mathematical analysis of satellite images · 2017-01-08 · mathematical analysis of satellite...

Mathematical analysis of satellite images

Part 1

István László*

István Fekete**

Summer School in Mathematics Institute of Mathematics, Eötvös Loránd University,

Budapest, Hungary, June 6 - 10, 2016

* Institute of Geodesy, Cartography and Remote Sensing http://www.fomi.hu

** Eötvös Loránd University, Faculty of Informatics http://www.elte.hu

http://www.fomi.hu/

http://www.elte.hu/

http://www.elte.hu/

http://www.elte.hu/

Contents

2 Mathematical analysis of satellite images

Summer School in Mathematics, ELTE, Budapest, 06-10 June 2016

•Part 1: An overview of remote sensing

• 1. Introduction: the evolution of remote sensing

• 2. Raw material: acquisition and pre-processing

• 3. Evaluation: only visual approach?

• 4. Practical applications

• 5. Education and university collaboration

•Part 2: Advanced methodology

• 1. Numerical evaluation of satellite images

• 2. The whole process of evaluation

• 3. Segment-based thematic classification

• 4. Data fusion

• 5. Object-based image analysis (OBIA)

1. Introduction: the evolution

of remote sensing



Needs?

- The exhausting of local and global resources (first: oil)

- Global problems

- The extinction of species

- Club of Rome

Therefore natural resources should be managed, based on exact survey!

Opportunities?

- Space technology

- Sensors

- High speed data transfer

- 1972: launch of the first LANDSAT (ERTS) satellite

- Fast computers with graphical capabilities

- Digital image processing



Remote sensing:

from data acquisition to thematic evaluation

The 3 main

components of remote

sensing system



The parts of optical wavelength interval



The first series

of Landsat

satellites

(1, 2, 3)

(1972-1983)



SPOT 5 HRG

sensor



Very high resolution satellite images

(0,5 m x 0,5 m – 4 m x 4 m-es ground resolution)

IKONOS satellite image



A part of IKONOS multispectral satellite image (4 m) Sapporo, Japan, 1999 October 6 Space Imaging



The first Hungarian satellite: MASAT-1 http://cubesat.bme.hu

2012 February 13 – 2015 January 9



Long-term European programme: Copernicus

…and Sentinel satellites

Sentinel-1A (2014. 04. 03.): weather independent, radar sensor

Sentinel-2A (2015. 06. 23.): high resolution multispectral optical sensor



2. Raw material:

data acquisition and pre-processing



• Passive, reflective systems:

- Sun is the source of electromagnetic (EM) radiation

- Sampling “windows” of the whole EM spectrum

The physical background of remote sensing

Different land covers reflect differently:

- crops, - water, - soil Sensors measure the intensity of electromagnetic

radiation arriving from the Earth’s surface



Spectral response curves

What are the acquisition bands used for?



The diversity of remote sensing - Carrier

- Height

- Time

- Sensors

- Wavelength

Satellites: Geostationary orbit: 36 000 km

(Near) polar orbit: 450-1000 km

Airplanes: 300 m-10 km

Drones: UAV – Unmanned Aerial

Vehicle or RPAS - Remotely

Piloted Aircraft Systems:

30-600 m

Gliders: 100-300 m

Terrestrial observation:

5 m



Scanning acquisition:



What does a remote sensing image contain?

Pixels <--> elementary pieces of surface

Band values within a pixel



The main parameters

of remote sensing systems Spatial:

- Spatial coverage

- Ground resolution

Spectral:

- Spectral resolution

- Radiometric resolution

Temporal:

- Temporal resolution (cycle length)

- Other factors: data access (availability, speed, price)



Scene ID Date

1333 2012.08.06.

8364 2012.08.18.

8481 2012.08.18.

4565 2012.08.25.

The mosaicking of image tiles



Geometric correction

It is clear that an accurate Digital Elevation Model is inevitable.

Scene 8481, RGB 423 Scene 8364, RGB 412



Two complete coverages of VHR images (Pléiades)

Pléiades, 2013.05.18. + Pléiades, 2013.07.17.



