Premium
A robust approach for determination of the macro‐porous volume fraction of soils with X‐ray computed tomography and an image processing protocol
Author(s) -
Keyes S. D.,
Boardman R. P.,
Marchant A.,
Roose T.,
Sinclair I.
Publication year - 2013
Publication title -
european journal of soil science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.244
H-Index - 111
eISSN - 1365-2389
pISSN - 1351-0754
DOI - 10.1111/ejss.12019
Subject(s) - porosity , thresholding , gravimetric analysis , segmentation , soil water , filter (signal processing) , computer science , volume fraction , histogram , soil science , volume (thermodynamics) , biological system , artificial intelligence , mathematics , materials science , environmental science , computer vision , image (mathematics) , chemistry , physics , composite material , biology , organic chemistry , quantum mechanics
Soil structure is known to govern aspects of hydration, aeration, faunal activity and root growth, which influence plant development. Industrial X‐ray computed tomography ( μ CT) has been used for over 15 years for the elucidation of soil structure, leading to a number of valuable insights. However, there is evidence of a need for more robust, repeatable methods for segmentation of significant structural features, which are essentially free from operator interference. We develop in this paper an automatable approach using a seeded region growth (SRG) algorithm for segmentation of the connected, macroporous domain of homogenized, real soils. Furthermore, we demonstrate methods for user‐independent selection of seed point and tolerance values, leading to a fully automated segmentation regime. The stability of this approach to different seed locations has been assessed, as well as the impact of X‐ray target and filter choice upon mitigation of artifacts, which are particularly detrimental to accuracy of SRG methods. Estimated porosity derived using this method has been compared with values from a gravimetric protocol and histogram thresholding approaches. It is seen that substantial differences exist in porosity quantified by such methods, with these differences probably the result of varying categorization of different porosity domains.