compared with reference field measurement data, the coefficient of determination R2 was ≥ 79.93% for tree crown width estimation, and the accuracy of our deep learning method was not influenced by the values of key parameters, yielding 3.9% greater accuracy than the traditional watershed method. The results demonstrate an enhancement of tree crown segmentation in the form of a heightmap for different forest plot types using the concept of deep learning, and our method bypasses the visual complications arising from aerial images featuring diverse textures and unordered scanned points with irregular geometrical properties.

Evaluation of UAS LiDAR data for tree segmentation and diameter estimation in boreal forests using trunk- and crown-based methods

Very high point density laser scanning data from unmanned aerial systems (UAS) can be used to segment the trunks of individual trees. Such segmentation (e.g., individual trunk segmentation (ITS)) is useful, for example, in the estimation of diameter at breast height (DBH), which in turn is needed for the estimation of other tree and stand attributes, such as stem volume and diameter of the basal area median tree (DGM). In this paper, we assess the estimation of DBH directly from UAS LiDAR data in open and closed canopy conditions that represent the range of operational conditions encountered in boreal forests in Finland. We also compare trunk-based DBH estimates to corresponding estimates from individual tree crown segmentation (ITC) and fuse the results from the trunk- and crown-based estimates. The results showed that trunk segmentation performed slightly better than ITC in open canopy areas, whereas ITC performed better in closed canopy areas. The DBH prediction error was smaller for ITS (3.0 cm) than ITC (3.9 cm) when considering the trees that were recognized by both methods. We also conclude that a hybrid method, where both segmented tree trunks and tree crowns are fused, considerably increases the number of correctly segmented trees but does not decrease the prediction error associated with DGM compared to using either ITC or ITS individually.

Towards low vegetation identification: A new method for tree crown segmentation from LiDAR data based on a symmetrical structure detection algorithm (SSD)

Obtaining low vegetation data is important in order to quantify the structural characteristics of a forest. Dense three-dimensional (3D) laser scanning data can provide information on the vertical profile of a forest. However, most studies have focused on the dominant and subdominant layers of the forest, while few studies have tried to delineate the low vegetation. To address this issue, we propose a framework for individual tree crown (ITC) segmentation from laser data that focuses on both overstory and understory trees. The framework includes 1) a new algorithm (SSD) for 3D ITC segmentation of dominant trees, by detecting the symmetrical structure of the trees, and 2) removing points of dominant trees and mean shift clustering of the low vegetation. The framework was tested on a boreal forest in Sweden and the performance was compared 1) between plots with different stem density levels, vertical complexities, and tree species composition, and 2) using airborne laser scanning (ALS) data, terrestrial laser scanning (TLS) data, and merged ALS and TLS data (ALS + TLS data). The proposed framework achieved detection rates of 0.87 (ALS + TLS), 0.86 (TLS), and 0.76 (ALS) when validated with field-inventory data (of trees with a diameter at breast height ≥ 4 cm). When validating the estimated number of understory trees by visual interpretation, the framework achieved 19%, 21%, and 39% root-mean-square error values with ALS + TLS, TLS, and ALS data, respectively. These results show that the SSD algorithm can successfully separate laser points of overstory and understory trees, ensuring the detection and segmentation of low vegetation in forest. The proposed framework can be used with both ALS and TLS data, and achieve ITC segmentation for forests with various structural attributes. The results also illustrate the potential of using ALS data to delineate low vegetation.

Oblique view individual tree crown delineation

Individual tree crown (ITC) segmentation supports numerous applications in forest management and ecology. In the latter context, special attention is dedicated to the study of angular reflection effects, caused by the interaction of incident sunlight with a canopy. High precision airborne analysis of these effects requires multi-view sensor systems and multi-view ITC segmentation. In particular oblique view image segmentation is difficult and has been addressed by numerous template based methods. This contribution identifies persistent shortcomings in the state of the art and tackles the problem by a multi-step workflow utilizing the digital surface model (DSM), derived from multi-view stereo data. A slightly revised version of the previously published levelset-watershed segmentation of the DSM is presented as the first step. In the second step, the contour of the visible part of a candidate tree in images with known orientation is obtained by means of ray casting and concave hull calculation. The method was tested on a deciduous, mixed and coniferous plot, whose aerial images were acquired using the 3K camera system in 2018 at Kranzberg Forest, Bavaria, Germany. Accuracies were assessed at hand of human operator generated groundtruth tree tops for the DSM as well as images with zenit angles of approximately 0,45 and 52 degrees. The resulting F1-scores, averaged over the plots, are 0.909/0.902/0.886/0.876 for the DSM/near-nadir-/oblique-/maximum-oblique-images, respectively.

