Short-Interval Monitoring of Land Use and Land Cover Change Using a Time Series of RADARSAT-2 Polarimetric SAR Images - Space-Time Integration in Geography and GIScience - page 355

After changed objects were detected using CVA, PCC was performed on the
changed objects to determine the type of changes. Many classification methods
for PolSAR data have been explored (Chen et al. 1996 ; Barnes and Burki 2006 ;
Lee et al. 1999 ; Shimoni et al. 2009 ). However, so far most of the classification
methods are pixel-based. These methods are prone to be affected by speckles
in PolSAR images and are hard to utilize textural and spatial information of
PolSAR images. Moreover, they cannot take fully use of polarimetric information
of PolSAR data for LULC classification. Qi et al. ( 2012 ) proposed a new classifica-
tion method, which integrates polarimetric decomposition, PolSAR interferometry,
object-oriented image analysis, and decision tree algorithms, for the classification
of PolSAR images. The results show that the proposed method can achieve much
higher accuracy than conventional pixel-based classification methods. Polarimetric
information has significant implications for identifying different vegetation types
and distinguishing between vegetation and urban/built-up. Object-oriented image
analysis is helpful in reducing the effect of speckle in PolSAR images by imple-
menting classification based on image objects, and the textural information extracted
from image objects is helpful in distinguishing between water and barren land.
The decision tree algorithm is efficient in selecting features and implementing
classification. In this study, the classification of RADARSAT-2 images was per-
formed using the method proposed by Qi et al. ( 2012 ). First, different polarimetric
techniques were used to extract polarimetric parameters, and the extracted polari-
metric parameters were then combined with the elements of the coherency matrix
to form a multichannel image. Second, different features were extracted from the
multichannel image for image objects to support LULC classification. Third, a
decision tree algorithm was used to select features and create a decision tree for
LULC classification. Finally, the final LULC classification was implemented using
the constructed decision tree. After the independent classification of RADARSAT-2
PolSAR images, PCC was performed on the changed objects to determine the type
of change.

19.4 Results and Discussion

Monthly LULC changes detected using the time series of RADARSAT-2
PolSAR images are shown in Fig. 19.4 . Detection accuracy, false alarm rate,
and overall error rate are commonly used statistics for evaluating change
detection results. The detection accuracy is the percentage of correctly labeled
“change” samples. The false-alarm rate is the percentage of erroneously
labeled “no-change” samples. The overall error rate is the percentage of erroneously
labeled validation samples. These three statistics were calculated using the
confusion matrix that was determined using the validation samples of change
and no-change. As shown in Fig. 19.5 , high accuracies were achieved for monthly
LULC change detection by using the time series of RADARSAT-2 PolSAR images.
The average overall error rate, average detection accuracy, and average false alarm
rate were 1.97 %, 91.29 %, and 1.37 % respectively.