Multitemporal Classification Under Label Noise
Based on Outdated Maps
Alina E. Maas, Franz Rottensteiner, Abdalla Alobeid and Christian Heipke
Abstract
Supervised classification of remotely sensed images is a clas-
sical method for change detection. The task requires training
data in the form of image data with known class labels. If the
training labels are acquired from an outdated map, the classi-
fier must cope with errors in the training labels. These errors
(label noise) typically occur in clusters in object space, be-
cause they are caused by land cover changes over time. In this
paper we adapt a label noise tolerant training technique for
classification, so that the fact that changes affect larger clus-
ters of pixels is considered. We also integrate the existing map
into an iterative classification procedure to act as a priori in
regions which are likely to contain changes. Additionally we
expand the model for multitemporal data, making it appli-
cable for time series. Our experiments are based on four test
areas, including a multitemporal example. Our results show
that this method helps to distinguish between real changes
over time and false detections caused by misclassification
and thus improve the accuracy of the classification results.
Introduction
The updating of topographic databases (referred to as
maps
for brevity) is typically based on a classification of current
remote sensing imagery. Comparing the results to the map,
areas of change can be detected and the map can be updated
accordingly. Supervised classification is commonly used for
that purpose, requiring representative training data that are
typically generated in a time-consuming manual process. The
latter could be avoided by using the existing map to derive
the class labels of the training samples. As the map may be
outdated, classifiers using the class labels derived from the
map for training must take into account the fact that some of
these labels will be wrong. Nevertheless, changes typically
only affect a relatively small part of a scene, so that one can
assume the majority of the training data to be correct.
From the point of view of training the classifier, changes
will correspond to errors in the training labels if the outdated
map is used to obtain the training samples. In machine learn-
ing, such errors in the class labels of training data are referred
to as
label noise
(Frénay and Verleysen, 2014). In remote
sensing, the problem has mostly been dealt with by data
cleansing, i.e., by detecting and eliminating wrong training
samples, e.g., Radoux
et al
. (2014). An alternative is to use
probabilistic methods for training under label noise which
also estimate the parameters of a noise model. An example for
such an approach is the label noise tolerant logistic regres-
sion (Bootkrajang and Kabán, 2012), which has been applied
successfully in the context of remote sensing in (Maas
et al
.,
2016). However, the underlying noise model of that technique
assumes wrong labels to occur at random positions in the
image. This is not a very realistic model for change detection,
where changes typically occur in spatial clusters in object
space, e.g., due to the construction of a new building, and
may lead to a degradation of the classification performance.
Using the existing map has another potential benefit. As
change is usually a rare event, the existing class labels can be
seen as providing observations for the prediction of the new
class labels. This may be particularly useful in areas where
the classifier cannot distinguish the class label by the given
features well, e.g., at object borders. The corresponding prob-
abilities for the classes to be correct are related to the prob-
ability of observing a wrong label and, thus, to the parameters
of a probabilistic noise model (Bootkrajang and Kabán, 2012).
However, such an assumption also neglects the fact that
changes typically occur in compact clusters. It would typi-
cally lead to a strong bias for maintaining the class label of
the map, which is desired in areas without changes, but may
limit the prospects of detecting real changes.
The parameters of a probabilistic noise model can also
be used in a multitemporal setting. The trained parameters
describe the change between two epochs, namely the epoch of
the map creation and the epoch of recording the current data.
If there are remotely sensed data for the first epoch as well,
e.g., because the map was created by classifying remotely
sensed data, the parameter of the noise model can be used for
temporal transitions in multitemporal models like the multi-
temporal
CRF
described in (Hoberg
et al
., 2015).
In this paper, we propose a new supervised classification
method that tries to extract as much benefit as possible from
the availability of outdated information about the area to be
classified, such the existing map and the remotely sensed
data of earlier epochs. First, our method uses the class labels
from the map for training. This is achieved by expanding the
method by Bootkrajang and Kabán (2012) to take into account
that changes typically occur in clusters, which we expect to
improve the results in scenes with a large amount of change.
Second, the class labels of the existing map are included as
observations in a classification procedure based on Condition-
al Random Fields (
CRF
). We propose an iterative procedure
to reduce the impact of the observed class labels in compact
areas that are likely to have changed, which we expect to
improve the classification results in areas of weak features
without affecting the detection of real changes too much.
Third, we integrate more epochs into the
CRF
model to see if
the results improve due to use of time series.
To evaluate the inclusion of map information in training
and classification we use four datasets with different degrees
of changes. One of these datasets also contains several epochs,
which is used to evaluate the multitemporal model.
This is an extended version of (Maas
et al
., 2007). Com-
pared to the original submission, we have expanded the
methodology to a multitemporal
CRF
-based model. We have
also expanded the experimental evaluation by adding a new
multitemporal dataset of Las Vegas, which, unlike the data in
Institute of Photogrammetry and Geoinformation, Leibniz
Universität, Hannover, Germany (
Photogrammetric Engineering & Remote Sensing
Vol. 84, No. 5, May 2018, pp. 263–277.
0099-1112/18/263–277
© 2018 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.84.5.263
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
May 2018
263