Individual Tree Crown Segmentation Directly from UAV-Borne LiDAR Data Using the PointNet of Deep Learning

Accurate individual tree crown (ITC) segmentation from scanned point clouds is a fundamental task in forest biomass monitoring and forest ecology management. Light detection and ranging (LiDAR) as a mainstream tool for forest survey is advancing the pattern of forest data acquisition. In this study, we performed a novel deep learning framework directly processing the forest point clouds belonging to the four forest types (i.e., the nursery base, the monastery garden, the mixed forest, and the defoliated forest) to realize the ITC segmentation. The specific steps of our approach were as follows: first, a voxelization strategy was conducted to subdivide the collected point clouds with various tree species from various forest types into many voxels. These voxels containing point clouds were taken as training samples for the PointNet deep learning framework to identify the tree crowns at the voxel scale. Second, based on the initial segmentation results, we used the height-related gradient information to accurately depict the boundaries of each tree crown. Meanwhile, the retrieved tree crown breadths of individual trees were compared with field measurements to verify the effectiveness of our approach. Among the four forest types, our results revealed the best performance for the nursery base (tree crown detection rate r = 0.90; crown breadth estimation R2 > 0.94 and root mean squared error (RMSE) < 0.2m). A sound performance was also achieved for the monastery garden and mixed forest, which had complex forest structures, complicated intersections of branches and different building types, with r = 0.85, R2 > 0.88 and RMSE < 0.6 m for the monastery garden and r = 0.80, R2 > 0.85 and RMSE < 0.8 m for the mixed forest. For the fourth forest plot type with the distribution of crown defoliation across the woodland, we achieved the performance with r = 0.82, R2 > 0.79 and RMSE < 0.7 m. Our method presents a robust framework inspired by the deep learning technology and computer graphics theory that solves the ITC segmentation problem and retrieves forest parameters under various forest conditions.

Individual tree crown segmentation from airborne LiDAR data using a novel Gaussian filter and energy function minimization-based approach

Accurate segmentation of individual tree crowns (ITCs) from airborne light detection and ranging (LiDAR) data remains a challenge for forest inventories. Although many ITC segmentation methods have been developed to derive tree crown information from airborne LiDAR data, these algorithms contain uncertainty in processing false treetops because of foliage clumps and lateral branches, overlapping canopies without clear valley-shape areas, and sub-canopy crowns with neighbouring trees that obscure their shapes from an aerial perspective. Here, we propose an approach to crown segmentation using computer vision theories applied in different forest types. First, a dual Gaussian filter was designed with automated adaptive parameter assignment and a screening strategy for false treetops. This preserved the geometric characteristics of sub-canopy trees while eliminating false treetops. Second, anisotropic water expansion controlled by the energy function was applied for accurate crown segmentation. This utilized gradient information from the digital surface model and explored the morphological structures of tree crown boundaries as analogous to the maximal valley height difference from surrounding treetops. We demonstrate the generality of our approach in the subtropical forests within China. Our approach enhanced the detection rate of treetops and ITC segmentation relative to the marker-controlled watershed method, especially in complicated intersections of multiple crowns. A high performance was demonstrated for three pure Eucalyptus plots (a treetop detection rate r ≥ 0.95 and crown width estimation R2 ≥ 0.90 for canopy trees; r ≥ 0.85 and R2 ≥ 0.88 for sub-canopy trees) and three plots dominated by Chinese fir (r ≥ 0.95 and R2 ≥ 0.87 for canopy trees; r ≥ 0.79 and R2 ≥ 0.83 for sub-canopy trees). Finally, in a relatively complex forest park containing a wide range of tree species and sizes, a high performance was also achieved (r = 0.93 and R2 ≥ 0.85 for canopy trees; r = 0.70 and R2 ≥ 0.80 for sub-canopy trees). Our method demonstrates that methods inspired by the computer vision theory can improve on existing approaches, providing the potential for accurate crown segmentation even in mixed forests with complex structures

Multisensor Data Fusion for Improved Segmentation of Individual Tree Crowns in Dense Tropical Forests

Automatic tree crown segmentation from remote sensing data is especially challenging in dense, diverse, and multilayered tropical forest canopies, and tracking mortality by this approach is even more difficult. Here, we examine the potential for combining airborne laser scanning (ALS) with multispectral and hyperspectral data to improve the accuracy of tree crown segmentation at a study site in French Guiana. We combined an ALS point cloud clustering method with a spectral deep learning model to achieve 83% accuracy at recognizing manually segmented reference crowns (with congruence >0.5). This method outperformed a two-step process that involved clustering the ALS point cloud and then using the logistic regression of hyperspectral distances to correct oversegmentation. We used this approach to map tree mortality from repeat surveys and show that the number of crowns identified in the first that intersected with height loss clusters was a good estimator of the number of dead trees in these areas. Our results demonstrate that multisensor data fusion improves the automatic segmentation of individual tree crowns and presents a promising avenue to study forest demography with repeated remote sensing acquisitions.

