Search

To reduce the cost of manually annotating training data for supervised classifiers, we propose an automated approach to extract training data of urban objects in six classes: buildings, fences, man-made poles, vegetation, vehicles, and low objects. In this study, two segmentation algorithms are firstly implemented to generate meaningful objects from the non-ground point cloud. Then, we generated valid strict rules to label partial RANSAC (Random Sample Consensus) planes and meaningful objects as training data. The strict rules are built upon the semantic knowledge formed by the features of geometric, eigenvalue, RANSAC plane, multidimensional slice, and relative location. The accuracy of strict rule-based (SRB) training data is higher than 98.5 % for buildings, man-made poles, vegetation, and vehicles. The accuracy of low objects and fences reaches 97.10 % and 94.99 %, respectively. Finally, we compared the performance of the KPConv and PointNET++ networks trained by SRB and manually labeled training data to evaluate the effectiveness of our training data. The KPConv overall accuracy using manually labeled and (SRB) training data are 91.5 % and 86.8 % in the Paris dataset, 95.6 % and 92.0 % in the Freiburg dataset, respectively. The experiments demonstrate that automatically labeled training data can achieve similar accuracy compared to manual labels when coupled with two deep learning networks. Therefore, SRB training data extraction can effectively deal with the problem of training data scarcity and provide significant advancements in urban point cloud classification, where manual labeling of training data remains a crucial challenge.

Land-cover mapping (LCM) at a fine scale would be useful for forest management across heterogeneous natural landscapes. However, the heterogeneity of land covers at such scales results in complex spectral and textural properties that hinder the applicability of LCM. Besides, the method suffers from, e.g. inconsistent representation of different land-cover types, lack of sufficient and balanced training samples, and instability of classifiers trained by a high number of predictor variables. Even well-known object-based classification approaches are challenged with an objective evaluation of segmentation outputs. Here we classified partially ambiguous land-cover types across heterogeneous forest landscapes in the Bavarian Forest National Park (Germany) by combining metrics from airborne light detection and ranging (LiDAR) and colour infrared (CIR) imagery data and a random forest classifier implemented in an object-based paradigm. We evaluated the segmentation results by creating a global quality score based on inter- and intra-measurements of variance and the number of segments. Selected segmentation outputs were combined with balanced training samples to run the classification algorithm based on representative blocks within the national park. The entire processing chain was implemented in an open-source domain. The final segmentation consisted of LiDAR-based height, image-based Normalized Difference Vegetation Index (NDVI) and red band, with 20 cluster seeds and a minimum segment size of 40 pixels. In the classification, the most important variables included the height of the top layer, NDVI, Enhanced Vegetation Index (EVI) and Green–Red Vegetation Index (GRVI). The average values of 500 random forest runs indicated an overall accuracy of 86.6% and an estimated Cohen’s kappa coefficient of 85.2%, with different probabilities of correct classification for land-cover classes. Mature deciduous, standing deadwood, fallen deadwood, meadow, and bare soil classes were classified most accurately, whereas classification of young coniferous, intermediate-age coniferous, mature coniferous, young deciduous, and intermediate-age deciduous were associated with the highest uncertainties. Our methodology is sufficiently robust to be applied to other similarly structured sites across temperate forested landscapes. The versatility of the method is partially guaranteed by the proposed segmentation quality score, which satisfactorily corrects under- and over-segmentation.

Accurate and updated knowledge of forest tree heights is fundamental in the context of forest management. However, measuring canopy height over large forest areas using traditional inventory techniques is laborious, time-consuming and excessively expensive. In this study, image-based point clouds produced from stereo aerial photographs (AP) were used to estimate forest height, and compared to Airborne Laser Scanning (ALS) data. We generated image-based Canopy Height Models (CHM) using different image-matching algorithms (SGM: Semi-Global Matching; eATE: enhanced Automatic Terrain Extraction), which were compared with a pure ALS-derived CHM. Additionally, plot-level height and density metrics were extracted from CHMs and used as explanatory variables for predicting the Lorey’s mean height (LMH), which was measured at 296 reference points on the ground. CHMSGM and CHMALS showed similar results in predicting LMH at sample plot locations (RMSE% = 8.54 vs. 7.92, respectively), while CHMeATE had lower accuracy (RMSE% = 13.23). Similarly, CHMSGM showed a lower normalized median absolute deviation (NMAD) from CHMALS (0.68 m) compared to CHMeATE (1.1 m). Our study revealed that image-based point clouds using SGM in the presence of high-resolution ALS-derived digital terrain model (DTM) provide comparable results with ALS data, while the performance of image-based point clouds using eATE is poorer than ALS for forest height estimation. The findings of this study provide a viable and cost-effective option for assessing height-related forest structural parameters. The proposed methodology can be usefully applied in all those countries where AP are updated on a regular basis and pre-existing historical ALS-derived DTMs are available.