Pixel-based fusion („merge”)

- Pricipal Component Analysis (PCA)

- Modified Intensity-Hue-Saturation Merge (MIHS)

- In-house developed fusion, based on high pass filter (HPF)



3. Evaluation:

only visual approach?



The thematic evaluation of remote sensing images

2003. 03. 29., Landsat 7 ETM+

2003. 07. 27., Landsat 5 TM

A part of a crop map

Őszi búza Tavaszi árpa

Őszi árpa Kukorica Silókukorica Napraforgó Cukorrépa

Lucerna Vízfelszínek Nem mezőgazdasági területek Egyéb szántóföldi növények



The diversity of thematic categories

Őszi búza

Tavaszi árpa

Őszi árpa

Kukorica

Silókukorica

Napraforgó

Cukorrépa

Lucerna

Vízfelszínek

Nem mezőgazdasági területek

Egyéb szántóföldi növények



Texture: the regular changes of intensity values



Feladat Vizuális interpretáció

(szem + agy rendszer)

Számítógépes

kiértékelő rendszer

Geometriai összefüggések, struktúrák

felismerése

kitűnő gyenge

Textúra felismerése, azonosítása jó gyenge

Textúra mérése gyenge kitűnő

Tónusok elkülönítése közepes kitűnő

Megbízhatóság, objektivitás,

reprodukálhatóság

közepes jó

Feldolgozási sebesség gyenge kitűnő

Bonyolult szakértelem, egyéb ismeretek

alkalmazása

jó közepes

Több adatforrás vagy időpont együttes

kiértékelése

gyenge kitűnő

The two basic methods of remote sensing data evaluation:

visual interpretation

and digital image processing



3.1. Numerical evaluation



The basic task of image processing

and the elementary solutions of classification

Pixels belonging to categories

The intensity vectors of categories

show a typical probability distribution

in certain parts of intensity space



Clusters and thematic categories



3.2. Visual evaluation



Application in remote sensing subsidy control (CwRS)

Sensor Date

Ikonos 2013.06.21.

Ikonos 2013.07.02.

GeoEye 2013.07.03.

GeoEye 2013.07.06.



CwRS, Example#1: tree density counting

Ikonos, 2012.07.02. Pléiades, 2012.08.18.



CwRS, Example#2: the detection of sunflower

GeoEye, 2012.07.06. Pléiades, 2012.08.18.



CwRS, Example#3: cereal stubble (weed-free)

Ikonos, 2012.07.02. Pléiades, 2012.08.18.



3.3. Combined solutions



The detection of weed-infected areas

Weed infection in soybean parcels, detected with the quantitative evaluation of NDVI

map. The grades of weed infection can be observed with Pleiades images on

soybean stubble. The extent of infection can be well measured within parcels.



Automatic delineation of forest areas Object-based Image Analysis (OBIA)



The quantitative comparison of Pleiades images

2012.08.06. 2012.08.18.:

Left to right:

difference, 2012.08.06, 2012.08.18.



4. Practical applications



4.1. The National Operational

Crop Monitoring and Production Forecast

Programme (CROPMON)



26th EARSeL Symposium May 29 – June 2, 2006, Warsaw

Data Flow in CROPMON

CROPMON

INFORMATION

EXTRACTION

SYSTEM

Precalibration,

historical data

Reference

data

Low resolution

satellite data

High resolution

satellite data

Crop maps and

area figures

Development

assessment

Yield forecast

Production Reports

W in ter w heat

L egend

Spring barley

W in ter barley

M aize

Silage m aize

Sunflow erSugarbeet

A lfalfa

Peas

O ther cereals

Spring fodder crops

1. Megye Őszi búza

(ha)

Őszi árpa

(ha)

Tavaszi árpa

(ha)

