remote sensing image data analysis system ......remote sensing 工mage data analysis system...

24
ISE-TR-86”54 ・懸 REMOTE SENSING IMAGE DATA ANALYSIS AND IT’S APPLICATION 一Especially Classifing Coastal Area around by Takashi Hoshi Kiyoshi Torii Tomoyuk日shida January 6,1986

Upload: others

Post on 19-Sep-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

ISE-TR-86”54

・懸  .

REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM{cTSUKUSYS”

               AND IT’S APPLICATION

                                            一Especially Classifing Coastal Area around a Bay Using↑M I-

                      by

                  Takashi Hoshi

                   Kiyoshi Torii

                 Tomoyuk日shida

                 January 6,1986

Page 2: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

     Remote Sensing 工mage Data Analysis System ”Tsukusys”

                    and 工tIs Appl ication

一 Espec ially ClassiEing Coastal Pllrea around a Bay Using [lhvi 一

                 Takashi Hoshi

工nstitute of 工nfo rma tion Sc iences and Electronics

             Uni▽ersity of Tsukし止)a

               Ibaraki 305 Japan

                                                      Klyoshl Torll

Depar七menし of Agricultural Engineering

                   コ                             KyOt=O UnlVerSlty

          Kyoし0 666 Japan

         Tomoyuk i 工shida

Depa rtment of Research Developrnent

      University of Tsukuba

        工baraki 305,Japan

Page 3: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

     Rem・te Sensingエmage Data袖a・ysiS SYstem ’lTSukusys・

                     and 工tls Application

一一一@Especially Classifing Coastal Area around a Bay Using IXvr 一一一

                       Takashi Hoshi

     Institute of.Inforrnation Sciences and Electronics                   University of Tsukuba

                     Ibaraki・ 305 JAPAN

            Kiyoshi Torii

Department of Agricuitural          Kyoto University          Iくyoto 606 JaDan                        L

Engineering

         Tomoyuki Ishida

Department of Research[)evel・μnent        コ                   

      Unlversit二y of Tsukuba

        Iba raki 305 JAPAN

Abstract

     A system called as ”ISukusys” has been developed to

distinguish and analyze land use and land cQver patterns by

uslng   remote   senslr阿  1mage  data r especially  Landsat二

Multispectral Scgnner(IV!SS) irnage data. As, in Japanr

Landsat二 Thematic  IYIapper (Tひ4)  image  dat二a can be acquired.

slnce last year, the appl ication of TM data by the Tsukusys

was attempted. As a result, the ad▽ant二age of the T岡data is

coinpletely vertified. This investigation is to interprete

wat二er  quality condit二ion around t二he Koj ima Bay, where water

pollution becomes a social issue.

Page 4: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

Contents

   Introduction

   Configuration of [Vsukusys 一一一一一一一一一一一一一一一一一一一一

   2-l Configuration oE’hardware 一一.一一一一一一一一一一一一一一一一一

   2-2 Configuration of software 一一’一一一一一一一一一一一一一一一一

   2-3 Sorne fea tures o f the Tsukusys 一一一一一一一一一一一一一

   An Exarnple of Tsukusys’s appl ication 一一一一一一一一一一・・一

   3-1 0bゴective to classify coastal area   一一一一一一一一一

   3-2 Description of test site 一一一一一一一一一一一一一一一一

   3一一3 Hydraulic structure in test site 一一一一一一一一一一一一一一

   3-4Analyticai procedure 一一一一一一一一一一一一一一一一・一一一一一一一・一一一一一一一

   3-5 Results and discussion 一一一一一一一一一一一一一一一

   3-6 Conclusion of appl ication 一一一一一一・一一一・一一一一一一一一一一一一

4 (lonclusion 一一一一一一一一一一一“”一....一www”.w

1

2

3

1

1

1

4

9

9

9

Acknowledgments

References

,夢9曜,劇,e-po-oo響〇一一一一一,一一一}一画岬,,,,

嘲■脚ρ●嘔幽P●■P■■■■■「刷■P■一■-騨門■■■■脚ρ口■■「,■■r騨■r■騨P■■’朝■■■■9「■■9r●■■■■■引■闘-■V燭■脚7■■’■■騨■■F■口脚q願t刷圏ρ■■PO■艶闘■■’■■F■■■■馴■P■■騨

10

12

12

13

17

18

18

18

Page 5: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

1. 工ntroduction

      an invesしigaしi・n・n regi・nal scale is indispensaple f・rしhe research・f earch

science’   for  the  invenヒories  of  natural  resource  and  for  the  monition  of

                                                つ                          ロ

env1「o㎜ent・It ls also lM’P()rtant to invesしigate the temporal variaしion on the

「egion・da「th”・bservati・nal sens・r sys七ems・・aded in sate・・‡しes ha▽e rnet these

requirements since・97・Is・画・wぬndsat雌S data are c・㎜ercia・・Y・bしainab・e and

extensively used in Japan・T・st・re the multiしemp・ral lV6S data may be。ne。f

necessary conditions  to  develop  t士1e  apPl icaしion  of  remote  sensing  しechnique.

