igarss 2011.ppt

Archaeological Land Use Characterization Archaeological Land Use Characterization using Multispectral Remote Sensing Datausing Multispectral Remote Sensing Data

Dr. Iván Esteban Villalón Turrubiates, Dr. Iván Esteban Villalón Turrubiates, Member,Member, IEEE IEEE María de Jesús Llovera TorresMaría de Jesús Llovera Torres

UNIVERSIDAD DE GUADALAJARAUNIVERSIDAD DE GUADALAJARACENTRO UNIVERSITARIO DE LOS VALLESCENTRO UNIVERSITARIO DE LOS VALLES

Monitoring Hidrological Variations using Multispectral Monitoring Hidrological Variations using Multispectral SPOT-5 Data: Regional Case of Jalisco in MexicoSPOT-5 Data: Regional Case of Jalisco in Mexico

Dr. Iván Esteban Villalón Turrubiates, Dr. Iván Esteban Villalón Turrubiates, Member,Member, IEEE IEEE

OverviewOverview

- AbstractAbstract- Remote Sensing DefinitionRemote Sensing Definition- Sensor ResolutionSensor Resolution- Introduction to Image ClassificationIntroduction to Image Classification- Model FormalismModel Formalism- Verification ProtocolsVerification Protocols- Simulation ExperimentsSimulation Experiments- Concluding RemarksConcluding Remarks

AbstractAbstractProposition - A new and efficient classification approach of remote sensing signatures extracted from large-scale multispectral imagery.

Contribution - This approach exploits the idea of combining the spectral signatures from a remote sensing image to perform a novel and accurate classification technique.

Verification - Simulation results are provided to verify the efficiency of the proposed approach.

REMOTE SENSING REMOTE SENSING DEFINITIONDEFINITION

Remote SensingRemote SensingRemote Sensing can be defined as:

"The arte and science to obtain data from an object avoiding direct contact with it” (Jensen 2000).

There is a transmission medium involved?

Remote SensingRemote SensingOf the Environment:

… is the collection of information regarding our Planet surface and its phenomena involving sensors that are not in direct contact with the studied area.

The main focus is in recollected information from a spatial perspective throughout electromagnetic radiation transmission.

Remote SensingRemote Sensing

Sensor election.

Reception, storage and digital signal processing of the data.

Analysis of the resulting information.

A) Illumination Source

B) Radiation

C) Interaction with the object

D) Radiation sensing

E) Transmission, reception and data processing

F) Analysis and interpretation

G) Application

ProcessProcess

SENSOR RESOLUTIONSENSOR RESOLUTION

ResolutionResolution All remote sensing systems use four types of

resolution:

Spatial

Spectral

Temporal

Radiometric

Spatial ResolutionSpatial Resolution

Spectral ResolutionSpectral Resolution

Time

July 1 July 12 July 23 August 3

11 days

16 days

July 2 July 8 August 3

Temporal ResolutionTemporal Resolution

6-bits Range0 63

8-bits Range0 255

010-bits Range

1023

Radiometric ResolutionRadiometric Resolution

INTRODUCTION TO IMAGE INTRODUCTION TO IMAGE CLASSIFICATIONCLASSIFICATION

Image ClassificationImage Classification Why classify?

Make sense of a landscape Place landscape into categories (classes)Forest, Agriculture, Water, Soil, etc.

Classification scheme = structure of classes Depends on needs of users.

Typical usesTypical uses Provide context

Landscape planning or assessment Research projects Natural resources management Archaeological studies

Drive models Meteorology Biodiversity Water distribution Land use

Example: Near Mary’s PeakExample: Near Mary’s Peak•Derived from a 1988 Landsat TM image

•Distinguish types of forest

Open

Semi-open

Broadleaf

Mixed

Young Conifer

Mature Conifer

Old Conifer

Legend

Classification: Critical PointClassification: Critical Point LAND COVER not necessarily equivalent to LAND USE We focus on what’s there: LAND COVER Many users are interested in how what is there is being

used: LAND USE

Example Grass is land cover; pasture and recreational parks are

land uses of grass

Basic Strategy: How to do it? Basic Strategy: How to do it? Use radiometric properties of remote sensor Different objects have different spectral signatures

In an easy world, all “vegetation” pixels would have exactly the same spectral signature.

Then we could just say that any pixel in an image with that signature was vegetation.

We could do the same for soil, water, etc. to end up with a map of classes.

Basic Strategy: How to do it? Basic Strategy: How to do it?

