use of the classified satellite imagery for updating topographic maps

use of the classified satellite imagery for updating topographic maps

PRACE INSTYTUTU GEODEZJI I KARTOGRAFII
Tom XLII, zeszyt 92, 1995
ZBIG NIEW BO CH EN EK
USE OF THE CLASSIFIED SATELLITE IMAGERY
FOR UPDATING TOPOGRAPHIC MAPS
1. INTRODUCTION
The presented work was a joint Polish-Belgian study, carried out within
agreem ent established between the Institute of Geodesy and Cartography in
W arsaw and SURFACES Laboratory, University o f Liege. The aim o f this
study was to assess and compare different methods for updating topographic
maps, using various types of satellite imagery. Two test areas were selected
for research works: Warsaw area as a Polish test site and Liege area as a
Belgian test site. Both test areas were analysed, using similar m ethods o f
data processing and results evaluation, in order to make comparable conclusions.
Two types of satellite data, i.e. SPOT multispectral and panchromatic imagery,
as well as KOSMOS images were used in this study. The article presents
works concentrated on applying the classified satellite imagery for updating
topographic maps; it includes the applied methodology and discussion o f the
results.
2. MATERIALS AND METHODS
In order to perform multispectral analysis of SPOT data, three im ages
covering W arsaw area were applied:
- SPOT multispectral image collected on M ay 4, 1986
- SPOT multispectral image collected on June 30, 1992
- SPOT panchrom atic image collected on July 30, 1992.
First SPOT image (May 1986) was analysed at first stage o f the works,
conducted at the SURFACES Laboratory in April 1993. After thorough analysis
it was classified into 11 land cover categories: water, coniferous forests,
deciduous forests, mixed forests, orchards/gardens, high-density built-up land,
low -density built-up land, grassland, wetland, soils and crops. The resultant
174
Zbigniew Bochenek
classification was printed at a scale of 1 : 100 000 with the use of Versatec
plotter. A t final part o f this work SPOT panchromatic image o f July 1992
was also entered into INTERGRAPH system installed at the SURFACES
Laboratory.
That image was next geometrically corrected, using image-to-map registration
software o f the INTERGRAPH system. Polish topographic maps at a scale
o f 1:50 000 were used for this purpose. About 60 points distributed evenly
through the whole scene were chosen, to ensure sub-pixel registration accuracy.
Next the whole image was transformed applying linear transform ation and
resam pled with the use of nearest neighbour method, which keeps original
pixel values. Extract of that image was finally transferred to the tape, to be
further used in Warsaw.
Further works were continued at the Remote Sensing and Spatial Information
Centre in W arsaw with the use of ERDAS image processing/GIS system.
The geometrically corrected panchromatic image of Warsaw was transferred
from the tape into the system. It was the base for correcting multispectral
image of Warsaw, collected on June 30,1992. Process of geometric correction
has been done with the use of ERDAS image processing functions. Image-to-image registration was performed with one-pixel accuracy, using 36 points
for transform ation procedure. As a result of this work, four-band file for
W arsaw area has been created; it contained three multispectral bands and
panchrom atic band of 1992 images.
In order to maintain high matching accuracy 1986 SPOT multispectral image
was registered to the rectified 1992 four-band image. As a result o f that
procedure two image files were created: for 1986 and for 1992 with mutual
registration accuracy better than 10 m. Two extracts (512 x 512 pixels) were
made from these files, to perform detailed classification and change detection
analysis. This extract covered central-eastern part of Warsaw, characterized
by changeable urban landscape.
The main part of the work was concentrated on finding the optimum method
for classification of both SPOT images and for their comparison. At first
phase it was decided to perform supervised classification based on original
SPOT spectral bands. Training fields, representing 12 land cover categories
existing within the selected image, were created with the use o f ground-truth
inform ation. Next, statistical parameters were derived from these fields, in
order to assess class separability. After this analysis it was decided to
distinguish the following land cover categories:
- high-density built-up land
- low-density built-up land
- m ixed stands
- orchards/gardens
- grasslands
Use of the classified satellite imagery.
175
- open areas (dark soil, asphalt, etc.)
- sand (sandy soils)
- water (3 sub-classes).
