You are on page 1of 5

2015 8th International Congress on Image and Signal Processing (CISP 2015)

A New Method for Discovery of Vegetation Changes


Based on Satellite Ground Photographs
Haiyan Xiao Chuang Tong, Qiang Liu
College of Geographical Sciences Key Laboratory of Humid Sub-tropical Eco-geographical
Fujian Normal University Process (Fujian Normal University) , Ministry of Education
Fuzhou, 350007, China Fuzhou, 350007, China
xhy@fjnu.edu.cn

Abstract—The normalized difference vegetation index (NDVI) is military satellites the resolution achieves less than 0.02m.
widely used to detect ground vegetation based on remote-sensing Therefore, it has advantages to discover vegetation changes
satellite images. However, it is affected by deficiencies such as based on satellite ground photographs. In this paper, we
sensation to change in sun angle, atmospheric effects and noise propose a new method for discovery of vegetation changes
contamination. As developing of high-resolution satellite ground based on satellite ground photographs. The HSV color space is
photographs, a new approach of vegetation detection on ground used to analyze vegetation areas, and a new complexity index
photographs is proposed to automatically discover changes of is defined to separate grass fields and forests. The changes of
vegetation on a specific area. The HSV color space is used to vegetation areas are detected from differences between the
analyze vegetation areas, a new complexity index is defined to
vegetation areas on ground photographs of the same place in
identify forest or grass. The experimental results show that the
vegetation areas can be well separate from the Baidu satellite
different times. The experimental results show that the
ground photographs and the changes of vegetation areas can be proposed method can efficiently discover the changes of
discovered using a set of new definitions of vegetation changes. vegetation areas.
The proposed method has potential to use in monitoring of plant
ecology changes in agriculture, forestry human life environment. II. COLOR ANALYSIS OF VEGETATION
The new method can be used not only on the satellite ground
photographs but also on the unmanned aerial vehicle images. A. HSV Color space
Keywords- Vegetation detection; image segementation; satellite
HSV (Hue, Saturation and Value) defines a type of color
ground photographs; environment; plant ecology space, also known as the hex-cone color model shown in
Fig.1(a), which is created by Alvy Ray Smith in 1978. The
HSV color space is quite similar to the way in which humans
I. INTRODUCTION perceive color. Hue represents a color that corresponds an
The normalized difference vegetation index (NDVI) is angle from 0 to 360 degrees, as shown in Fig.1(b).
proposed by Rouse et al in 1974 [1], which is defined using
reflectance values in near infrared band and red bands because
chlorophyll in vegetation reflects wavelength in near infrared
band ((0.7ȝm) and absorbs wavelength in red or visible band
(0.645ȝm) [2]. NDVI is widely used to extract vegetation areas
from Satellite remote-sensing images [3, 4]. Since the near
infrared and red radiance of surface transmit through the
atmosphere, they are susceptible to large sources of error and
uncertainty. In order to solve the problems, researches are
trying to improve the vegetation index [5, 6]. As developing of
high resolution satellite ground photographs, a new method can
be used to extract vegetation areas from the ground photos
instead of remote-sensing images. In [7], H. Hashiba et al
(a)The hex-cone color model (b) Hue and colors
proposed an edge-based approach to detect small scale
vegetation in city region from high-resolution satellite Figure 1. HSV color spaces
photographs. As satellite ground photographs are different
from remote-sensing images and no vegetation indices can be Saturation indicates the range of grey in the color space. It
used, the image processing technology is used to identify ranges from 0 to 100%. Sometimes the value is calculated from
vegetation areas and discover their changes [8, 9]. As the 0 to 1. When the value is ‘0,’ the color is grey and when the
development of technology, the resolution of satellite value is ‘1,’ the color is a primary color. A faded color is due to
photographs becomes higher and higher. In the published a lower saturation level, which means that color contains more
ground photographs in Google maps and Baidu maps the grey. Value is the brightness of the color and varies with color
resolution is about 0.5m. For some ground photographs from saturation. It ranges from 0 to 100%. When the value is ‘0’ the

