Snímka 1 - InSAR.sk

Katedra geodetických základov
Monitoring Landscape Changes Using Satellite Radar Imagery
Introduction
Edge Detection
All Synthetic Aperture Radar (SAR) images are inherently degraded by granular noise known as radar speckle. Speckle is caused by random
Edge detection and edge extraction was achieved by MATLAB Image Processing Toolbox. For the purpose of edge detection and extraction
constructive and destructive interference from multiple backscatters and causes difficulties for analysis, classification and image
from georeferenced SAR images in GeoTiff format a simple MATLAB function was proposed. The function is called Saredge. This function
interpretation for various RS applications, including object detection. Before using SAR images, the very first step is to reduce the effect of
enables reading images in GeoTiff format (both optical and SAR), perform edge detection on greyscale images and export map coordinates
speckle. One of the common methods which is used to reduce speckle noise is spatial filtering. For the purpose of this research, five speckle
of every single pixel belonging to the detected edge in text file format. Saredge was gradually applied to all test SAR images previously
filters were quantitatively evaluated. The filter showing the best results was latter used to reduce the speckle effect on SAR images used for
smoothed by Gamma MAP filter. The attempt to detect edges in Sentinel-1 imagery met with the poorest results and the edges were
detection of landscape elements. Radar imagery was provided by the TerraSAR-X (TSX) mission and by the new European Space Agency
incomparable with the TSX dataset. The main drawback of the Sentinel-1 imagery was the lower resolution then the TSX imagery resolution.
initiative, the Sentinel-1 mission. The obtained results were compared with in situ measurements using global navigation satellite systems
The best and only usable results were achieved on the TSX SAR images from December 2008 and June 2013.
(GNSS). The results have been presented by filtered images, statistical tables and diagrams.
Study Area
The area of interest (AOI) is situated northeast
(NE) of Bratislava next to class II road 503
between the D1 highway and the city of Pezinok.
This location was selected due to multiple
available SAR images of this area and for its
good accessibility via the D1 highway. The study
(b)
(a)
area is located in the Danubian Lowland at the
Figure 4: Detected edges on the TSX image from December 3, 2008 (a) and June 2, 2013 (b).
intersection of Danubian Hills and Danubian Flat.
The AOI's center geographic coordinates are
48° 14´ 52´´ N and 17° 19´ 20´´ E in World
In Situ Measurements
geodetic system 1984. According to the territorial
For evaluation of edge detection accuracy it was necessary to perform in situ measurements in the AOI. The borders of the AOI were mapped
and administrative system of Slovakia it is
with Real Time Kinematic (RTK) GNSS method on March 20, 2015 with three TOPCON GNSS receivers. Two receivers TOPCON Hiper
situated in Bratislava Region, Pezinok District
GD were used in base/rover setup and the receiver TOPCON Hiper II was connected to the Slovakian Spatial Observation Service (SKPOS)
and it falls under the cadastral areas of Slovenský
which represents active geodetic controls in Slovakia. The boarder was mapped with approximately 50 meter step. Single types of borders
Figure 1: Area of interest
(Adapted from Google Earth)
Grob and Viničné.
are shown in Figure 5 from (a) to (d).
Data used
TSX radar imagery for this master's thesis was kindly provided by the German Aerospace Centre (DLR) within the project ID LAN1583:
Object Recognition Based on High-Resolution Radar Imagery [1] and the project LAN2833: Utilization of the High Spatio-temporal
Resolution of TerraSAR-X Observations for Recent Ground Deformation Monitoring and SAR Image Processing [2]. Sentinel-1 dataset was
provided by the Sentinel-1 Scientific Data Hub of the European Space Agency [3].
(a)
(b)
(c)
(d)
Figure 5: Representative images of the NW border (a), SW border (b),
Speckle reduction
SE boarder (c), NE boarder (d).