The same set of training fields was used for performing supervised
classifications for both images collected in 1986 and 1992. The m aximum
likelihood m ethod was applied as a type of classifier. In case o f 1992 scene
two classification images have been created; first was based solely on three
multispectral bands, while the second was formed with inclusion of
panchrom atic band for classification.
A fter completing this stage of the work it was decided to perform the
same analysis and classification for both sets o f SPOT multispectral data,
re-computed from original DN values into reflectance values. The aim o f this
work was to verify, if this procedure can improve results o f classification
and to ensure better inputs for change detection analysis. Re-calculation was
done with the use of ATCOR program, working within ERDAS environment.
This program gives reflectance values of terrain features, taking into account
atm ospheric parameters and solar zenith angle at a particular date of image
acquisition. As a result of that action two new sets of atmospherically corrected
data have been created for May 1986 and June 1992. They were next classified,
using previously chosen training fields for 8 land cover categories.
Besides the main approach concentrated on using original spectral bands,
additional analysis has been done, in order to assess usefulness o f the
transformed data for classification. Principal component analysis was performed
for both scenes and next supervised classification was done, using the same
set o f training fields, as previously.
Third approach applied in this work assumed use of unsupervised
classification for obtaining land cover map for the selected part o f W arsaw.
The special unsupervised procedure existing within ERDAS software called
ISODATA was utilized for that purpose. 20 clusters were obtained as a result
o f this procedure; they were finally merged into 8 land cover inform ation
classes.
The separate part of the work was aimed at assessment o f usefulness o f
KOSMOS images combined with SPOT XS data for land cover classification.
At first phase KOSMOS space photograph collected by KFA-1000 cam era
in A ugust 1985 was used for this purpose. It was scanned at the Rem ote
Sensing and Spatial Information Centre in W arsaw; as a result o f this
procedure digital image with 5 m ground resolution has been obtained. Next,
that image was geometrically corrected to 1992 SPOT multispectral+panchromatic
image. As a result o f these transform ation procedures all different im ages
were geometrically matched, enabling further comparative analysis. KOSMOS
KFA-1000 image was combined with three channels of 1986 SPOT multispectral
image; this four-band file was next used for further analysis and classification,
based on the same set o f training fields.
176
Zbigniew Bochenek
At the second phase o f these works, high-resolution KOSMOS image
acquired by KVR-1000 camera in 1992, was ulilized. This image is
characterized by very high 2 m ground resolution. It was rectified to previously
corrected SPOT image and resampled to 5.8 m pixel size for further works.
Next, it was com bined with SPOT bands and classified, using the same set
o f training fields, as previously.
So, applying various sets of data and their transformations 9 classification
im ages were finally created. They were based on:
- 1986 original SPOT multispectral data
- 1992 original SPOT multispectral data
- 1992 original SPOT multispectral + panchromatic data
- 1986 atmospherically corrected SPOT multispectral data
- 1992 atmospherically corrected SPOT multispectral data
- 1986 SPOT data transformed through PC analysis
- 1992 SPOT data transformed through PC analysis
- 1985/86 - KOSMOS K FA -1000 + SPOT multispectral data
- 1992 - KOSMOS KVR-1000 + SPOT multispectral data.
All classifications were kept at the same digital database, which enabled
comparison of different scenes and approaches. Such a comparison was done
for two supervised classifications based on original 1986 and 1992 SPOT
data. Using overlaying procedure existing within ERDAS GIS module two
layers were intersected. As a result of this procedure new classification map
has been created. This map comprises two new classes representing high-density and low-density built-up land, which appeared between 1986 and 1992.
Tw o other approaches have been also tested in order to make change
detection analysis. First approach was based on studying results of principal
component analysis, performed for the data set, combined from 1986 and 1992
multispectral data. This study revealed, that PC4 band includes residual image,
which illustrates changes between two analyses scenes.
Second approach utilized regression analysis for change detection. Such
an analysis was done using in-house developed software. 1986 SPOT image
was correlated with 1992 SPOT image through regression equation
Y = 0.751X+ 39.59.
Next, both images were subtracted forming residual image, which reflects
changes between two analysed scenes.
At final stage of the works comparison of the obtained classifications with
map information was performed. For this purpose topographic map of Warsaw
at a scale o f 1 : 50 000 , covering study area, was digitized using scanning
process and matched with satellite images. Next, polygons representing four
land cover categories, i.e. water, grasslands, gardens and mixed stands were
Use of the classified satellite imagery.