The automatic extraction of single trees from remotely sensed data is approached in numerous studies, but results are still insufficient in areas of dense temperate forest. Common watershed-based algorithms using digital surface models tend to produce erroneous results in difficult constellations because the treetop determination lacks an exact criterion for smoothing. In this article, a new approach is introduced that classifies crown size in advance and uses this information as prior knowledge for single-tree extraction. Crown size is classified from texture with an improved grey-scale granulometry followed by a crown size adapted watershed segmentation of single trees. The method is applied on a large area of 10 km2 and verified on six reference plots reflecting diverse and difficult compositions. The accuracy varies between 64% and 88%, and shows an average improvement of about 30% for deciduous and mixed stands compared to a non-crown-size-dependent algorithm.

This article compares three methods for forest resource estimation based on remote sensing features extracted from Airborne laser scanning and CIR orthophotos. The estimation was made exemplarily for the total stem volume of trees for a given area, measured in cubic metres per hectare [m³ ha−1] (as one of the most important quantitative parameters to characterise a forest stand). The following methods were compared: Regression Analysis (RA), k-NN (nearest neighbour) method and a method that utilises regional yield tables, referred to as the yield table method (YT-method). The estimation of stem volume was examined in a mixed forest in Southern Germany using 300 circular inventory plots, each with a size of 452 m². Remote sensing features relating to vegetation height and structures were extracted and used as input variables in the different approaches. The accuracy of the estimation was analysed using scatter plots and quantified using absolute and relative root mean square errors (RMSE). The comparison was made for all plots, as well as for averaged plot values located within forest stands that have the same age class. On “plot level” the RMSE yielded 79.79 m³ ha−1 (RA), 81.93 m³ ha−1 (k-NN) and 81.78 m³ ha−1 (YT-method) and for the averaged values 35.75 m³ ha−1 (RA), 35.06 m³ ha−1 (k-NN) and 42.98 m³ ha−1 (YT-method). Advantages and disadvantages, as well as requirements, of the methods are discussed.

A literature review of new publications in the field of 3D data for forest applications shows that the application of airborne laser scanner data (ALS) is in the focus of research today due to its great potential for practical applications. While there is a lot of research carried out to derive forest management parameters based on laser metrics deduced from a single tree assessment or a statistical area based assessment, the delineation of stand or sub-stand units derived from laser metrics itself is a rather new approach. In order to describe stand characteristics statistical grid cell approaches or single tree approaches have been developed. The LIDAR based segmentation of stand or sub-stand units is rarely documented. This article provides information on enhanced processes to delineate stand or sub-stand units and to extract different forest information based on airborne laser derived parameters. For the stand delineation an automatic process was developed which provides a stand or sub-stand unit delineation which is according to the first results sufficiently uniform within stands and sufficiently different in species, age class, height class, structure and composition between stands in order to be distinguishable from adjacent areas. With a combined method the stand boundaries as they are established by the mapping units today, as well as sub-stand units which have in common physical characteristics indicating the same management disposition, were assessed. Finally a first validation of the forest stand unit delineation is provided, indicating the high potential of ALS data for separating stand units.

Laser scanning provides a good means to collect information on forest stands. This paper presents an approach to delineate single trees automatically in small footprint light detection and ranging (lidar) data in deciduous and mixed temperate forests. In rasterized laser data possible tree tops are detected with a local maximum filter. Afterwards the crowns are delineated with a combination of a pouring algorithm, knowledge-based assumptions on the shape of trees, and a final detection of the crown-edges by searching vectors starting from the trees’ tops. The segmentation results are assessed by comparison with terrestrial measured crown projections and with photogrammetrically delineated trees. The segmentation algorithm works well for coniferous stands. However, the implemented method tends to merge crowns in dense stands of deciduous trees.