|
Land Use/Land Cover
|
A study on land use/cover classification with textural analysis using Multi-Temporal JERS-1 (SAR) L- band in arid and semi-arid areas.(A case study in Northeastren Jordan)
Methodology
Figure 2 shows the general methodology of this study. A subset of each of the of the JERS- 1 SAR L-band
images acquired in October 1993, and March 1995, respectively, covering the studied area which is
a part of the northeastern Jordan, including rewaished city (Fig. 1) were used. The digital images were
geometrically and radiometrically calibrated to each other to facilitate their comparison. An important
component to SAR data is the speckle noise reduction; a characteristic feature of SAR imagery is the
presence of speckle (Ulaby et al. 1986). This effect originates from the way in which the electromagnetic
radiation is produced by the sensor and how it interacts with the target. Consequently, speckle reduction
is a useful first step in the processing of SAR data. Images were georeferenced to UTM E012 of WGS
1984.
The textural features generated images (in this study) were those used in previous studies (e.g. Takashi
et al. 2001, Ulaby et al. 1986, and Holmes et al. 1984). The textural features used were Contrast, Energy,
Entropy and Homogeneity. The displacement vectors to obtain the Gray Level Co-occurrence Matrix
(GLCM) needed to calculate the textural features were (1,0þ), ( 1,45þ), (1,90þ) and (1,135þ) in (d,þ), where
d denotes the distance and þ denotes the angle. Each textural feature was obtained by averaging four
values calculated from GLCMs corresponding to the above four directions, the four textural features were
calculated at each pixel in each text ure image.
The original SAR data, for both dates were geometrically calibrated to each other; then smoothed images
were generated by using Lee filtering from the original SAR data, and used to drive the GLCM textural
features images, both original and smoothed images were used as in put for maximum likelihood
classifier (Fig.2).
The GLCM textural features images (Contrast, Energy, Entropy and Homogeneity) were generated
using 3,13,23,33 and 45 windows, and were used also as in put for the maximum likelihood classifier
with Principal Component Analysis (PCA). PCA is a statistical technique that rotates the axes of a
multi dimensional image space in the direction of maximum variance. The components or axes
generated by this rotation are simple linear combinations of the original image data, or orthogonal to
each other and, thus, have no mathematical relation (i.e., their correlation coefficient is approximately
zero) (Kwarteng and Chavez, 1997).
Classification of SAR L- band pixels were clustered into five basic land use/cover classes using (Sato-Tateishi,
land cover legend, personal communication, 2002), these classes are: 1- spares
herbaceous/shrub, 2- hardpan, 3- gravels, stones and boulders, 4- bare rock and 5- bare soils and
other unconsolidated materials.
Results
Figure 3 shows the results of the classification accuracy for the original SAR images , the smoothed
images and the textural features images. The results indicate that for the original images the over all
accuracy is unsatisfactory; where it is 53% and 56% for 1993 and 1995, respectively, and for the
smoothed images the speckle noise is reduced and the over all accuracy is a little bit improved , where
it is 63% and 69% for 1993 and 1995, respectively.
Figure 3: Classification Accuracy Results.
If the over all accuracy of the textural features images is compared, the contrast images give the best
accuracy for both dates, where 75% and 76% had been obtained for 1993 and 1995, respectively, the
energy images second with 73% and 70% for 1993 and 1995, respectively, the entropy images third with
67% and 63% for 1993 and 1995, respectively and the Homogeneity images give the worst over all
accuracy results, where 66% and 62% had been obtained for 1993 and 1995 respectively.
The combination of all textural features images give the best over all accuracy results where 89% and
93% had been obtained for 1993 and 1995, respectively (Figures 4 and 5)
Figure 4: Supervised classification results(SAR 1993)
Figure 5: Supervised classification results (SAR 1995)
Conclusion
The objective of this study was to investigate the performances of the textural features images. It was
found that the textural features generated images improved the classification accuracy, for the arid and
semi-arid areas, more than that of the original SAR images and their smoothed images. The combination
of all textural features images give the best over all accuracy results.
Nevertheless, textural features analysis is important because it can improve the classification accuracy in
arid and semi-arid areas.
Reference
- Holmes, Q. A., Nuesch, D. R., and Schuchman, R. A., 1984: Textural analysis and real-time
classification of sea-ice types using digital SAR data. IEEE Transaction on Geoscience and Remote
Sensing, GE- 22, pp. 113-120.
- Kurosu, T., Fujita, M., and Chiba, K., 1997: The identification of rice fields using multi-temporal ERS-1
C-band SAR data. International Journal of Remote Sensing, 18, pp. 2953-2965.
- Kurosu T., Yokoyama, SHI., and Fujita, M., 2001: Land use classification with textural analysis and
the aggregation technique using multi-temporal JERS-1 L -band SAR images. International Journal of
Remote Sensing, 22, No. 4, pp. 595 -613.
- Kwarteng, A. Y., and Chavez, P. S., JR., 1997. Change detection study of Kuwait City and environs
using multi -temporal Landsat Thematic Mapper data. INT. J. Remote Sensing, vol.19, No. 9,pp.
1651- 1662.
- National Information System (NIS). 1999. The official web site of the Hashemite Kingdom of Jordan
database. http://www .nis.gov.jo and http://www.dos.gov.jo.
- The Hashemite Kingdom of Jordan. 1991. National Strategy for Jordan. Ministry of Municipal and
Environment, Amman -Jordan.
- Ulaby, F. T., Kouyate, F., Brisco, B., and Williams, T. H. L., 1986: Textural information in SAR
images. IEEE Transaction on Geoscience and Remote Sensing, GE- 24, pp. 235 -245.
|
|
|
|
|