Ill
constructed on the basis of this map. They were finally compared qualitatively
and quantitatively with the same classes, obtained on classification images.
3. DISCUSSION OF THE RESULTS
The resultant classification images were evaluated qualitatively and
quantitatively, using set of test fields. At first phase o f evaluation different
classification approaches and various data sets were compared. The evaluation
procedure based on test fields revealed, that some land cover categories can
be classified with acceptable accuracy exceeding 80% (e.g. water, grasslands,
high-density built-up land). Nevertheless, some spectral overlaps exist between
low-density built-up land and orchards/gardens, decreasing accuracy o f their
classification. Comparison o f two classifications based on original bands and
on principal com ponent transformed bands revealed only slight changes in
accuracy. Use o f atmospherically corrected data did not im prove overall
classification results. The unsupervised approach applied in the study proved
to be inferior comparing to supervised procedure.
Tem poral comparison of SPOT data collected in May 1986 and in June
1992 revealed, that image collected in summer 1992 gave higher overall
classification accuracy (84.7% for 1992 and 79.8% for 1986). In particular,
low-density built-up land, open areas and grasslands were classified better,
while decrease of accuracy was observed for orchards/gardens. These diferences
were caused by changes in vegetation development between two analysed
seasons.
Inclusion of SPOT panchromatic band into multispectral data file did not
influence results o f classification. Almost the same percentage accuracies were
obtained for both XS and XS+PAN data sets.
Incorporation of high-resolution KOSMOS images into classification
process gave satisfactory results in case of KVR-1000 image, im proving
slightly overall accuracy (86.8%). In particular, urban classes were classified
more precisely, due to fine details, represented by KVR-1000 image.
Quantitative results of post-classification acuuracy assessment are presented
in table 1.
At the second phase evaluation of change detection map produced from
1986 and 1992 classification images was done. Analysis of that m ap gave
unsatisfactory results, comparing to groud-truth information. Too m uch area
is occupied by newly built-up land on that image. There are two main reasons
explaining this situation. First reason relates to the dates of acquisition of
both SPOT scenes. First scene was collected at the beginning of M ay, while
the second at the end o f June. Differences in response from vegetation cover
both SPOT scenes. First scene was collected at the beginning o f M ay, while
178
Zbigniew Bochenek
Table la . P o st-classificad o n accu racy assessm en t - users a c cu racy *
Name of class
1986
1986COR
1986PC
1986+KFA
High-density
urban land
79.3
79.3
82.1
71.4
Low-density
urban land
76.9
76.9
76.2
79.5
Mixed stands
67.3
68.0
70.2
75.6
Gardens
83.8
81.6
83.8
86.1
Grasslands
68.4
71.4
72.2
75.7
Open areas
80.0
80.0
73.3
68.0
Sand
100.0
100.0
100.0
75.0
Water
100.0
100.0
100.0
100.0
Overall
accuracy
79.8
80.2
80.9
80.2
Table lb. P o st-classificatio n accuracy assessm ent - users accuracy*
N am e o f
class
1992
1992COR
1992PC
1992X S+P
1992+K V R
H igh-density
u rb an land
77 .4
67.6
70.0
70.6
9 6 .7
L ow -density
urb an land
80.4
7 8 .7
85.4
78.7
7 0 .8
M ixed
stands
87.5
89.7
87.9
90.3
73.5
G ardens
78.1
78.1
78.6
81.2
9 1 .9
G rasslands
77.6
73.1
74.5
77.6
100.0
O pen areas
9 1 .7
9 5 .7
91.7
91 .7
9 3 .7
Sand
100.0
100.0
100.0
100.0
100.0
W a te r
100.0
100.0
100.0
100.0
94.8
O verall
a c cu racy
84.7
82.4
83.6
84.0
86.8
* U sers accuracy: p e rce n tag e o f co rrectly classified pixels in each class
o f G IS file
Use of the classified satellite imagery..
179
between these two dates caused, that they influenced classification o f low-density built-up land (comprising a certain percentage o f vegetation). This
situation is reflected by measures of class separability, higher for M ay than
for June scene.
The second reason is connected with character of low-density built-up land
within W arsaw area, which is often close to resolution of SPOT multispectral
image. It causes mixed pixels, which can be classified correctly random ly,
form ing "fringe" or "salt-and-pepper" effect at the change detection image.