This project is funded by Teaching Reform Foundation of Fujian Normal


University (Grant No. I201502025).
*Corresponding author Haiyan Xiao: xhy@fjnu.edu.cn

978-1-4673-9098-9/15/$31.00 ©2015 IEEE 851


color space will be totally black. With the increase in the value, The histograms of trees, water and soil are shown in the second
the color space brightness up and shows various colors. row, the third row and the forth row respectively. As shown in
Fig.2, comparing hue histogram in the first column, hue
Analysis of vegetation in HSV space degrees of grass are from 70 to 140. Hue degrees of trees are
After the ground photographs are transferred to HSV color from 90 to 180. Hue degrees of water are from 130 to 230. Hue
space, a histogram analysis is applied to different type of image degrees of soil are less than 60. The hue of grass, trees and
blocks from satellite ground photographs. For example, water and soil can be shown in Fig.3.
different grass block images are put together and the HSV
histograms are conducted as shown in the first row of Fig.2.

Figure 2. HSV histogram for grass,trees, water and soil

to represent the roughness. Since saturation S has large


complexity within the grass and tree areas, the definition is
based on the saturation S. Let I represent an color image with
N×M dimension and S(x,y) represent saturation value at point
(x,y) in the image, x=1, …, N, y=1, …, M. Cs(x,y) represents
complexity index for the image pixel at point (x,y).
Figure 3. Hue of Vegetation
d = (( S (i, j ) − S (i + 1, j )) 2 + (( S (i , j ) − S (i, j + 1)) 2 , (1)
x+ p y+ p β
1 −
Cs(x, y) = 2 ¦ ¦
ed , (2)
III. SEGEMENTATION OF VEGETATION (2 p +1) i=x− p j =x− p
According to the analysis in Section II, the vegetation areas where p is the half size of the pixel block (e.g. p=6), (x,y) is the
can be separated using the hue degrees from 70 to 180. center point of the block, β is a constant, which is set to 0.03 in
However, the water areas contain hue degrees from 120 t0 180. our experiment. The index of Cs(x,y) can be regarded as an
If only hue range is used in vegetation segmentation, the average of saturation changes in the image block, and it can be
extracted areas contain some water areas. As known, water used to indicate the saturation roughness within the image
surface is smooth, but the grass and tree areas are rough and block. The Cs value is very small for water surfaces, a little bit
complicated structure. Therefore, we define a complexity index

852
large for grass areas, and much large for tree areas. Using this The grass area is represented by GH(x,y) and it can be
complexity index, the water surfaces, grass and tree areas can obtained as follows.
be separated. Some experimental results are shown in Fig.4.
GH (x, y) =VH (x, y) −TH (x, y) (6)

Suppose that there is an image I with dimension N×M. The


algorithm of the segmentation is as follows.
Algorithm of vegetation segmentation
1. Initialize matrices Cs, H, RH, VH, TH, GH.
2. Transfer image I to HSV color space.
3. For x=1 to N
4. For y=1 to M
5. Using (1) to (6), calculate Cs, H, RH, VH, TH, GH.
6. End for x
7. End for y
8. Output Cs, H, RH, VH, TH, GH.
This is very efficient algorithm. The vegetation areas, tree
areas, grass areas can be obtained in one run. For example,
a satellite ground photograph of a village is shown in Fig.5.
The segmentation of tree areas is shown in Fig.6. The
segmentation of grass areas is shown in Fig.7. The
segmentation of vegetation areas is shown in Fig.8.

Figure 4. Comparison of histograms of tree, grass and water areas

Based on the above analysis, a segmentation algorithm is


proposed. Firstly, the related pixels with hue degree from 70 to
180 are filtered out from an image. Let H(x,y) represent hue
degree at the point (x,y) of the image. RH(x,y) represents the
pixel set. We have
Figure 5. satellite ground photograph of a village
­1 , if 70 < H ( x, y ) < 180
RH ( x, y ) = ® (3)
¯0 , otherwise
In order to remove the water surface from RH(x,y),
vegetation area is represented by VH(x,y), it can be obtained as
follows.

­ 0, if Cs(x, y) < 0.3 and S(x, y) >130


VH (x, y) = ® (4)
¯RH (x, y), otherwise
The tree area is represented by TH(x,y) and it can be
obtained as follows.

