Course objective:

The objective of the course is to

–          provide the specific conception of space by means of its registration by satellite images and to present new research conceptions in space analysis from satellite images and their derivative products, combined with auxiliary data such as topographic, thematic and other at varying spatial, temporal, spectral and radiometric resolutions.

–          Provide hand-on experience to student for methods of satellite data analysis by means of modern software such as IDRISI KILIMANJARO and ERDAS IMAGINE 8.7. These applications are based on satellite images of Greece and other countries.

Course description

Theoretical knowledge such as satellite platforms, multispectral satellite images, geometric and radiometric corrections, spectral signatures, methods of supervised and unsupervised classification and accuracy assessment complete the theoretical background. Practical experience in management and analysis of satellite data is realized by means of laboratory exercises on personal computers, with modern software and satellite imagery.

In detail:

• Display the spectral channels of satellite images.
• Structure of digital images. Spatial location of pixels and spatial unities.
• Creation of color palettes and application on satellite images.
• Color graduation of digital images.
• Color composites of spectral channels.
• Histograms of digital images and analysis.
• Manual classification of digital images, based on histograms and spatial investigation of radiometric values of a satellite image.
• Algebraic and logical operations on digital images.
• Linear regression of satellite images.
• Visual location of Ground Control Points in satellite images and satellite image overlaying.
• Geodetic transformation of satellite images.
• Spectral signatures: location of polygons per land cover class, spectral signatures extraction and mutual comparisons.
• Supervised and Unsupervised classifications of satellite images.
• Assessment of the accuracy of the classification based on the principles of accuracy matrix.
• Introduction of neural networks for the classification of remotely sensed data
• Fusion of remote sensing and Geographical Information Systems (GIS)

Introduction to neural networks in remote sensing. The simplest neural network, known as the perceptron, is introduced. The structure and different components (nodes, links, weights etc) as well as the function of the perceptron (summation function, threshold function etc) are explained. The students experiment in solving simple classification problems using the perceptron. The discussion of results includes the decision line that is formed and the shape it can take for perceptrons with or without bias. The limitations of perceptrons are discovered by experiment with classification problems that are not linearly separable (e.g. the XOR problem). The multilayer perceptron is the briefly introduced. The practical exercise includes a case study with real remote sensing data. The objective is to use the multilayer perceptron (MLP) with the standard backpropagation training algorithm and compare several configurations (learning rate, topologies, training termination criterion etc) in order to maximize classification accuracy.

Remote sensing and GIS fusion. Remotely sensed data are often used as an additional source in geographical information systems. Usually the satellite data have to classified and the output is entered to the GIS system. Satellite images are also used for topographic map update. The reverse information flow, from the geographical information system towards the satellite data, is also frequently realized when some GIS layers are used as additional parameters (ancillary information) in the classification of remotely sensed data. A typical example is the use of topographic information (DEM, slope, aspect) to perform a classification for land use mapping.

Background required

Knowledge on PC use and on basic statistics

Course presentation

The theoretical and methodological parts of this course are presented in different conferences. The application part is realized by laboratory exercises using satellite images as well as state of the art remote sensing software.Recent references research is demanded especially in scientific journals (rf. References)

Schedule

Bibliography :

1) Lillesand, M.T. and Kiefer, W.R. Remote sensing and image interpretation, 3rd Edit. J. New York: Wiley and Sons Inc, 1994.

2)                   Stathakis D. and K. Perakis, Feature evolution for classification of remotely sensed images, IEEE Geoscience and Remote Sensing Letters, accepted.

3)                   Lippmann R. P., An introduction to computing with neural nets”, IEEE ASSP Magazine, vol. 2, pp. 4 – 22, 1987

4)                   Kanellopoulos I. and G. G. Wilkinson, Strategies and best practice for Neural Network image classification, International Journal of Remote Sensing, vol. 18, no. 4, pp. 711-725, 1997

5)                   Richards, A.J., Remote sensing digital image analysis, Springer-Verlag, 1995

Theoretical knowledge counts 20% of the total grade and laboratory exercises and applications 80%. Theory examination is written and laboratory exercises are oral or realized by means of specific software on PC. During the academic semester the use of diskettes or flash disks for data backup is demanded. Voluntary or directed research or computational exercise, theoretical or laboratory application, will contribute to the total grade. This exercise is OPTIONAL. A bonus to students present in every course will be given as an optional written test. Attendance and performance will be taken in consideration