Final evaluation was related to assessment o f usefulness o f digital
classifications based on satellite data for revision of topographic maps.
Distribution and acreage of four land cover categories: water, grasslands,
gardens and m ixed stands were compared. The remaining land cover classes
were not compared, due to difficulties in extracting from maps inform ation
concerning their distribution. Results of that evaluation pointed out, that two
classes: water and grasslands can be updated with acceptable accuracy. For
water class 1986 SPOT multispectral image proved to be optimum, while for
grasslands 1992 SPOT multispectral + panchromatic image gave the best
results. Discrepancies between two other analysed classes: gardens and mixed
stands were large; it can be partly explained by non-comprehensive information
about these vegetation categories on topographic maps at a scale
o f 1 : 50 000, which would imply opportunity of their updating with the
use o f satellite images.
Summing up, application of digital classifications of satellite data for
revision o f medium-scale topographic maps proved to be in this study quite
limited, restricted to some land cover categories. The supporting solution can
be use o f satellite image map produced from high-resolution satellite data
(including KVR-1000 images) as a basemap for delineating the desired map
categories through digitizing procedures. Such a hybrid approach, which
combines analog and digital methods for information extraction, can significantly
enhance usefulness of high-resolution satellite data for map revision.
LITERATURE
[1]
[2]
A h earn S .C., W ee., 1991: D a ta space vo lu m es an d classificatio n o p tim iz a tio n o f
S P O T and L andsat T M data. Photogram m etric Eng. and R em ote Sen sin g, V ol. L V Ü ,
N o. 1
B o ch e n ek Z., P o ław sk i Z ., 1992: U se o f satellite data and G IS for ev a lu a tin g u rb an
e n v iro n m e n t. P ro c e e d in g s o f the In tern atio n al S pace Y ear C o n feren ce , M u n ich ,
G e rm a n y , 3 0 M arch - 4 A pril, 1992
180
[3]
Zbigniew Bochenek
C saplo v ics E ., K an o n ier J., S indhuber A., 1993: H ig h-resolutio n K F A -p h o to g rap h y
fo r in teg ra tio n w ith g eo in fo rm atio n sy stem s - a case stud y o f F erto - to N atio n al
Park. P roceedings o f the 13th E A R SeL Sym posium , D undee, Scotland, U K , 28 June-1 Ju ly , 1993
[4]
D o n n a y J.P ., B inard M ., M archal D ., N adasdi I., 1992: D ev elo p m en t urbain . R apo rt
F in al. T E L S A T /II/0 6 , D ecem b re 1992
[5]
K ac z y ń sk i R., 1994: O b razo w a m apa satelitarna W arszaw y w skali 1:25 000. P race
In sty tu tu G e o d ezji i K arto g rafii, T o m X LI, z. 89
[6]
O rm sb y J.P ., 1992: E valuation o f natural and m an-m ade features u sing L and sat T M
data. In te rn a tio n a l Jo u rn a l o f R em ote S ensing, V ol. 13, N o. 2
Przyjęto do opublikowania >v styczniu 1995
Recenzent: prof. d r hab. A. Ciołkosz
ZBIG N IEW BOCHENEK
ZASTOSOW ANIE ZDJĘĆ SATELITARNYCH DO AKTUALIZACJI
MAP TOPOGRAFICZNYCH
S t r e s z c z e n i e
P rz e d sta w io n a p ra c a je s t w ynikiem w sp ó ln eg o p o lsk o -b elg ijsk ie g o p ro je k tu ,
realizo w an ego w latach 1993-94 przez pracow ników Instytutu G eodezji i K artografii
w W a rsz aw ie o ra z S U R F A C E S L ab o rato ry , U n iv ersity o f L iege. C elem tego p ro je k tu
b y ła o c e n a m o żliw o śc i w y k o rz y sta n ia w y so k o ro zdzielczych zdjęć satelitarn y ch do
ak tu alizacji m ap to p o g rafic zn y ch . Do p rac b ad aw czych zostały w y b ran e d w a o b sza ry
ag lo m e rac ji m iejskich: W arszaw y o raz L iege. D o analizy o b szaru aglom eracji
w arszaw sk iej w y k o rz y sta n o następ u jące typy zdjęć satelitarnych:
- zd jęcie w ielo sp ek tra ln e S P O T z 4 m aja 1986 r.