But in reality, that is not the case. Looking at several pixels with vegetation, you’d see variety in spectral signatures.

The same would happen for other types of pixels, as well.

Basic Strategy: How to do it? Basic Strategy: How to do it?

The Classification Trick: The Classification Trick: Deal with variabilityDeal with variability

•Different ways of dealing with the variability lead to different ways of classifying images.

•To talk about this, we need to look at spectral signatures a little differently.

Think of a pixel’s brightness in a 2-Dimensional space. The pixel occupies a point in that space.

The vegetation pixel and the soil pixel occupy different

points in a 2-D space.

With variability, the vegetation pixels now

occupy a region, not a point, of n-Dimensional space.

Soil pixels occupy a different region of n-Dimensional space.

• Classification: • Delineate boundaries of classes in n-dimensional space• Assign class names to pixels using those boundaries

Basic Strategy: Basic Strategy: Deal with variabilityDeal with variability

Classification StrategiesClassification StrategiesTwo basic strategies:

Supervised Classification We impose our perceptions on the spectral data.

Unsupervised Classification Spectral data imposes constraints on our interpretation.

Digital Image

Supervised ClassificationSupervised Classification

The computer then creates...

Supervised classification requires the analyst to select training areas where he knows what is

on the ground and then digitize a polygon within that area…

Mean Spectral Signatures

Known Conifer Area

Known Water Area

Known Deciduous Area

Conifer

Deciduous

Water

Multispectral ImageInformation

(Classified Image)

Mean Spectral Signatures

Spectral Signature of Next Pixel to be

Classified

Conifer

Deciduous

Water Unknown

Supervised ClassificationSupervised Classification

Water

Conifer

Deciduous

Legend:

Land Cover Map

The Result: Image SignaturesThe Result: Image Signatures

Unsupervised ClassificationUnsupervised Classification In unsupervised classification, the spectral data imposes constraints on our interpretation.

How? Rather than defining training sets and carving out pieces of n-Dimensional space, we define no classes beforehand and instead use statistical approaches to divide the n-Dimensional space into clusters with the best separation.

After the fact, we assign class names to those clusters.

Unsupervised ClassificationUnsupervised Classification

Digital Image

The analyst requests the computer to examine the image and extract a number of spectrally distinct

clusters… Spectrally Distinct Clusters

Cluster 3

Cluster 5

Cluster 1

Cluster 6

Cluster 2

Cluster 4

Saved Clusters

Cluster 3

Cluster 5

Cluster 1

Cluster 6

Cluster 2

Cluster 4

Unsupervised ClassificationUnsupervised ClassificationOutput Classified Image

Unknown

Next Pixel to be Classified

Unsupervised ClassificationUnsupervised Classification

Conif.

Hardw.

Water

Land Cover Map Legend

Water

Water

Conifer

Conifer

Hardwood

Hardwood

Labels

It is a simple process to regroup (recode) the clusters into

meaningful information classes (the legend).

The result is essentially the same as that of the

supervised classification:

MODEL FORMALISMMODEL FORMALISM

Multispectral ImagingMultispectral Imaging Is a technology originally developed for space-based imaging.

Multispectral images are the main type of images acquired by remote sensing radiometers.

Usually, remote sensing systems have from 3 to 7 radiometers; each one acquires one digital image in a small band of visible spectra, ranging 450 to 690 nm, called red-green-blue (RGB) regions: Blue -> 450-520 nm. Green -> 520-600 nm. Red -> 600-690 nm.

The combination of the RGB spectral bands generates the so-called True-Color RS images.

Statistical Approach.

Assume normal distributions of pixels within classes.

For each class, build a discriminant function For each pixel in the image, this function calculates the

probability that the pixel is a member of that class. Takes into account mean and variance of training set.

Each pixel is assigned to the class for which it has the highest probability of membership.

Weighted Pixel Statistics MethodWeighted Pixel Statistics Method

Blue Green Red Near-IR Mid-IR

Mean Signature 1

Candidate Pixel

Mean Signature 2

It appears that the candidate pixel is closest to Signature 1. However, when

we consider the variance around the signatures…

Rel

ativ

e R

efle

ctan

ce

Weighted Pixel Statistics MethodWeighted Pixel Statistics Method

Blue Green Red Near-IR Mid-IR

Mean Signature 1

Candidate Pixel

Mean Signature 2

The candidate pixel clearly belongs to the signature 2 group.