2. Pest, Budapest 69 694 13 522 7 871

Közép-Magyarország 69 694 13 522 7 871

Fejér 82 809 8 603 4 659

Komárom-Esztergom 30 598 5 621 4 744

Veszprém 36 982 15 751 9 654

Közép-Dunántúl 150 389 29 975 19 057

Győr-Moson-Sopron 68 062 13 965 24 257

Vas 39 011 7 456 13 853

Zala 22 241 7 441 6 030

Nyugat-Dunántúl 129 314 28 862 44 140

Baranya 55 873 13 734 5 959

Somogy 50 241 11 666 2 018

Tolna 54 666 10 264 1 965

Dél-Dunántúl 160 780 35 664 9 942

Borsod-Abaúj-Zemplén 58 269 5 249 20 885

Heves 52 188 6 397 10 906

Nógrád 22 031 2 040 6 673

Észak-Magyarország 132 488 13 686 38 464

Hajdú-Bihar 68 156 8 238 8 487

Jász-Nagykun-Szolnok 116 323 14 016 15 198

Szabolcs-Szatmár-Bereg 36 212 5 087 3 284

Észak-Alföld 220 691 27 341 26 969

Bács-Kiskun 93 202 27 864 6 043

Békés 124 146 21 279 3 893

Csongrád 70 870 17 375 2 930

Dél-Alföld 288 218 66 518 12 866

1. Megye Őszi búza

(kg/ha)

Őszi árpa

(kg/ha)

Tavaszi árpa

(kg/ha)

2. Pest, Budapest 3 650 3 146 2 540

Közép-Magyarország 3 650 3 146 2 540

Fejér 4 180 3 802 3 298

Komárom-Esztergom 3 909 3 486 2 834

Veszprém 3 760 3 407 3 031

Közép-Dunántúl 4 022 3 535 3 047

Győr-Moson-Sopron 3 712 3 378 3 307

Vas 3 626 3 250 3 245

Zala 3 861 3 610 3 152

Nyugat-Dunántúl 3 712 3 405 3 266

Baranya 4 346 3 867 2 934

Somogy 3 718 3 572 2 959

Tolna 4 179 3 957 3 040

Dél-Dunántúl 4 093 3 796 2 960

Borsod-Abaúj-Zemplén 3 328 2 912 2 721

Heves 3 116 2 977 2 614

Nógrád 3 193 2 841 2 541

Észak-Magyarország 3 222 2 932 2 659

Hajdú-Bihar 3 681 3 148 2 396

Jász-Nagykun-Szolnok 3 365 3 261 2 572

Szabolcs-Szatmár-Bereg 3 636 3 156 2 663

Észak-Alföld 3 507 3 207 2 528

Bács-Kiskun 3 700 3 207 2 234

Békés 3 461 3 265 2 387

Csongrád 3 421 2 976 2 582

Dél-Alföld 3 528 3 165 2 360

Drought Alert

Reports



26th EARSeL Symposium May 29 – June 2, 2006, Warsaw

• Basics:

• combination of spatial with

spectral/temporal information

(high res. + AVHRR)

• NOAA AVHRR images

and crop maps =>

crop specific temporal profiles

• Features:

• generic: works for several crops

• year independent

• area independent

• reliable, accurate, timely

The FÖMI RSC crop yield forecast model

Part of a crop map with 1.1 km grid overlay,

corresponding to the NOAA AVHRR pixel size

Corresponding subset of a NOAA AVHRR colour composite (211 RGB)

good

bad

bad



Area measurement: Crop maps –

detailed analysis at pixel level

Kukorica megyei területadatok 1991-2000

FÖMI TK (távérzékeléssel mérve) - KSH adat

R2 = 0,91

20000

40000

60000

80000

100000

120000

140000

160000

20000 40000 60000 80000 100000 120000 140000 160000

FÖ MI TK területadatok (ha)

KS

H t

erü

leta

da

tok

(h

a)

Őszi búza megyei területadatok 1991-2000

FÖMI TK (távérzékeléssel mérve) - KSH adat

R2 = 0,97

0

20000

40000

60000

80000

100000

120000

140000

160000

180000

0 20000 40000 60000 80000 100000 120000 140000 160000 180000

FÖ MI TK területadatok (ha)

KS

H t

erü

leta

da

tok

(h

a)