Therefore r a  daヒabank  cons i sting  of  a  large amoしmt of the image data should be

establ ished, based on daしabase concepts. Wi t士ししhe realization  of  the  databank  in

mind,  remote  sensing image data analysis system ca11$d as lITsukusys闘 was developed

in1983.

     This paper g ives an outl ine of the Tsukusys. Segしion 2 describes  the  hardware

organl zaしlon  and  software  organization  of  the  system. Section 3 discusses the

apρlication of the system’ esρecially t血e usefulness of Landsat TM daしa t二〇  classify

coasta]. area  around  tne Koゴima Bay that includes a desalted reser▽oir where water

quality has been degraded。

2. Configuration of TsuK)usys

2-i Configuration of hardware

a. Host Cornputer System

     Ahost computer oE 48 M bytes memory capacity r which runs at Science

エnf・rllati。n Pr・cessing Cenしer’Uni▽ersity・・f TsukUba(Fig一・),is used as rem・te

sepsing image analysis systern. Here the most important design was how can transfer

data frqm. Iarge capacity d i sk memory to irnage rnernory at high speed. This problem was

resolved by developing an interface between thern and by connecting them onl ine

                                     -1 一一

Page 6: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

I

N曜

Other Faculties

ComputerNet Work

Remote Station(27 Stations )

TSS Terminal(300 Terminals )

N-1 Network (15

  Universities)

Open Ba tch

       Station

Science lnformation Process’ing Center

Hos t Cornpu ter

CPU Memory(48 M bytes.)

BMC

DASD DASD

ccp BMC

Optical Cable

TEK 4025

TEK 4027

× 16

×4

Computer Sys tem for Education

Hos t. Computer

 CPU・Memory (24 M bytes)

  のStandal.one

Termina・1・

DASD

5 GB

MTI 2)3

× 90

⑤…一……〈動

78 GB 1600/6250 BPI

OpticalDisk

g}t15ffil{fiaging’

Disk

  Mass

StorageSys tem

101.7 GB

Imqge Processing System

lmage  Memory

1.00 MB

Image

  Memory.O.78 MB

512 ×512

ColorPisplay

×

512 ×512

ColorDisplay

Smalt Computei’

CPU Memory   (O.77 MB)

1024 × 1024

ColorDisplay

1024 × 1024

ColorDisplay

×

×

×

1

1

DASD

320 MB

1

1

MT )MT Termina1 ×3

2000 × 3000

CCD Camera

Fig-1 Con.figuration of Host Copmputer Sy$tern

Page 7: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

through central process ing units (CPU). [Ehe aux i l iary storage oE the host computer

cons i sts of 9 magnetic tape (MT) units, a d i sk systein (DASD) (78 G bytes) and a mass

storage systern (IOI.7 G bytes).

     工nterdata  8/23 system is also available at the center as a sub-compu仁er syst:em

for analyzing remote sens ing image data. [[he sub-systern includes a high-density

drurn-scannerr which the Tsukusys uses for digitizing color aerial photographies.

b. mass Storage System

     工mage databank of Landsat MSS data was constructed and has been arranged so far

as to .sto re 106 tandsat irnages taKen wi th in and outside Japan

〈 Hosh i et al (1983) 〉. ’IWo important features of the databank a re to bring tandsat

image data inし。 uniしy data format, that is, B工L fo rmat:and to  be  able  to  process

thern onl ine. The later fea ture i s ’pe rformed by the mass sto rage systern.’IXAio

cartridges as shown in Photo-1 (size of cartridge is 4.7 crn in diarneter and 8.8 orn

in height) can efficiently and systematically store a whole image taken by tandsat.

Photo-1 Ca rtridge of itass Storage Systern

一3一

Page 8: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

t

Using the daしabank. comp◎sed of mass storage system, it; is・ possible  to  select  an

interested  image thrdugh image ret=rieval system〈 工shii et;aユ. (1983) 〉.,、 to take ouし

the dat二a stored in mass storage syst(『m to stag ing d isk, and t len to convert the dat:a

stored  in sしaging disk t二〇 disk system・ This process reduces operaしion in respect;しo

labour and time, by comparing wi th しhe operat二ion t=hat a dat二a stored in magneしic ヒape

ロ              リ                                                                                                                                    コ      

