This function performs segmentation of scattered 2D data based on sampling density and location.
doSegmentation(object)A modified Intamap object which additionally includes the
list element clusters. This element is a list that contains
(i) the indices of removed points from observations; (ii)
the indices of the clusters to which the remaining observation
points are assigned
and (iii) the number of clusters detected.
list element added to the original object containing the segmentation results.
Indices of removed points.
Index array identifying the cluster in which each observation point belongs.
Number of clusters detected.
An Intamap type object containing the element (list)
observations, which includes the coordinates of the
observation locations
A. Chorti, Spiliopoulos Giannis, Hristopulos Dionisis
This function performs segmentation of scattered 2D data
based on sampling density and location. Let us assume that
No is the number of observation locations. If No< 200,
then a single cluster is returned.
(1) The segmentation algorithm first removes isolated distant points, if there are any, from the observation locations. Points $(xi,yi)$ are characterized as 'isolated' and 'distant' if they satisfy the following conditions : $abs(xi-mean(x)) > 4 *std(x) or abs(yi-mean(y)) > 4 *std(y)$ and distance from closest neighbor $> sqrt((std(x)/2)^2+(std(y)/2)^2)$. After the first step the size of the original dataset is reduced to N (N= No - isolated points) points.
(2) A sampling density matrix (lattice) consisting of N cells that cover the study area is constructed. Each cell is assigned a density value equal to the number of observation points inside the cell. In addition, each observation point is assigned the sampling density value of the containing cell.
(3) Unsupervised clustering edge detection is used to determine potential cluster perimeters.
(4) Each closed region's perimeter is labeled with a different cluster (segment) number.
(5) All observation points internal to a cluster perimeter are assigned to the specific cluster.
(6) Each cluster that contains fewer than 50 observation points is rejected.
(7) The observation points that have not initially been assigned to a cluster and those belonging to rejected (small) clusters are assigned at this stage. The assignment takes into account both the distance of the points from the centroids of the accepted clusters as well as the mean sampling density of the clusters.
Note: The No< 200 empirical constraint is used to avoid
extreme situations in which the sampling density is concentrated
inside a few cells of the background lattice, thereby inhibiting the
edge detection algorithm.
[1] D. T. Hristopulos, M. P. Petrakis, G. Spiliopoulos, A. Chorti (2009). Non-parametric estimation of geometric anisotropy from environmental sensor network measurements, StatGIS 2009: Geoinformatics for Environmental Surveillance Proceedings (ed. G. Dubois).
library(gstat)
data(walker)
# coordinates(walker)=~X+Y
object=createIntamapObject(observations=walker)
object=doSegmentation(object)
print(summary(object$clusters$index))
Run the code above in your browser using DataLab