Rel

ativ

e R

efle

ctan

ce

Weighted Pixel Statistics MethodWeighted Pixel Statistics Method

Weighted Pixel Statistics MethodWeighted Pixel Statistics Method

Weighted Pixel Statistics MethodWeighted Pixel Statistics Method

VERIFICATION PROTOCOLSVERIFICATION PROTOCOLS

Verification ProtocolsVerification ProtocolsA set of three synthesized images are used as verification protocols.

All synthesized images are True-Color (RGB), presented in 1024-by-1024 pixels (TIFF format).

Each synthesized image contains three different regions (in yellow, blue and black colors) with a different pattern.

The developed Weighted Pixel Statistics (WPS) algorithm is compared with the most traditional Weighted Order Statistics (WOS) method [S.W. Perry, H.S. Wong, 2002].

Results:Results:11stst Synthesized Scene Synthesized Scene

Synthesized SceneSynthesized Scene WOS ClassificationWOS Classification WPS ClassificationWPS Classification

Quantitative ComparisonQuantitative Comparison11stst Synthesized Scene Synthesized Scene

Results:Results:22ndnd Synthesized Scene Synthesized Scene

Synthesized SceneSynthesized Scene WOS ClassificationWOS Classification WPS ClassificationWPS Classification

Qualitative ComparisonQualitative Comparison22ndnd Synthesized Scene Synthesized Scene

Synthesized SceneSynthesized Scene WOS ClassificationWOS Classification WPS ClassificationWPS Classification

Quantitative ComparisonQuantitative Comparison22ndnd Synthesized Scene Synthesized Scene

Results:Results:33rdrd Synthesized SceneSynthesized Scene

Synthesized SceneSynthesized Scene WOS ClassificationWOS Classification WPS ClassificationWPS Classification

Qualitative ComparisonQualitative Comparison33rdrd Synthesized Scene Synthesized Scene

Synthesized SceneSynthesized Scene WOS ClassificationWOS Classification WPS ClassificationWPS Classification

Quantitative ComparisonQuantitative Comparison33rdrd Synthesized Scene Synthesized Scene

RemarksRemarksThe quantitative study is performed calculating the classified percentage obtained with the WOS and WPS methods, respectively.

The WOS method uses only 1 spectral band.

The WPS method uses the information from the three spectral bands to analyze the pixel-level neighborhood means and variances.

The results shows a more accurate and less smoothed identification of the classes.

SIMULATION EXPERIMENTSSIMULATION EXPERIMENTS

Archaeological Land UseArchaeological Land UseA Remote Sensing Signatures (RSS) electronic map is extracted from the multispectral image. Three level RSS are selected for this particular simulation process, defined as:

██ – Archaeological land use zones.

██ – Modern land use zones.

██ – Natural land cover zones.

██ – Unclassified zones.

Archaeological SiteArchaeological Site"Guachimontones", Jalisco Mexico"Guachimontones", Jalisco Mexico

Simulation ResultsSimulation ResultsScene from "Guachimontones"Scene from "Guachimontones"

Original SceneOriginal Scene WPS ClassificationWPS Classification

Hidrological VariationsHidrological VariationsA Remote Sensing Signatures (RSS) electronic map is extracted from the multispectral image. Three level RSS are selected for this particular simulation process, defined as:

██ – Humid zones.

██ – Dry zones.

██ – Wet zones.

██ – Unclassified zones.

Simulation ResultsSimulation ResultsScene from "La Vega" dam, Jalisco MexicoScene from "La Vega" dam, Jalisco Mexico

Original SceneOriginal Scene WPS ClassificationWPS Classification

CONCLUDING REMARKSCONCLUDING REMARKS

RemarksRemarksThe WOS classifier generates several unclassified zones because it uses only one spectral band in the classification process.

The WPS classifier provides a high-accurate classification without unclassified zones because it uses more robust information in the processing.

The qualitative and quantitative analysis probe the efficiency of the proposed approach.

Future WorkFuture WorkComparison with several classification techniques.

A more extensive performance analysis of the proposed approach with different synthesized images.

Application to remote sensing imagery and the study of its performance.

Hardware implementation of the proposed approach.

Dr. Iván Esteban Villalón Turrubiates, Dr. Iván Esteban Villalón Turrubiates, Member,Member, IEEEIEEE

UNIVERSIDAD DE GUADALAJARAUNIVERSIDAD DE GUADALAJARACENTRO UNIVERSITARIO DE LOS VALLESCENTRO UNIVERSITARIO DE LOS VALLES

THANK YOU!THANK YOU!Questions?Questions?