エs  d↓rectly  coρied to dat;a file of DASD. 工n additlon, this mass storage system・・haS

tne merit;of smaller cost than the disk sy$tern in performing。,

c・ Color graphic disρlay

     Two color graρhic displays are included  in  the  image  I/0  ρrocessing  units

connected to Che host compute r. The d i splay uniしs have l.78 iyi byt二es (512 x 512 byヒes

x 4 frames and 3 frames) image memory. The reason why two dispユ.ays are adopt;ed is to

be  able  し。  operate  them  indeρendenしly and t;o compare the image before and aft;er

processing・ This is convenienし fQr image processing・

d. Co:Lor ink一一ゴさt printdr/plotter

     エtis des;rable t。 preser▽e s。me images・r maps represgnted。n・c・1・r graphic

displays・  Of co]・or pr:t.nter/plotter on sale’ a color ink一ゴet printer/plot;しer made by

Applicon Comp. in U.S.A. was thought having enough capability for use. lhe  maximum

picture  size produced by it is more t;han four times than (重uart:o paper. 工し is larger

than an ordinary map in Japan。 Although the design ofl ha rd wa re is planed t;o  process

。n・ine as menti・ned ab・▽er the printer/P・・tヒer is n・t c。n「iected・n・ine・pecause

ρ「inting and。thr「p「ocessing can p「ocee昌at甑e same time’total P「。cessing time

will be reduced’・wing t・。ffline pr。cessing・As a s・IUble ink in water.jets’ヒhe

printer/pl・tter has a demerit that c。1。r。f pictures.fades。u七as m◎istening.エt had

bette「・laminate picture串in。rdrr t。 prevent e。1・r fr・m fading「・u七・

2-21 bonfiguration of soEtware

a・ Usable inpuし image data sets              .           ・                   、

     If only・・工andsat MSS data are concerned,the mass st;orage syst二em is enough for

pracしical use. But there are other remote sensing image data sets’ so  工/O  soft壷are

is installed f・r analyzing these image daしa sets・Usable input image data sets qre

                                     -4一

Page 9: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

飴ndsat雌S data st・red in fl・pPy disketteダdigiしaiized aerial c・IQr ph・t・graphic

data’and airborne tv!SS data.

(1) Landsat fv!SS data

     In Japan r £or rnicro-computer user, 8 inch floppy diskette (1 M bytes) in which

irnage data belonged to a portion of a irnage taken by tandsat tvlSS is storeq can be

acquired.

(2) Digitalized aerial coior pnotographic data

     An aerial color photography is digitalized as 8 bits data with Red-Green一一Blue

color.coding system r using above一一mentioned drurn-scanner 〈 Hoshi et al (1981) 〉.

(3) Airborne MSS data

     The airborne MSS data l i ke iY!SS data produced by Jirco Comp. in Japan or

Daedalus Cornp. in U.S.A. are also available.

     Pre-processing includes 4 different cornponents, that is, speciEication of test

site, elimination of striping noiser geometrical registration and correction to

norrnal distribution.

(1) Spef iEication of test site

     工n  order  to  specify a しest siしe aしwill by using truck-ball, a whole Landsat;

LYISS dat;a or oしher image daしa are ext二racted at intervaユ of 7 ρixels, so  all  of  しhe

selected pixels can be displyed on a coior graphic display sirnultaneously. At this

operation r a mesh at 512 pixels interval is displayed together with remote sensing

image data r so the specification of a test site is easily perforrned.

(2) Elimination of striping noise

     MSS embarked on Landsat has 6 detectors peF channel. Due to the difference of

response , しhe lag of offset二and A/D con▽ersion wit二h t;hese deteqしors’ stripes  t;urn

up systernatically. [[he Tsukusys has capability to correct them.

(3) Geometrical registration

     As rernote sensing image data do not correspond geornetrically with ordinary map,

it is diEficult to cornbine other data with some resulting irnages or maps produced

                                       一一 5 一一

Page 10: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

fr。m t・he image daしa.エn・rder t。 res・1▽e this pr。blem’the Tsukusys P「epa「esξ

software especially for geornetrical registration t hrough universal transvers(

mercat。r(U珊pr・ゴecti・nくNagamine et al(1982》》・U獅pr・ゴecti。n was ad・pted b:

the reason that ordinary rnaps on the scales of 1/25,000 and 1/50,000 in Japan hav(

been produced using this projection. , ,(4) Correction to norrnal distribution

     The rnernory of host computer and irnage mernory are capable of assigning a dat,

into a byte. Therefore, previous acquired Landsat rvlSS data which are forined as

                                                                 /

bits data are converted. to 8 bits ones and corrected to norma1 distribution wit

taking the effect oE enhancement into account.

c. unsupervised classfication ・,     [the unsupervised classfication analysis applied in this systern is t h

hierarchical clustering method which includes tne 6 different componenPs as follows

(1) Wa rd rnethod

(2) Group average rnethod

(3) Centro id rnethod

(4) Med i an rnethod

(5) Furthest ne ighbour method

(6) Nearest ne ighbour rnethod

     Although it is desirable t hat all pixels within a test site are submitted to

clustering algorithm for producing a resulting map, it is not efficient in terms e

calculation time. Hence, in this systern, an approach to improve classification spee

is ad・ptedぴacc・rding t・bel・w・rnenti・ned pr・cedures〈H。shi et al(in press)〉.

