Image Processing and Pattern Recognition Applied to Soil Structure

Luo, Daisheng (1995) Image Processing and Pattern Recognition Applied to Soil Structure. PhD thesis, University of Glasgow.

Full text available as:
[thumbnail of 11007664.pdf] PDF
Download (10MB)


This thesis represents a collaborative research between the Department of Electronics & Electrical Engineering and the Department of Civil Engineering, University of Glasgow. The project was initially aimed at development of some theories and techniques of image processing and pattern recognition for the study of soil microstructures. More specifically, the aim was to study the shapes, orientations, and arrangements of soil particles and voids (i.e. pores): these three are very important properties, which are used both for description, recognition and classification of soils, and also for studying the relationships between the soil structures and physical, chemical, geological, geographical, and environmental changes. The work presented here was based principally on a need for analysing the structure of soil as recorded in two-dimensional images which might be conventional photographs, optical micrographs, or electron-micrographs. In this thesis, first a brief review of image processing and pattern recognition and their previous application in the study of soil microstructures is given. Then a convex hull based shape description and classification for soil particles is presented. A new algorithm, SPCH, is proposed for finding the convex hull of either a binary object or a cluster of points in a plane. This algorithm is efficient and reliable. Features of pattern vectors for shape description and classification are obtained from the convex hull and the object. These features are invariant with respect to coordinate rotation, translation, and scaling. The objects can then be classified by any standard feature-space method: here minimum-distance classification was used. Next the orientation analysis of soil particles is described. A new method, Directed Vein, is proposed for the analysis. Another three methods: Convex Hull, Principal Components, and Moments, are also presented. Comparison of the four methods shows that the Directed Vein method appears the fastest; but it also has the special property of estimating an 'internal preferred orientation' whereas the other methods estimate an 'elongation direction'. Fourth, the roundness/sharpness analysis of soil particles is presented. Three new algorithms, referred to as the Centre, Gradient Centre, and Radius methods, all based on the Circular Hough Transform, are proposed. Two traditional Circular Hough Transform algorithms are presented as well. The three new methods were successfully applied to the measurement of the roundness (sharpness of comers) of two-dimensional particles. The five methods were compared from the points of view of memory requirement, speed, and accuracy; and the Radius method appears to be the best for the special topic of sharpness/roundness analysis. Finally the analysis and classification of aggregates of objects is introduced. A new method. Extended Linear Hough Transform, is proposed. In this method, the orientations and locations of the objects are mapped into extended Hough space. The arrangements of the objects within an aggregate are then determined by analysing the data distributions in this space. The aggregates can then be classified using a tree classifier. Taken together, the methods developed or tested here provide a useful toolkit for analysing the shapes, orientation, and aggregation of particles such as those seen in two-dimensional images of soil structure at various scales.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Additional Information: Adviser: James E S MacLeod
Keywords: Electrical engineering, Soil sciences
Date of Award: 1995
Depositing User: Enlighten Team
Unique ID: glathesis:1995-74745
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 27 Sep 2019 16:41
Last Modified: 27 Sep 2019 16:41

Actions (login required)

View Item View Item


Downloads per month over past year