- zd jęc ie w ielo sp e k tra ln e S P O T z 30 czerw ca 1992 r.
- zd jęc ie p a n c h ro m a ty c zn e S P O T z 30 lipca 1992 r.
- zd jęc ie p an c h ro m a ty c z n e K O S M O S K F A -1 0 0 0 z sierpn ia 1985 r.
- zd jęcie p a n c h ro m a ty c z n e K O S M O S K V R -1 0 0 0 z sierp nia 1992 r.
W p ie rw sz e j, p rz y g o to w a w cz ej fazie p rac w y k o n an o d o sto so w an ie zd ję c ia
p an ch ro m a ty cz n eg o S P O T d o m ap y topograficznej w skali 1 : 50 000. Z g eom etryzow an e
zdjęc ie p a n c h ro m a ty c z n e b y ło p o d staw ą d o rek ty fikacji p ozo stały ch ty pó w d an ych
satelitarnych. Proces rektyfikacji b y ł przeprow adzany z w ykorzystaniem dw óch system ów
p rz e tw a rz a n ia o b razó w : IN T E R G R A P H (U n iversity o f L ieg e) o raz E R D A S (In sty tu t
Use o f the classified satellite imagery..
181
G e o d e zji i K arto g rafii). W w y n ik u tych p rac u tw o rzo n o je d n o ro d n ą b a z ę d a n y c h ,
u m o ż liw ia ją c ą an alizę p o ró w n aw c zą ró żn y ch o b razó w k lasy fik a c y jn y c h .
G łó w n ą fazę p ierw szeg o etapu p rac stan o w iła k la sy fik a c ja i an aliza p o ró w n a w c z a
ró ż n y c h ty p ó w zd jęć satelitarnych. W niniejszej p ra c y p o sta n o w io n o za sto so w a ć ja k o
p o d sta w o w e n arzęd zie analizy m eto d ę k lasy fik ac ji n ad zo ro w an ej. W ty m celu w y b ra n o
n a o b sz a rz e W arszaw y p o la tren in g o w e rep rezen tu jące 12 k ateg o rii p o k ry c ia teren u . P o
w y k o n a n iu s ta ty sty c z n e j a n a liz y ro z d z ie lc z o ś c i k la s d o w y k o n a n ia p ro c e s u
k la sy fik a c y jn e g o w y k o rzy stan o 8 k ateg o rii p o k ry c ia terenu, a m ianow icie:
- o b sz a ry o zw artej zab u d o w ie
- o b sz a ry o ro zlu źn io n e j zab u d o w ie
- d rz e w o sta n y m ieszane
- sad y i o g ro d y d ziałk o w e
- u ży tk i zielo n e
- teren y o tw arte (z o d k ry tą glebą)
- o b sz a ry piaszczy ste
-■w ody.
T en sam zestaw pól tren in g o w y ch zo stał zasto so w an y d o sk la sy fik o w a n ia ró żn y ch
ty p ó w d a n y ch satelitarn y ch . W p ierw szy m etapie p o d d a n o k lasy fik acji zb io ry o p a rte na
o ry g in a ln y c h 3 -k an alo w y ch dan y ch w ielo sp ek traln y ch S P O T z 1986 i 1992 r. o ra z na
d a n y c h p a n c h ro m a ty cz n y ch S P O T z 1992 r. N astęp n ie p rz e tw o rz o n o w y jścio w e d a n e
w ielospektralne SPO T, poddając je korekcji atm osferycznej o raz stosując do przetw o rzen ia
m etodę składow ych głów nych. Przetw orzone dane S P O T zostały następnie sklasyfikow ane
z w y k o rzy stan iem stałego zestaw u p ó l treningow ych. W k o lejn y m etapie p rac u tw o rzo n o
zb io ry w ielosp ek traln e, p o ch o d zące z d w óch typów satelitów : S P O T i K O S M O S (K F A -1 0 0 0 i K V R -1 0 0 0 ) o raz p o d d an o je p ro ceso w i k lasy fik acy jn em u . W w y n ik u tych p rac
u tw o rz o n o 9 o b ra z ó w k la s y fik a c y jn y c h , b a z u ją c y c h na w ie lo s p e k tra ln y c h i
p a n c h ro m a ty c z n y c h dan y ch z d w óch ty p ó w satelitów .