     The 256 pixels that are randomly located are extracted frorn interested sub一一tes

site. They are regarded as rePresentative of test site. Ctr)e method chosen frorn thos

mentioned’ ≠b盾魔?@is performed within the 256 pixels. [Ehe 256 pixels are ・flnall

combined into 20 clusters. Every pixel of test site is then assigned to the ciuste

sAzhose mean vector ’is closest to the pixel vector. For display, each pixel assigne

to s珈e cluster is labeled wiしh one corresp()nding color, a color sch㎝e is Inanuall

altered so as to match well.                                   一一 6一

Page 11: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

d. Supervised ciassification

     ll)e well-known rnaxirnum l i kel ihood method i s implernented to ・assign the pixels to

one of pre-deterrnined classification iterns according to the analysis objectives. 2

criterions are adopted to assess the degree of success oE the classification

results.

(1) Average perforrnance

     [[he average performance is obtained on the basis of correct recognition for

each i tem and is expressed as

                 n

            S=.Z.:.1 ri/n .一一一一一一一一一一一一一一一m一一 (2-1)

     Vghere r and n refer to the percentage of correct recognition for each item and

the number of i terns respectively.

(2) Equivocation quantification

     .1he equivocation・qunatification is ’based on the concept of entropy by

information theory and considers the infuluence of rejection items. 〔【『be fしmction  is

denoted by [V(G,G) as expressed in equation (2-2).

T(G,召) = 工(G,ご)’/H(G》

        = 1 一 { H(Gr ’G一’) 一 H(G一) }/ H(G) ’一”一一一一” 〈2‘”’2)

     Vthere 1(G,G)r H(GrG), H(G) and H(G) refer to rnutual inforination, joint entropy,

entropy of a prior itern and entropy of the posterior item respectively.

     T(G’G) varies from o’七〇 1・ エf the value is obtained to be higher than o.65, the

classi[ication result is generally considered as good.

e. EValuation of・distance and area

    -The estimation in distance and area can be carried out on the basis of counting

the nuMbeF of line and colume in interested image data. A distance can be evaluated

                                     -7 一一

Page 12: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

by  specifying two ρoints on CRT wit士1しruck・一ball・Also, an area can be sized up from

sum total of pixels belonged し。 same classification item not only in  a  rectangular

region but also in an irregular region. A disしance or an area evaluaしed in this way

will be compared wit血 bygone data acquired by ground survey or image dat二a.

f. 工mage descripしion

     Ouしput devices for describipg image data include color graphic  display, coユ.or

ink一ゴeし prinしer/plotter and X-Y plotter.

(1) Display on color graphic display

     trh6 subr。utines f・r pr。▽iding a funとti。n t・dispユay。n CRT were devel・ped.

These subroutines include the data ponversion from one「CRT  to  ot二her‘CRT  and  the

representat二ion of image data depend ing on some formats.              ’,

(2) Display on color printer/plotter

     Although  the  color  ink一つet  printer/plotter  installed  aし ーヒhe center adopts

Lyragent;a, Yellow, and Cyan (M.Y.C) color coordinate, Red r Green, and Blue  (R.G.B)

c・1・rc・ding systern can beしransf・rmed t。 LYi.Y.C acc・rd ingし・an expressi・n bel・w’

         M  al「bl cl R

         Y   =   a2   b2  c2  .  G    一一。一一一一一一一一一一一一一一(2-3》

         C  a3 b3 c3 B

     Where the matrix elements are experimentally determined.

     Hisしograms  for  each  of M, Y and C are construc=ヒed for each image. The M-Y・一C

levels are transforrned to color plotter patch which is composed of l6 dots either in

equal disしance basis・r in equal pr・babiliしy basis.16 d・ts patch can represenしi7

Cし止×∋ color le▽eユS.

