Premium
Investigation of the performances of advanced image classification‐based ground filtering approaches for digital terrain model generation
Author(s) -
Yilmaz Volkan
Publication year - 2021
Publication title -
concurrency and computation: practice and experience
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.309
H-Index - 67
eISSN - 1532-0634
pISSN - 1532-0626
DOI - 10.1002/cpe.6219
Subject(s) - support vector machine , computer science , artificial neural network , terrain , artificial intelligence , process (computing) , class (philosophy) , point (geometry) , pattern recognition (psychology) , image (mathematics) , data mining , machine learning , mathematics , geography , geometry , cartography , operating system
The majority of the ground filtering techniques proposed so far use several user‐defined parameter values. Since no standard protocols exist to define these parameters, obtaining the optimum filtering performance is very hard, especially in large‐extent areas with abrupt topography changes. This, of course, reveals the necessity of some more efficient strategies to ease the ground filtering process in such areas. Utilizing classified images for ground filtering purpose may be of help to achieve this. Hence, this study, for the first time in the literature, investigated the performances of the state‐of‐the‐art machine learning algorithms maximum likelihood (ML), artificial neural network (ANN), support vector machines (SVM), and random forest (RF) in ground filtering of a UAS‐based point cloud. The used approaches were based on the assignment of the points corresponding to the ground‐related classes to the ground class. Evaluations showed that the SVM‐based ground filtering approach achieved the optimum filtering result. The SVM‐, ML‐, RF‐, and ANN‐based ground filtering methods achieved the total errors of 13.2%, 16.4%, 19.6%, and 21.9% in the test site, respectively.