­ 1, if Cs(x,y)>0.3 and RH (x,y)=1


TH (x, y) = ® (5) Figure 6. The segmentation of tree areas
¯0, otherwise

853
The proposed algorithm of vegetation segmentation is applied
to two photographs so that two segmentation images are
obtained and are represented by Vt1 and Vt2 respectively. From
time t1 to t2, vegetation is in growth for some places. In order to
describe this property Vegetation Growth Vgrth is defined as
follows.
­Vt 2 ( x, y ) − Vt1 ( x , y ), ifVt 2 ( x, y ) − Vt1 ( x , y ) > 0 (7)
Vgrth ( x, y ) = ®
¯0, ifVt 2 ( x, y ) − Vt 1 ( x, y ) ≤ 0
The area of vegetation growth can be calculated as follows.
N M
Vga = α ¦ ¦ Vgrth ( x , y ) , (8)
x =1 y =1
Figure 7. The segmentation of grass areas
where Į is scaling factor. For example, the photographs are
from Baidu map. Each pixel corresponds 0.5m. The scaling
factor is Į=0.5×0.5=0.25m2/pixel. However, in other places
vegetation may be in decrease. Therefore, Vegetation Decrease
Vdecr is defined as follows.
­Vt1 ( x, y ) − Vt 2 ( x , y ), ifVt 1 ( x, y ) − Vt 2 ( x , y ) > 0 (9)
Vdecr ( x , y ) = ®
¯ 0, ifVt1 ( x , y ) − Vt 2 ( x, y ) ≤ 0
Similarly, we have the area of vegetation decrease Vda as
follows.
N M
Vda = α ¦ ¦ Vdecr ( x, y ) (10)
x =1 y =1
Figure 8. The segmentation of vegetation areas
This area can be represented by the percentage of the whole
photograph area as follows.
IV. DISCORVERY OF VEGETATION CHANGES
N M
1
Change of vegetation is related to plant ecology. It is very
important issue in agriculture, forestry and human life
Vdp =
N ×M
¦ ¦
x =1 y =1
Vdecr ( x , y ) (11)

environment. Based on the segmentation method, changes of


The total changes of vegetation can be obtained as follows.
vegetation can be efficiently monitored. An algorithm of
vegetation change detection is proposed as follows. For Vchng ( x, y ) = abs (Vgrth − Vdecr ) (12)
example, the two photographs of the village have been taken
time t1 before the road construction and time t2 after the road These changes include both of the growth and decrease places.
construction, as shown in Fig.9 and Fig.10. In order to indicate whether the total area of vegetation is
increasing or decreasing, the change area of vegetation is
defined as follows.
Vca = Vga − Vda (13)
The negative Vca indicates decrease of the total vegetation
area. The positive Vca indicates increase of the total vegetation
area. Using equations (7), (9) and (12), the change places can
Figure 9. Image of the village at time t1.
be figured out, shown in Fig.11 to 13. Using equations (8),
(10), (11) and (13), the change areas of vegetation are
calculated and shown in the last column of Table I.

Figure 10. village at time t2


Figure 11. Vgrth: new vegetation growth in two sides of road

854
TABLE II. MONITRING VEGETATION AREAS IN A CITY

Tree Grass Vegetation


Items
Area(M2) % Area(M2) % Area(M2) %
OldCityCentre 49933.25 15.08 2971.00 0.9 52904.25 15.98
NewCityArea 70876.00 20.67 13814.25 4.03 84690.25 24.70
Old-dormitory 45371.25 13.12 3722.50 1.08 49093.75 14.20
New-dormitory 96916.00 28.02 1918.00 0.55 98834.00 28.58
This method can be used to vegetation investigation for a
city. The figure 14 and 15 are shown the samples of old city
Figure 12. Vdecr: vegetation destroyed by construction of the road center, old dormitory block, new city area and new dormitory
block. The Table II is output from the proposed algorithm.