4.2. Waterlog and flood monitoring



Monitoring the impact of waterlog

using satellite image time series



Waterlog maps derived from IRS WIFS

medium resolution satellite data

• good overview at country and at county level

• frequent (3-4 days), good for change detection

• cover the whole country with low cost

• quick processing



Monitoring the change of waterlog

March 12, 1999 March 3, 1999

182 ha 86 ha

1384 ha 596 ha

38 ha 17 ha

The dynamics of waterlog

and its impact can be

quantified. The affected

areas can be assessed by

villages as well.

standing water

saturated soil

crop in water

no waterlog

natural water bodies



Real-time flood monitoring, 2001



4.3. The building up

and maintenance

of Land Parcel Identification System (LPIS; MePAR in Hungarian)



Orthophoto 2011, MePAR 2012


Review and change management of physical blocks

Example #1 The reduction of eligible area because of road construction




Example #2 The reduction of eligible area because of urban development






Example #3



The reduction of eligible area because of the extension of industrial area (opencast)



The MePAR land cover system

Scale 1 : 5000



4.4. Control with Remote Sensing

of Agricultural Subsidies



The comparison of claims and real situation:

• Cultivated crop

• Parcel area

• Good Agricultural and Environmental Conditions (GAEC)

Remote sensing control of area-based agricultural subsidies

Claims

(electronic)

Control in GIS

using satellite images

Result: control

documents

(electronic)



Spring 1

Crop identification

Satellite images

Hig

h r

es

olu

tio

n

(10-2

5m

) ti

me

se

rie

s

Ve

ry h

igh

re

so

luti

on

(0,5

-1m

)

Spring 2 Summer 1 Summer 2

SPOT 2 XS

SPOT 4/5/6/7 Xi

Landsat 5/7/8 (E)TM

IRS-1C/D/P6/R2 LISS

RapidEye

VHR

Area measurement

Ikonos

QuickBird

Pléiades 1A/1B

GeoEye

WorldView 1/2/3

Basic data of CwRS:

high and very high resolution satellite images (HR, VHR)

CwRS

central

database



Computer-aided Photo-interpretation (CAPI)

with GIS software developed within FÖMI

Digitised parcel drawing

Claim

database

High resolution satellite

image time series for

crop determination

Very high resolution (VHR)

satellite images for area

measurement



Different crop types can be distinguished at parcel

level using Very High Resolution images

rape seed

cereal

row crop

cereal

rape seed

rape seed



The protection of permanent grassland: at least one mowing per year or regular

grazing

Control of minimum cultivation practice



2009-04-22, SPOT4 2009-05-20, VHR 2009-07-24, SPOT4 2009-08-17, SPOT5

Encroachment of unwanted weeds must be prevented

Control of minimum cultivation practice



GAEC - the checking of prevention of soil erosion with DEM+GIS (DEMsteep parcels, CAPIparcels with row crops, intersection: problem!)



5. Education and

university collaboration



• 1983-84: Joint development of a system evaluating satellite images

• 1985-2002: Occasional collaborations, joint publications

• 1999, 2005: Segment-based classification appears in PhD projects

• 2003: The establishment of Faculty of Informatics. Further joint research

projects start

• 2004: The launching of Geospatial Information Systems educational

module. It contains the subject Analysis of Remote Sensing Images

(2+2), maintained jointly by ELTE and FÖMI

• 2006-2011: Twelve students attended at cooperative education in FÖMI

Collaboration in research and education

between ELTE and FÖMI



Education: the course

„Analysis of Remote Sensing Images”

- Started in 2005, 5th year MSc students. The curriculum contains a series

of about 400 slides within 15 lectures.

- Presentation (I. László, FÖMI) includes the theoretical background of

remote sensing (pre-processing, image analysis, statistical classification)

and covers wide variety of practical applications.

- Lab seminars (R. Giachetta, ELTE): students implement programming

tasks in connection with remote sensing (transformations, filtering,

segmentation, clustering, classification).



Innovative benefits of collaboration

- Main contact point: Department of Algorithms and Applications,

István Fekete assoc. prof.

- Joint research, development and educational results

- Students gain insight into current operational applications

- Introduction of new scientific results into projects

- Students may get a professional practice at FÖMI

- Rising generation of highly educated colleagues



Thank you for your attention!



mathematical analysis of satellite images · 2017-01-08 · mathematical analysis of satellite...

Documents