Speckle forms a main obstacle to analyse, interpret and classify SAR images for various RS applications. One of the common methods which
is used to reduce speckle noise is spatial filtering. This thesis evaluates the effect of Frost, Lee, Refined Lee, Gamma MAP and Perona-Malik
equation based filter. To assess the capability of the tested filters to remove speckle noise, and their effectiveness in successfully preserving
Comparison with In Situ Measurements
the real structure of the scene backscatter, ENL, SSI, SMPI, EEI and FPI performance measures were used. The main criterium was to reduce
The points observed by GNSS served as a flawless reference for the detected edges. The detected edges consist of many single points. These
speckle noise and to preserve sharpness of the edges as much as possible.
points represent pixels that were identified as edges. The nearest detected points to the in situ observed points were selected and the
differences between them were computed. Note that the SAR images were acquired in 2008 and 2013 while the in situ measurements took
place in 2015. Since 2008 the boundaries of the AOI may have slightly changed due to agricultural activity. The differences between GNSS
observations and edges detected on the TSX imageries are shown in the following graphs.
NE Border
NW Border
5
20
4
(c)
15
2
Difference (m)
(b)
(a)
Difference (m)
3
1
0
-1
-2
10
5
-3
0
-4
-5
1
3
5
7
9
11
13
15
17
19
21
23
25
27
-5
29
1
3
5
7
9
11
13
Point ID
(f)
(e)
20
Refined Lee (d), Perona-Malik (e) and Frost filter (f).
15
19
21
23
25
27
29
31
33
TSX December 3, 2008
TSX June 2, 2013
SE Border
Figure 2: Original image (a) and smoothed images by Lee (b), Gamma MAP (c),
17
Point ID
TSX December 3, 2008
TSX June 2, 2013
(d)
15
SW Border
15
10
preservation ability according to EEI and FPI were Refined Lee and Perona-Malik. However applying the edge detector and comparing the
detected edges revealed some interesting new facts. Best edge detection results were achieved on images that have been previously smoothed
by Gamma MAP filter. For this reason, the final decision was made in favour of the Gamma MAP filter. The conclusion is that there is no
single best filter for every possible scenario. The most suitable filter depends on many parameters and variables e. g. SAR viewing geometry,
5
Difference (m)
According to ENL, SSI and SMPI measures the best speckle reduction was achieved by Gamma MAP filter. The filters with the best edge
Difference (m)
10
0
-5
-10
5
0
-5
-10
-15
-15
-20
-25
-20
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47
polarization, land cover, prerequisites for certain applications, etc.
Point ID
TSX December 3, 2008
TSX June 2, 2013
1
3
5
7
9
11
13
15
17
19
21
23
Point ID
25
27
29
31
33
TSX December 3, 2008
TSX June 2, 2013
Conclusion
J
The results imply that edge detection on SAR images can't compete with automatic detection on optical remote sensing images due to the
distortive effect of the speckle noise. The results are still acceptable for certain applications like agriculture, forestry, monitoring large scale
(a)
(b)
events such as earthquakes, floods or oil spill. In addition SAR imagery carries more information that can be utilized besides object detection
in different applications such as radar interferometry or radar polarimetry.
References
[1] PAPČO, J., BAKOŇ, M.: Object Recognition Based on High Resolution Radar Imagery. German Aerospace Center (DLR),
(c)
TSX-Archive-2012 project ID LAN1583 2012.
(d)
[2] PAPČO, J.: Utilization of the High Spatio-Temporal Resolution of TerraSAR-X Observations for Recent Ground Deformation Monitoring
Figure 3: Smoothed images by Gamma MAP filter (a), Perona-Malik filter (b), and detected edges on
Gamma MAP smoothed image (c) and Perona-Malik smoothed image (d).
and SAR Image Processing. German Aerospace Center (DLR), TSX-Archive-2015 project ID LAN2833 2015.
[3] Sentinel-1 Scientific Data Hub: https://scihub.esa.int/ (2015-03-15).
Pedagóg: Ing. Juraj Papčo, PhD.
Akademický rok: 2014/2015
Poslucháč: Bc. Jakub Vanko
Študijný program: Geodézia a kartografia
Diplomová práca