V. I. CONCLUSION
In this paper we have analyzed the hue distribution of
vegetation in HSV color space. A complexity index on
saturation is defined to separate the tree, grass and water areas.
Based on the results of the proposed algorithms for the
segmentation of vegetation, grass, tree areas, the vegetation
growth, decrease and changes are defined. These definitions
Figure 13. Vchng: total changes caused by the road construction can be used to discover the change details of the vegetation.
The satellite ground photographs are used as examples to
demonstrate the processing results of the proposed method.
TABLE I. VEGETATION CHANGES OF THE VILLAGE This method can apply to not only satellite ground photographs
but also unmanned aerial vehicle photographs. In the further
Tree Grass Vegetation
Items study unmanned aerial vehicles can be used to take
Area(M2) % Area(M2) % Area(M2) % photographs of a specific area and the proposed method can be
Vallige t1 31605.25 13.02 80870.00 33.32 112475.25 46.34
used to discover the changes of vegetation in some specifically
Vallige t2 30425.75 12.53 70551.50 29.07 100977.25 41.60
important areas.
Growth t2-t1 936.00 0.39 505.25 0.21 935.00 0.39
Decreaset1-t2 2115.50 0.87 10823.75 4.46 4.46 5.12
Changes t2-t1 -1179.50 -0.49 -10318.50 -4.25 -11498.00 -4.74 REFERENCES
Similar to calculation of vegetation areas, we have defined [1] J. W. Rouse, R. H. Haas, J. A. Schell, D. W. Deering and J. C. Harlan,̌
growth, decrease and changes for tree area and grass area. The Monitoring the vernal advancement of natural vegetation,” NASA
results of calculation are shown in Table I. Goddard Space Flight Center, Greenbelt, MD, Final Rep., 1974.
[2] Paul J. Gibson, Introductory Remote Sensing: Principles and
As shown in Table I, changes of vegetation, tree area and Concepts,London: Routledge, 2000.
grass area are minus because the decrease of the vegetation, [3] M. P. Ponti, “Segmentation of Low-Cost Remote Sensing Images
tree and grass is larger than growth. Applying the proposed Combining Vegetation Indices and Mean Shift,” Geoscience and
algorithms to two photographs of the same place in different Remote Sensing Letters, IEEE,Vol.10,no.1,2013,pp.67-70.
times, we can obtain a result table similar to Table I and a set [4] Paul M. Mather, Computer Processing of Remotely-Sensed Images: An
of figures which can indicate the change places similar to Introduction, 3rd ed., New York: John Wiley & Sons, 2004.
Fig.11 to Fig.13. If the two photographs are not exactly in the [5] H. Q. Liu and A. Huete, “A feedback based modification of the NDVI to
same field of view, the algorithm of image registration can be minimize canopy background and atmospheric noise,” IEEE Trans.
Geosci. Remote Sensing, vol. 33, no. 2, pp. 457–465, March 1995.
used to extract the same view in the two photographs.
[6] M.H.Anvar, S.M.T.AlModarresi,"A new method to reduce the sun angle
effects and noise contamination in extracting the vegetation indices from
satellite images", proceedings of IEEE International Symposium on
Geoscience and Remote Sensing, 2007,pp.1909-1914.
[7] H. Hashiba, T.Sugimura, S.Tanaka, “Evaluation of small-scale
vegetation in city regio nfrom high-resolution satellite images with
different ground resolution”, Proceedings of IEEE International
Symposium on Geoscience and Remote Sensing. IEEE, 2004, pp. 3140 -
Figure 14. Old city center and old dormitory block 3143
[8] Y. Zheng, H. Qiu, "Mapping urban landuse types in Los Angeles using
multi-date Moderate-Resolution Imaging Spectroradiometer vegetation
image products", Proceedings of Second International Workshop on
Earth Observation and Remote Sensing Applications (EORSA), IEEE,
2012, pp.282-286.
[9] Guangxing Wang, George Gertner, Shoufan Fang, and Alan B.
Anderson, "Mapping Vegetation Cover Change Using Geostatistical
Figure 15. New city area and new dormitory block Methods and Bitemporal Landsat TM Images",IEEE Transactions on
geoscience and remote sensing, vol.42, no.3, 2004, pp.632-643.

855

You might also like