(3) Display on X-Y plotter

                    

     工し 1s lmportant;to observe features of training  data  utilized  by  suρervised

classification  analysis. The Tsukusys has capabilit;y to represent these data on X-Y

pl・tter’using.1’c。nstellaしi・n meth・d”by which rnulti▽ariate data can be displayed。n

plane  or  it 4-axis  methodi1 by  which  4 efficient channel da ta can be disρlayed on

ort血ogonal coordinaしe system, so distribution of these data becomes clear。

                                      一8一

Page 13: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

h

Z2-3 Sorne features of the Tsukusys

     Landsat tvlSS data cover all over the werld and are . represantive of rernote

sensingネmade da ta・エt is imp。rtant as a system f。r educati・n and training t・

analyze the popular irnage data by using the authorized unsupervised and supervised

cl.assiEication analysis. ln designing digital image data analysis system,. most of

designers only have an inclination to construct hardware systern and to develop

software system, not including image databank. The reason is that the databank is

too expensive to be werthy of study obj’ective. However, in future, it may be

indispensable Eor.remote sensing irnage data analysis systern to contain an image

databank and the retrieval systern.

     A Color graphic dispiay plays an important role for observing image data. So,

the Tsukusys efficiently ㎝ployes the 2 displays.工t is also important to shorten

the time for displaying image data. Here, high-speed processing is realized by

developing the inte r face between host cornputer and image rnernory.

     Citing other features, the way how to extract the data appl ied in cluster

a19。砒㎞with unsupe「vised classificati・n’new criteri・n with accgracy。f

supervised classification (equivocation quantification) and so on are included.

3. An example of Tsukusys’s appl ication

     Here, the attempt to classifY coastal area around a bay using [ITvl is introduced

as one of applications 〈’Ishida et al (1985) 〉.

3一一i Qbjective to classify coastal area

     [[he objective oE this study is to investigate the usefulness of Tlvr data for

lnte「p「etユng職e「quality・Especially’the study was E・cused・n classifying water

zones and extracting flow patterns.

一9一

Page 14: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

Phoし。-2 Natural Color 工mage of Test Site and Training Classes

3-2 Description of test site

     The area stud ied includes the Koゴima Bay in Japan. 工t  i s  encompassed  by  t」ae

K・jima Peninsula Wnich is 1・cated in the cenヒralρart。f Seし・エnland Sea and several

Kiiorneters away from Okayama City in the southern direction・A  reclamation  proゴect

aroしmd  the  bay  by  drainage  has  been  begun more than one hundred years before.

However e because there is no sufficient water for the reclairned agricultural land,

crops  frequently  sustained drought damage・ 工n order し。 sol▽e this problem’ a levee

across the middle of the bay was constructed in l962・ 工し ρrevents sea  wat二er from

flowing into tne bayr and a desalted reservoir with an area of 1,088 ha was formed.

As a resultr suf£icient irrigation water supply got guarantee and reclaimed land

became available.

一 10 一

Page 15: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

    tately, Kurashiki and Sasagase River catchrnents proceeded to be urbanized, wi th

the inc rease of population on the outskirts of ekayama City. Accord ingly,

urbanizaVion caused water pollution in the reservoir. Because both industry waste

water and dornestic sewage of cities are ’poured ’into” rivers and water from the

yoshii, Asahi, Sasagase and Kurashiki River Elows into the bay (Fig-2)i So various

chernical materials move into the reservoir and deposit there. Water quality of the

reservo i r comes to be deg raded. :t i s thought that i t is irnpo rtant to construct a

reg ional sewerage net to draw clear watet from the upper reach and to drain water

flowing in lower stratum by desalination underdrain, as to irnprove the water

Sea ef Japan

D

、欝t確書

Se

’Kyes.hLu

   Di st ri ct

Shikoku

 District

Pacfic Ocean

    ioo KmN一一,.一一..一,.一一dve

Fig-2 Geographical ILocation of Test Site

一 11 一一

Page 16: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

quality. The present study was attempted  to  extract  the  information  about; flow

pat;しerns  aroしind  the  bay  from TM data 〈 Torii et al (1985) 〉’ in order to pro▽ide

scientific e▽idence for the proゴect of water purification・

3一・3 Hydraulic structure in test site

     The level of sluice gaしe for drainage in Koゴima desalted reservoir  is  usually

adゴusted acc。rding t。 the rise and fall・f tide・[[he sill。f the gaしe is f・unded at

the elevation of 3 meters below the sea-1e▽el and’ some  spots  are  deeper  しhan・ 3

rneters. Fig」3  shows the isobat血 of desalted reservoir at present二・Most area of tt)e

reser▽oir are.shallow, aboUt 2.5 m in depth. There are only two deeper spots along

しhe levee and two deeper spots nea亡 ’].efし and right shore respectively.

                                                                       ウ     Fig-4  shows  the distribution of ▽elocit二y of plan flow obtained by もol▽ing・two

dilnens ional  sha1ユow  water  equation  〈Kawachi et al (1983) 〉,  using  depth

disしribution・data  as boしmdary condition. The resuit is acquried on t士ie calculation

condiしion that much water water flows int二〇 tne reservoir, on t士1e  ot血er  hand r much

water emits from tne gate, and so the stream patterns、 can l be completely understood・

3-4Analytical procedure

     Of  softwares. mentioned   in   configuration  of  software, pre一一processing,

uni3up6r▽ised classification, supervised classification  and  image  descript二ion  are

adopted  and  t血e  image  processing proceeds in turn. T()beg in wi th, the TM digiしal

daしa t血at were acqui red on 8th May l984, are con▽erted for handling a§ MSS data  and

pre・一precesSing is performed on t血e convert二ed data set二s.