W drug im etapie prac przeprow adzono analizę po rów naw czą otrzym anych klasyfikacji.
S p o rząd zo n o także m apy zm ian u żytkow ania ziem i w latach 1986-92, p o ró w n u jąc obrazy
k la sy fik a c y jn e S P O T p o ch o d ząc e z tych dw óch term in ó w rejestracji. Z a sto so w a n o
w ty m celu trzy m etody: p rzecięcia 2 w arstw in fo rm ac y jn y ch , m e to d ę sk ła d o w y ch
g łó w n y c h o ra z m eto d ę k o relacji o b razó w . W k o ń c o w y m etapie p ra c p rz e p ro w a d z o n o
p o ró w n a n ie u tw o rz o n y ch o b ra zó w k lasy fik acy jn y ch z m a p ą to p o g ra fic z n ą w skali
1 : 50 0 0 0 , w celu o c e n y m o żliw o ści w y k o rzy stan ia tych o b ra z ó w d o a k tu a liz a c ji
śred n io sk alo w y ch m ap to pograficznych.
E ościo w a i jak o ścio w a ocena dokładności klasyfikacji poch o d zący ch z różn y ch typów
d a n y c h w y k azała, iż w y so k ą .d o k ła d n o ść p rz ek raczającą 80% o siąg n ięto d la n iek tó ry c h
klas p o k ry c ia terenu (w o d y , użytki zielone, obszary o zw artej zab u d o w ie). Z asto so w a n ie
p rz e tw o rz e ń d an y ch o ry g in aln y c h (m eto d ą k o rek cji atm o sfery czn ej lub sk ła d o w y c h
g łó w n y c h ), ja k ró w n ież w łączenie zd jęcia p an ch ro m a ty cz n eg o S P O T n ie z m ien iło w
sposób istotny w yników klasyfikacji. W ykorzystanie obrazu panchrom atycznego K O S M O S
K V R -1 0 0 0 sp o w o d o w a ło n iezn aczn e p o d n iesien ie d o k ła d n o śc i k lasy fik ac ji, z w ła szc z a
d la tere n ó w z ab u d o w an y ch .
O c en a ja k o ś c i m ap zm ian u ży tk o w a n ia ziem i w latach 1 9 8 6 -9 2 w y k a z a ła
n ied o sta te c z n ą p re cy zję o k reślan ia o b szaró w p o d leg ający ch z m ian o m . G łó w n y m i
182
Zbigniew Bochenek
przyczynami tego stanu rzeczy były różne okresy wegetacji roślin oraz rozdzielczość
danych wielospektralnych SPOT, wywołująca w przypadku terenów zabudowanych tzw.
efekty brzegowe.
Porównanie obrazów klasyfikacyjnych z mapą topograficzną wykazało, iż przydatność
tych obrazów dla aktualizacji map jest ograniczona do niektórych klas pokrycia terenu
(wody, użytki zielone). W podsumowaniu stwierdzono, iż najlepszą metodą dla
aktualizacji map na podstawie zdjęć satelitarnych byłaby metoda hybrydowa,
wykorzystująca wysokorozdzielcze obrazy typu KOSMOS KVR-1000 do wydzielania
głównych typów pokrycia terenu w sposób analogowy i wprowadzenia wydzieleń w
postaci cyfrowej do numerycznej mapy topograficznej.
ЗБИГНЕВ БОХЕНЕК
ПРИМЕНЕНИЕ СПУТНИКОВЫХ СНИМКОВ
ДЛЯ ОБНОВЛЕНИЯ ТОПОГРАФИЧЕСКИХ КАРТ
Резюме
Представленная работа является результатом совместного польско-бельгийского
проекта, осуществляемого в 1993-94 годах сотрудниками Института геодезии
и картографии в Варшаве и SURFACES Laboratory, Университета в Льеж.
Целью этого проекта была оценка возможности использования спутниковых
снимков с большой разрешающей способностью для обновления топографических
карт. Для исследовательских работ были избраны два фрагмента городских
агломераций: Варшавы и Льеж. Для анализа пространства Варшавской
агломерации были использованы следующие типы спутниковых снимков:
- многозональные снимки СПОТ от 4 .мая 1986 г.
- многозональные снимки СПОТ от 30 июня 1992 г.
