In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
of multi spectral, hyperspectral and SAR data. Nevertheless, due
to space constraints, we will focus the attention solely on a
representative change-detection problem referring to an intense
farming area experiencing several land-cover transitions located
in Barrax, a village close to Albacete (Castilla-La Mancha,
Spain). In particular, available data consist in: two Landsat-5
Thematic Mapper (TM) images acquired on 15 th July 2003 and
17 th July 2004, respectively (as generally done in the literature
among the 7 spectral bands we did not consider the low-
resolution band associated with the thermal infrared channel);
one PROBA CHRIS image (composed by 63 spectral bands
with centre wavelengths from 400 to 1050 nm) acquired on 16 th
July 2004; and one Envisat ASAR alternate polarization (VV,
HH) image (despeckled using a 3x3 Gamma filter) acquired on
18 th July 2004. All of them have been properly co-registered to
a common spatial geometry of 30 meters and a study area of
512 x 512 pixels has been selected. July 2003 will be referred
to as t\, whereas July 2004 will be referred to as h (no signifi
cant changes indeed occurred between 16 th and 18 th July 2004).
From the original available images we derived the following
three datasets (using the stacked vector approach for multi
source data fusion (Richards and Jia, 2006)) composed by: i)
the 6 Landsat TM bands at both t\ and h (i.e., A=A=6)
[Dataset I]; ii) the 6 Landsat TM bands at t\ and the merger of
the 2 Envisat ASAR backscattering intensity images with the 6
Landsat TM bands at h (i.e., A =6 and A =8) [Dataset II];
and iii) the 6 Landsat TM bands at t\ and the 63 PROBA
CHRIS bands at h (i.e., A =6 and A =63) [Dataset III],
However, it is worth noting that our objective is not to seek for
a set of features at both dates which could be more effective for
solving the investigated problem, but rather to demonstrate that
the presented method is even capable of effectively handling
different types of data at the two times.
As described in Section 3, the user is required to set the number
of Gaussian kernels K to be employed for approximating the
PDFs. Hence, in order to understand how significant the selec
tion of this free parameter is, we performed a series of experi
ments varying K from 20 to 120 with steps of 10.
According with a variety of experiments on toy and real data
sets we fixed f = 10" 4 and /3=10 2 .
In all the trials we employed the ¿-means clustering for initializ
ing both centres and variances of kernel functions. Neverthe
less, the very high complexity of the algorithm (i.e., approxi
mately 0(iV z>(X+1) logA r ), where N and D represent the num
ber of samples to be clustered and their dimensionality, respec
tively (Inaba et al., 1994)), prevented us from using all the pat
terns of each investigated image at a time, as this would have
required a very high computational burden. In order to over
come this limitation, for each image we ran the ¿-means algo
rithm on a random subset containing one third of the total
amount of samples. However, as this might affect the final
change-detection accuracies of the proposed technique, for each
number of considered kernels K , we performed 10 different
trials running each time the ¿-means clustering on a different
random subset. Moreover, we finally also combined the 10
resulting change-detection maps through a majority voting en
semble.
For validating the potentialities of the presented method, we
compared the results with those obtained by supervised PCC. In
particular, we considered ML and SVM fully-supervised classi
fiers trained by exploiting a complete ground truth for all the
land-cover classes characterizing each considered date. ML is a
simple yet generally rather effective statistical classifier, which
does not require the user to set any free parameter (Richards
and Jia, 2006). SVM are advanced state-of-art classifiers, which
proved capable of outperforming other traditional approaches
(Cristianini and Shawe-Taylor, 2000). For the selection of the
two free parameters (i.e., a penalization parameter and the vari
ance of considered Gaussian kernels) we employed a 10-fold
cross-validation strategy (Duda et al., 2000).
Available prior knowledge has been used for defining regions
of interest composed on the whole by 21941 pixels whose
ground truth was known at the two times, respectively. 10 land-
cover classes have been considered at t\, whereas 9 have been
taken into consideration at ti. At both dates, from all the avail
able labeled samples we defined two spatially-disjoint training
sets (see Table 1). This means that there is no overlapping be
tween training samples at U and h . All of them have been used
for training both the ML and SVM supervised classifiers at each
time.
Land-cover class
t\ (July 2003)
h (July 2004)
alfalfa
2031
634
bare soil
2585
-
com
1737
2664
garlic
101
302
grasslands
-
42
onions
213
220
poppy
336
-
potatoes
208
283
spring crops
-
3318
stubble
2416
2247
sugar beet
365
-
sunflower
-
449
wheat
369
total
10361
10159
Table 1. Number of spatially-disjoint training samples
considered at the two dates.
Change-detection results have been evaluated (over those sam
ples whose land-cover class is known at both dates) in terms of:
percentage overall accuracy OA% (i.e., the percentage of sam
ples correctly identified as both changed or unchanged over the
whole number of samples), and kappa coefficient of accuracy
(which also takes into consideration errors and their type)
(Richards and Jia, 2006).
Among the different land-cover transitions occurred between
the two dates, here we take into consideration (one at a time)
two of them, namely “bare soil to spring crops” and “alfalfa to
com” (experienced by 4583 and 1035 pixels, respectively, over
the whole available 21941 whose ground truth was known at
both times).
In our trials, we empirically experienced that a common range
for K resulting in average high detection accuracies spans
from 60 to 80. When instead nearing the lower or the upper
bound of the considered interval, performances tend to vary
depending on the specific land-cover transition of interest. Ac
cordingly, in Table 2 and Table 3 we show the results obtained
with the proposed technique for K = 40, 60, 80, 100. In par
ticular, we report the median over the 10 realizations with dif
ferent ¿-means clustering initialization. Moreover, also accura
cies finally obtained with the majority voting ensemble (de
noted as PSCD^), as well as those obtained by supervised PCC
using ML and SVM (denoted as PCC^ and PCC 5[ / M , respec
tively) are presented.
While investigating the “bare soil to spring crops” transition
with the proposed PSCD technique, from all the training pixels
reported in Table 1, we considered the only 2585 available for
bare soil at t\ and the only 3318 spatially-disjoint available for