Individual Tree Segmentation Based on Mean Shift and Crown Shape Model for Temperate Forest

Light detection and ranging (LiDAR) provides high-resolution geometric information for monitoring forests at individual tree crown (ITC) level. An important task for ITC delineation is segmentation, and previous studies showed that the adaptive 3-D mean shift (AMS3D) algorithm provides effective results. AMS3D for ITC segmentation has three components for the kernel profile: shape, weight, and size. In this letter, we present an AMS3D approach based on the adaptation of the kernel profile size through an ellipsoid crown shape model. The algorithm parameters are estimated based on allometry equations derived from 22 forest plots in two study sites. After computing the mean shift (MS) vector, we initialize the parameters of the ellipsoid crown shape model to derive the kernel profile size, and further tested two crown shape models for adapting the size of the superellipsoid (SE) kernel profile. These schemes are compared with two other MS algorithms with and without kernel profile size adaptation. We select the best algorithm output per plot based on the maximum F1-score. The ellipsoid crown shape model with a SE kernel profile of n = 1.5 presents the highest recall and the best Jaccard index, especially for conifers.

Tree species classification using UAS-based digital aerial photogrammetry point clouds and multispectral imageries in subtropical natural forests

Tree species composition of forest stand is an important indicator of forest inventory attributes for assessing ecosystem health, understanding successional processes, and digitally displaying forest biodiversity. In this study, we acquired high spatial resolution multispectral and RGB imagery over a subtropical natural forest in southwest China using a fixed-wing UAV system. Digital aerial photogrammetric (DAP) technique was used to generate multi-spectral and RGB derived point clouds, upon which individual tree crown (ITC) delineation algorithms and a machine learning classifier were used to identify dominant tree species. To do so, the structure-from-motion method was used to generate RGB imagery-based DAP point clouds. Then, three ITC delineation algorithms (i.e., point cloud segmentation (PCS), image-based multiresolution segmentation (IMRS), and advanced multiresolution segmentation (AMRS)) were used and assessed for ITC detection. Finally, tree-level metrics (i.e., multispectral, texture and point cloud metrics) were used as metrics in the random forest classifier used to classify eight dominant tree species. Results indicated that the accuracy of the AMRS ITC segmentation was highest (F1-score = 82.5 %), followed by the segmentation using PCS (F1-score = 79.6 %), the IMRS exhibited the lowest accuracy (F1-score = 78.6 %); forest types classification (coniferous and deciduous) had a higher accuracy than the classification of all eight tree species, and the combination of spectral, texture and structural metrics had the highest classification accuracy (overall accuracy = 80.20 %). In the classification of both eight tree species and two forest types, the classification accuracies were lowest when only using spectral metrics, indicated that the texture metrics and point cloud structural metrics had a positive impact on the classification (the overall accuracy and kappa accuracy increased by 1.49–4.46 % and 2.86–6.84 %, respectively).

A Comparison of Forest Tree Crown Delineation from Unmanned Aerial Imagery Using Canopy Height Models vs. Spectral Lightness

Improvements in computer vision combined with current structure-from-motion photogrammetric methods (SfM) have provided users with the ability to generate very high resolution structural (3D) and spectral data of the forest from imagery collected by unmanned aerial systems (UAS). The products derived by this process are capable of assessing and measuring forest structure at the individual tree level for a significantly lower cost compared to traditional sources such as LiDAR, satellite, or aerial imagery. Locating and delineating individual tree crowns is a common use of remotely sensed data and can be accomplished using either UAS-based structural or spectral data. However, no study has extensively compared these products for this purpose, nor have they been compared under varying spatial resolution, tree crown sizes, or general forest stand type. This research compared the accuracy of individual tree crown segmentation using two UAS-based products, canopy height models (CHM) and spectral lightness information obtained from natural color orthomosaics, using maker-controlled watershed segmentation. The results show that single tree crowns segmented using the spectral lightness were more accurate compared to a CHM approach. The optimal spatial resolution for using lightness information and CHM were found to be 30 and 75 cm, respectively. In addition, the size of tree crowns being segmented also had an impact on the optimal resolution. The density of the forest type, whether predominately deciduous or coniferous, was not found to have an impact on the accuracy of the segmentation.