     エnthis c昌se, pre-pr・ceSsing d・es n。t include ge。metrical registraしi・n and

correction of striping noise, because the registration is not necessary judging from

the  obゴective  of the ρresent study and striping no i se can not be foしmd. 4 bands TM

daしa corresp◎nding し。 wave range of 瞭∋Sグ that is, Band 2,3, 4 and  5. were  selecしed

for analysis. The  しest  site is disp].ayed by composing bands data in Photo一・2. And

t血en, unsupervised classification is performed using centroid method. The result  is

utilized  as reference[data for super▽ised classificat二ion・ Super▽ised classification

is performed using maximum likelihood method・

                                    一12一

Page 17: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

3-5 Results and discussion

     Phot二〇一3 is t二be classificat二ion map on the case that しraining siヒes amount二 t二〇  14

sites shown in Table-1 and Table-2. Water bQdies and land area were classiEied into

8 and 6 types respectively. Judg ing frorn the con;ingency table (Table-2), average

accuracy is high at 83.9 percentage of correct classiEication and at O.86

equivocaしion quantification ・ エヒ is considered as high as t;he results acquired  from

airborne  MSS  data 〈Hoshi (1979) and Torii et a1 (1979) 〉. However, the accuracy of

classiEication item 4t 5, 12 and 13 are extremly low. Lirniting nbtice to the types

concerning water body.e altnough 8 types are correctly classified, the types do not

lead t士1e resultanし map representing some features  of  filow.  4  combined  t二ypes  in

respect of land and 3 coinbined types of water body are displayed in PhQto-4 so as to

disしinguish definiしely the st二aしus of flow.

Photo-3 Supervised Classification Map (No. of Training Classes is 14)