- панхроматические снимки СПОТ от 30 июля 1992 г.
- панхроматические снимки КОСМОС КФА-1000 с августа 1985 г.
- панхроматические снимки КОСМОС КВР-1000 с августа 1992 г.
В первой подготовительной фазе работ была произведена подгонка
панхроматического снимка СПОТ к топографической карте в масштабе
1 : 50 ООО. Геометризированный панхроматический снимок был основой для
ректификации остальных типов спутниковых данных. Процесс ректификации
проводился с использованием двух систем преобразования изображений:
INTERGRAPH (Университет в Льеж) и ERD AS (Институт геодезии и
картографии). В результате этих работ была создана однородная база данных,
дающая возможность производить сравнительный анализ разных классификационных
изображений.
Главной фазой первого этапа была классификация и сравнительный анализ типов
космических снимков. В данной работе было решено применить в качестве основного
орудия анализа метод классификации "с учителем”. С этой целью были подобраны
Use o f the classified satellite imagery..
183
на территории Варшавы тренировочные поля, представляющие 12 категорий
покрытия местности. После проведения статистического анализа разрешающей
способности для проведения классификационного процесса были использованы 8
категорий покрытия местности, а именно:
- территории со сплошной застройкой
- территории с разреженной застройкой
- смешанные древостой
- сады и огороды-участки
- пастбищные угодья
- открытые территории (с открытой почвой)
- песчаные пространства
- воды.
Такой же набор тренировочных полей был применен для классификации
разных типов спутниковых данных. На первом этапе были классифицированы
файлы, основанные на оригинальных 3-х канальных многоспектральных данных
СПОТ 1986 и 1992 гг. и панхроматических данных СПОТ 1992 г. Затем были
преобразованы выходные многоспектральные данные СПОТ, подвергая их
атмосферной коррекции и применяя для обработки метод главных составляющих.
Преобразованные данные СПОТ были затем классифицированы с применением
постоянного набора тренировочных полей. На следующем этапе работ были
созданы многозональные файлы, выводящиеся с двух типов спутников: СГ10Т
и КОСМОС (К0А-1ООО и КВР-1000), и затем были подданы классификационному
процессу. В результате этих работ создано 9 классификационных изображений,
базирующихся на многозональных и панхроматических данных с двух типов
спутников.
На втором этапе был проведен сравнительный анализ полученных
классификаций. Были составлены также карты изменений в землепользовании
в 1986-92 годах, сравнивая классифицированные изображения Спот, происходящие
из двух сроков регистрации. С этой целью применено три метода: разреза
2-х информационных слоев, метод главных составляющих, а также метод
корреляции изображений. В конечной фазе работ произведено сравнение
составленных классифицированных изображений с топографической картой в
масштабе 1 : 50 ООО, с целью оценки возможности использования этих
изображений для обновления среднемасштабных топографических карт.
Количественная и качественная оценка точностей классификации, происходящих
из разных типов данных, показала, что высокую точность, превышающую 80,
получено для некоторых классов покрытия местности (воды, пастбищные угодья,
территории со сплошной застройкой). Применение преобразований оригинальных
данных (метод атмосферной корреляции или главных составляющих), а также
привлечение панхроматических снимков СПОТ, не изменило существенным
образом результатов классификации. Использование панхроматического изображения
КОСМОС КВР-1000 привело к небольшому повышению точности классификации,
в особенности для застроенных территорий.
Оценка качества карт изменения землепользования в 1986-92 гадах выявила
недостаточную точность определения территорий подвергающихся изменениям.
Главными причинами этого положения вещей были разные периоды вегетации
184
Zbigniew Bochenek
растений и разрешение многозональных данных СПОТ, вызывающее в случае
застроенных территорий, так называемые, береговые (граничные) эффекты.
Сравнение классифицированных изображений с топографической картой
выявило, что пригодность этих изображений для актуализации карт ограничивается
до некоторых класс покрытия местности (воды, пастбищные угодья). В итоге
установлено, что самым лучшим методом для актуализации карт на основе
космических снимков был бы гибридный метод, использующий изображения
с высоким разрешением типа КОСМОС КВР-1000 для выделения главных
типов покрытия местности аналоговым способом и введения выделений в
цифровом виде на цифровую топографическую карту.
Перевод: RóżaTołstikowa