Individual Tree Crown Segmentation in Two-Layered Dense Mixed Forests from UAV LiDAR Data

In forests with dense mixed canopies, laser scanning is often the only effective technique to acquire forest inventory attributes, rather than structure-from-motion optical methods. This study investigates the potential of laser scanner data collected with a low-cost unmanned aerial vehicle laser scanner (UAV-LS), for individual tree crown (ITC) delineation to derive forest biometric parameters, over two-layered dense mixed forest stands in central Italy. A raster-based local maxima region growing algorithm (itcLiDAR) and a point cloud-based algorithm (li2012) were applied to isolate individual tree crowns, compute height and crown area, estimate the diameter at breast height (DBH) and the above ground biomass (AGB) of individual trees. To maximize the level of detection rate, the ITC algorithm parameters were tuned varying 1350 setting combinations and matching the segmented trees with field measured trees. For each setting, the delineation accuracy was assessed by computing the detection rate, the omission and commission errors over three forest plots. Segmentation using itcLiDAR showed detection rates between 40% and 57%, while ITC delineation was successful at segmenting trees with DBH larger than 10 cm (detection rate ~78%), while failed to detect trees with smaller DBH (detection rate ~37%). The performance of li2012 was quite lower with the higher detection rate equal to 27%. Errors and goodness-of-fit between field-surveyed and flight-derived biometric parameters (AGB and tree height) were species-dependent, with higher error and lower r2 for shorter species that constitute the lowermost layer of the forest. Overall, while the application of UAV-LS to delineate tree crowns and estimate biometric parameters is satisfactory, its accuracy is affected by the presence of a multilayered and multispecies canopy that will require specific approaches and algorithms to better deal with the added complexity.

Individual Tree Crown Segmentation of a Larch Plantation Using Airborne Laser Scanning Data Based on Region Growing and Canopy Morphology Features

The detection of individual trees in a larch plantation could improve the management efficiency and production prediction. This study introduced a two-stage individual tree crown (ITC) segmentation method for airborne light detection and ranging (LiDAR) point clouds, focusing on larch plantation forests with different stem densities. The two-stage segmentation method consists of the region growing and morphology segmentation, which combines advantages of the region growing characteristics and the detailed morphology structures of tree crowns. The framework comprises five steps: (1) determination of the initial dominant segments using a region growing algorithm, (2) identification of segments to be redefined based on the 2D hull convex area of each segment, (3) establishment and selection of profiles based on the tree structures, (4) determination of the number of trees using the correlation coefficient of residuals between Gaussian fitting and the tree canopy shape described in each profile, and (5) k-means segmentation to obtain the point cloud of a single tree. The accuracy was evaluated in terms of correct matching, recall, precision, and F-score in eight plots with different stem densities. Results showed that the proposed method significantly increased ITC detections compared with that of using only the region growing algorithm, where the correct matching rate increased from 73.5% to 86.1%, and the recall value increased from 0.78 to 0.89.

Individual tree crown segmentation based on aerial image using superpixel and topological features

The individual tree crown (ITC) segmentation algorithm based on aerial images is a prerequisite for understanding tree growth, tree species competition, and biomass assessment. We combine superpixel segmentation and topological graph methods to separate the ITC effectively from aerial images. First, the aerial images of forest plots captured by drones was segmented by simple linear iterative clustering of superpixel algorithm, and the crown boundaries of aerial images were obtained by deep learning concept of holistically nested edge detection (HED) network. Second, the similarity weights of neighboring superpixels were measured by three indices, i.e., the difference in color value, the number of intersecting pixels, and the number of boundary pixels defined by HED network in the intersecting area. Finally, the minimum spanning tree topological method was adopted to generate the connected trees of aerial images at the superpixel scale, and the superpixels were merged to realize ITC segmentation depending on the calculated similarity weights. This method was tested on the aerial images of three forest plots with different stand structural features, and the accuracies of the algorithm were evaluated by comparing the results of our algorithm with field measurements. Mixed growth of the withered trees and healthy trees is in the forest plot 1, which complicates the ITC segmentation process and only achieves 86% accuracy. The forest plot 2 with same tree species and approximately sized tree crowns obtains the highest ITC accuracy of 92%. The forest plot 3 has various sizes of tree crowns and is influenced by the upper-right solar illumination, which increases the difficulty of ITC segmentation using our algorithm and obtains 87% accuracy. Overall, the method proposed has promising potential for ITC segmentation from forest aerial images, which provides a new concept based on image processing technique suitable for various types of forests.