一 13 一一

Page 18: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

Table-1 Characteristic of Training Classes

1

 心i

Classification it㎝ No.of pixels Band 2 Band 3 Band 4 Band 5

1. Wa ters in the reservo i r 216 105.6±1.2 42.2±0.7 41.4±0.8 28.8±0.5

2. Waters in the Koj ima Bay 200 115.2±1.5 49.1±0.7 45.7±0.9 27.3±0.8

3. Wa ters in the I n l and Sea 416 109.5±1.3 ’40.9±0.7 36.8±1.0 21.3±0.8

4. Wa ters in Sasagase River 120 117.3±3.2「 49.3±2.4 ’51.0±4。7 44.6±14.8

5. Waters in Kurashiki River 104 112.7±L4 47.3±1.4 48.2±1.9 49.1±20.9

6. Wa ters in reclaimed area 脳 110.8±1.5 45.8±0.8 44.1±1.2 33.2‡4.8

7. Waters in Yoshino River 192 119.3±1.5 49.6±1.1 50.4±1.9 32.0±3.3

8. Wa ters in Asahi River ㎜ 119.1±1.6 49乙5±1.2 48.6±3.1 33.3±7.6

9. Paddyfields in planting season 132 111.6±2.5 47.9±1.6 47.3±3.3 122.4±7.4

10.Urban areas 娚 137.6±4.1 60.5±2.1. 71.8±4.3 65.8±4.8

11.Trees 謝 103.0±2。0 46.7±1.9 43.6±1.8 107.3±15.0

12.Bare fields 45 140.9±6.9 .〔違.0±4.1 87.9±7.6 76.9±6.5

13. Concre te s t1Mc tures around coas t 80 126.3±4.7 56.7±2.9 65.5±5.5’ 69.3±8.5

14.Shad㎝s 140 122.6±4.6 55。7±2.6 63.4±4.6 84.5±9.6

Page 19: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

 1

Aor l

Table-2 Contingency Table of Supervised Classification

Classification it㎝ No.of pixels 1 2 3 4 5 6 7 8 9 10 11 12 13 14

1. Wa ters ■1n the reservoir 216 98.6 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.5

2. Wa ters 01n the Koj ima Bay 200 0.0 84.5 0.0 2.5 0.0 7.5 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0

3. Wa ters 01n the Inland Sea 416 0.0 0.0 99.5 0.5 0.0 0.0 0.0 0.0 0.0 0.0 「0.0 0.0 0.0 0.0

4. Wa ters ■1n Sasagase River 120 0.0 0.0 0.0 35.0 45.8 11.7 0.O 2.5 0.0 0.0 0.00.O P

0.0 0.0

5. Wa ters ■1n Kurashiki River 104 0.0 σ.0 0.0 1.9 2510 66.3 0.℃ 0.0 0.0 0.0 0.0 0.0 0.0 0.0

6. Wa ters ■1n rec l a i med area 204 0.0 0.0 0.0 0.0 0.0 10010 0.0 0.0 0.0 0.0 0.0 0.0 0.0 010

7. Wa ters ■1n Yoshino River 192 0.0 0.0 0.0 1.0 0.0 0.0 87.5 5.7 0.0 2.1 0.0 0.0 2.7 0.0

8. Wa ters ■1n Asahi River ㎜ O.0 1.0 0.0 0.5 0.0 0.0 22’.5 76.0 0.0 0.0 0.0 0.0 0.0 0.0

9. Paddyf ields in planting seaSOn 132 0.0 0.0 0.0 0.0 3.0 0.0 0.0 0.0 93.9 0.0 0.8 0.0 0.0 0.8

10. Urban areas 螂 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 85.5 0.0 8.1 4.8 0.0

11. Trees 304 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.7 0.0 98.0 0.0 0.0 0.3

12. Bare fields 45 0.0 0.0 0.0 2.2 0.0 0.0 0.0 0.0 0.0 6.7 0.0 22.2 66.7 0.0

13. Concre te struc tures around (x沿s t 80 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 62.5 35.0

14. Shadows 140 0.0 0.0 0.0 0.0 2.9 0.0 0.0 0.0 1.4 0.0 0.0 0.0 22.9 69.3

’Average performance=83.9% Equivbcation quantification = O.as

Page 20: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

rf7NE5i7  濫

    ’ }zi!

・Nl.5

N

q

憾ミf一

瞥蕎  N

1.5

2Z

O,5

                               Fig・一3 工sobath (m) in the Reservo1「

 VELOCITY(MIS) ト峰ト0.400 e■

工く)㎜「1盟N O.0100M/S

         ノ   る       1二,’一、・、

、こご二乞楽藁   \  一   ノーく「」二、    ヘ                 ノ

                    ノ

ー、 A  、 ! / //\_  \ イ / /

 \く、『/ !         ノ\       ’ \   噸     ’, ’  ,

、 、 ’ ! 1        ’ 、   ,

    鳴  ” ●     覧

     、    ●

・    ○

   ○    ■    .

.    o

   .   ,   ,

 ・   .    ●    .    ・

 .   .

    …  ■   ・    ●

  .   ,

    o     ●   φ

    ‘       ●     ●

く1.

蝿 ’、      3跡」層●印の一」

驚募=   ノ’t ”_,の一

    知

..ッ

髪纏..’、、” @ミミ1

糠’沁\  瓢\\   xx   N N    N    N

     .

                                          Fig ・一 4 Caluculated Velocities .in the Reservolr

一 16 一

Page 21: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

     It is very interesting to compare、the map wi th the distribution  of  calcu:Lated

velocity  (Fig 一4)●  Because  the  map represenしs waしer flowing out二side、t士lrough water

gate, so iヒ is supposed that二water from Sasagase and Kurashiki River flows into  the

        reserVOIr●

3rr6 Conclusion of appl ication

     From this test example, following conclusions can be madeラ

(1) The accuracy o:E the TM data is as well as airborne MSS data.

(2) 工f  successi▽e  TM digital data of a test;site are st二〇red, it may be’possible to

detect exactly future transition on the enviro㎜ent around theヒest site.

.(3) 工naddition, the どemote sensing’techniques may  bring  their  capabilities  into

full play’ taking  togeth〔学r with simulation model in respect of water and chemical

maしerial flow.

Photo-4 Supervised ClassiEication Dvlap (No. of Training Classes is 7)

e’ 17 ’

Page 22: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

4. Conclusion

     工し has been shown that;the Tsukusys was deVeloped mainly for  しhe  ana:Lysis  of

Landsat tvlSS data r could cope wi th the Landsat [“vr da ta. HOwever, in o rder to expand

the utilization of Landsat Tlvr data, it is necessary that the TsuKusys can sele¢t

spectral bands wi th anaiysis or rnore than 4 bands data can be processed, while the

TsukusYs presently only deals wi th 4 bands data.

     The second systern is being designed r taking notice of the selection and

eniargernent of the bands. The systern i s scheduled to process tandsat Trvl data, MOS-1

tv!ESSR data .and SpCrr XS data. When the new systern i s accornpl i shed, t he sys tern wi l l

aiso be open for education and training as well as the TsuKusys.

AcKnowledgments

     The authors would like to thank Mrs. Zhang Shaoxing Eor her enl ightening

comments and suggestions.

