Browsing by Author "Heinaro, Einari"

Filter results by typing the first few letters
Now showing 1 - 3 of 3
  • 3D-based tree detection in urban areas with airborne laser scanning
    (2018-12-10) Heinaro, Einari
     Insinööritieteiden korkeakoulu | Master's thesis
    Urban trees are a valuable resource, as they affect the climate of cities, provide aesthetic and recreational value and maintain the biodiversity in the cities. Thus, cities and municipalities often keep tree registers for monitoring the condition of urban trees. Updating these registers with field measurements is laborious and time-consuming and thus there is a need to automate the updating process. Airborne laser scanning (ALS) provides an efficient option for the tree registry updating process, as it enables acquiring detailed three-dimensional (3D) data from large areas at once. This thesis studied the ALS-based urban tree monitoring process starting from the extraction of vegetation points from the ALS point cloud and ending in the detection and delineation of individual trees. One method was developed and tested for removing falsely classified vegetation points from a pre-classified point cloud. In addition, three individual tree detection (ITD) methods were developed and tested. Method 1 detected trees using region growing, method 2 divided the point cloud into horizontal slices and delineated the trees by merging clusters of each slice, and method 3 detected trees from a surface model. The method for removing falsely classified vegetation points produced varying results. Some false vegetation points originating from flat man-made objects were detected rather well, whereas the detection of vertical and narrow objects was very poor. In conclusion, the method by itself was not sufficient, but it could be used as a part of the vegetation point extraction process. The accuracy of the ITD methods was assessed by calculating the tree detection rates with distance thresholds ranging from 0.5 m to 6 m. The distance threshold determined the maximum locational difference between a delineated tree and a reference tree for these trees to be matched. The detection rates of ITD methods 1,2 and 3 ranged from 0.09 to 0.79, 0.14 to 0.79 and 0.11 to 0.50, respectively. The study showed that none of the tested methods perform sufficiently well by themselves, but a combination of methods 1 and 3 could be a suitable method for detecting urban trees.
  • Jäätiköiden muutosten kartoittaminen ilmastonmuutoksen seuraamiseksi
    (2014-12-12) Heinaro, Einari
     Insinööritieteiden korkeakoulu | Bachelor's thesis
  • Remote sensing-driven biodiversity index for forest ecosystems
    (2024-11-18) Polvivaara, Antti
     School of Engineering | Master's thesis
    Biodiversity is essential for ecosystem resilience and stability. Remote sensing of forest biodiversity is a cost-effective method for large scale ecosystem mapping. Biodiversity is measured through functional, structural, and compositional metrics of the forests that can be captured through the use of different remote sensing sen- sors. Validating these metrics through ground-truth data allows the use of these traits to be used as proxies for biodiversity. This study aimed to produce a method for modelling biodiversity using a biodi- versity index that corresponds to the species richness of the forest. This is done with field reference data consisting of species information of bryophytes, lichens and polypores. Abundance and occurrence of these species constitute the metric for the biodiversity index that is in turn modelled using remote sensing derived features. I also had access to a validation dataset where the performance of the model could be observed in natural like forest sites. The remote sensing data used in this study consists of multitemporal aerial im- ages, LiDAR point clouds and auxiliary data products of topographic features and forest resource information that are freely available and provided by different Finn- ish organizations through various API:s. Structural features of forests were calcu- lated from LiDAR point clouds, indicator features (dead trees and aspens) were pro- duced using state-of-the-art convolutional neural networks and topographical fea- tures were off the shelf products from different service providers. The spatial unit for data-analysis and derived metrics was chosen to be a 16 m x 16 m grid which is a common unit for producing information in Finnish forestry organizations and is a compromise of desired accuracy and variation of forest and biodiversity metrics. The chosen model was a random forest regressor and its performance was quan- tified by splitting the data to training and tests (80/20%, respectively). The model was then evaluated by comparing Root Mean Square Error (RMSE) and R2 scores of different model permutations. The model with lowest RMSE and highest R2 score was the model with auxiliary forest resource information, and it outperformed the model with in-house metrics only by ~4% increase in R2 score. The topographical and indictor features were also shown to bring valuable information to the model. Results of the study imply that derivation of biodiversity index in scale of Finland using freely available data is possible.