References

1) Hoshi,T. (1979) ; Processing System ”USASi, of MSS Dat;a from an Aircraft;and  工し。s

Application, Proc. oE Japanese Societ’ 凵@gf.Civil Eng., ’No.285r pp.69-83. (in

Japanese)

2)H。sh・,T. andエs・hama,K.(1981);lnput/Output$・ftware・fエrnag・e Pr。cessing’JAFSA

Rep. 811012, pp.207-220. (in Japanese)

一 18 一

Page 23: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

3) Hosh i,T. and Nakayarna,K. (1983).; A 1!andsa t rmage Databank Us i ng iY!ass Sto rage

System-et TrsUKUSYSie 工lnage-bank一・一’r  Inter  Graphics I 83, An  工nternational  Conf. &

Exhibition on Computer Graphic Technical SessionS, D8-i, pp.1-15.

4) Hosh i,T. and Sato,K. ( in press ) ; Analys i s of iYlang rove Forest in OKinawa Us ing

Airborne Remote Sensing Data r Proc. of 6Vh Asian Conf. on Rernote Sensig.

5) lshida,T. and Hoshi rT. (1985) ; Estirnation of Regional Evapotranspiration Using

Land Classification wap Obtained frorn Landsat D4SS Data r Proc. of Annual Conf.’85r

Soc iety of Photog rammetry and Remote Sensing, pp.95-100. (in Japanese)

6) 工shii’Y・’Nakayama’K・’and Hoshi’T・ (1983); The Analysis of  Landsat (MSS)  工mage

Dat『亀 Using  工nしeractive  Information Retrieval L尋nguage SOAR, 工nter Graphics曹83, An

エnternati・nal C・nf・&Exhibiti・n・n’ C。mputer Graphic Technical Sessi。ns’D8・一4,

pp.1一一i4.

7) Kawachi,T.  and  Kooutangkulvong r S. (1983) ; Numerical Investigation iWo

Dimensi。na・Backwaしer by㎞e Finite Element Me伽d,」.・fエrri. hng. and Rural

Planning, No.3, pp.32・一51.

8)Nagamine’K・and H・shi’T・(1982);pre-pr・cessing。f tandsat lYISSエmage Data by

Using Graphic Display, Proc. of Annual Conf.’82, Society of Photogrammetry and

Remote Sens ing, pp.61一一64. (in Japanese)

9)[DoriirK. OkagawarN.r Kltarnura,T. and Hoshi rT. (1979) ; Analysis of Water Area

around the Suwa Lake Using CCCT, Proc. of Annual Conf.’79, Soc i ety o f lnstrument and

Control Eng.t pp..111-114. (in Japanese)

IO) ToriirK. HoShi,T., and lshida,T. (1985);Flow Patterns around the Kojima

Reservo i r Us ing [IM r Proc. o£ Annual Conf.’85, Society of Photog rammetry and Rernote

Sensing r pp.IOI-104. (in Japanese)

一 19 一一

Page 24: REMOTE SENSING IMAGE DATA ANALYSIS SYSTEM ......Remote Sensing 工mage Data Analysis System ”Tsukusys” and 工tIs Appl ication 一 Espec ially ClassiEing Coastal Pllrea around

INSTITUTE OF INFORIvrATION SCIENCE AND ELECTRONICS             UNIVERSITY OF TSUKUBA

    SAKURA-MURA, NIIHARI-GUN, I BARAKI, JAPAN

REPORT NUMBERREPORT DOCU凹ENTATION PAGE

ISE-TR-86-54

TITLERemo・ヒe Sensing工mage Data Analysis Sys’ヒem 町sukusys”

and工t 1 s Application

       ●                                    ・      ● und aBa. Usin MT一AUTHOR(S)

Takashi Hbshi

1くiyoshi Torii

TQmoydd工shida

REPORT DATE NUMBER OF PAGES

January 6, 1986 19

MAIN CATEGORY CR CATEGORIES

Remote Sensin9 4.0

KEY WORDS1血㎝atic Map鉾r, Stream Pattem ,Super▽ised Classification

ABSTRACT

A system ca】」. ed as“TSukusys.。 has been(皐eveloped to

distinguish arKヨanalyze 1and use and 1and cover patterns’by

  ・                                                         .

tSlrg   remote  SenSlng image δata, especially  Landsat

Mul tispectral  Scanner (MSS)  imag e data. As」  in  Japan,

Landsat Th㎝atiC Mapper (別) image data can be acquireq     .

SinCe laSt year, the apPIication of TM dat二a by the Tsuku寧ys

was attempted. AS a reSult, the ad▽antage of t二he TM data is

c㎝Pletely vertified・ This in▽estigation is  し。  interprete

water quality condition around the K・」ima臼ay油e「e mter

pollution becomes a social ●1ssue●

SUPPしEMENTARY NOTES