Remote Sensing

Satellite Introduction to the Remote Sensing System

Remote Sensing is the art science and technology to extract reliable information from images. The information extraction process is based on remote sensing methods while the reliability analysis of the information involves ground samples or ground truth combined with mathematical and statistical analysis.

Images used in remote sensing are all kinds of pictures in any form such as photographs using any camera, aerial photographs, video images, multichannel images, sonar images, radar images, x-ray images, magnetometry etc. Information extracted by remote sensing methods is related to the quality and the attributes of the object, while information related to the precise measurements on geometric elements of the object such as dimensions, horizontal coordinates, elevations, area etc. is performed by Photogrammetry.

Platforms used to locate image sensors and obtain pictures could be located: (a) on ground (vista points, cranes, high buildings), (b) in air (airplane, helicopters, balloons) and (c ) in space (rocket, spacecraft and satellite systems).

Pictures can be used in analog or digital form. A complete definition of a picture may be as follows: «the presentation of an object in a two dimensional surface where each point of the surface with coordinates (x, y) is given one gray scale value for single channel image or black and white image or a vector of grayscale values for multichannel images». If the points on the picture are of a finite number and they are ordered in rows and columns the picture is called digital picture or raster image. If the image points are infinite then the picture is an analog picture i.e. the classical photography. It must be noted that a three-channel picture can be presented using the three primary colors Red, Green and Blue (RGB). If the three colors of presentation (RGB) coincide with the true colors of the subject, then it is called a color image, other wise it is called a pseudocolor image. The raster cell of a digital picture is also called pixel (picture element) and its ground dimensions define the spatial resolution of the image, which represents the minimum size of the object to be identified in the image. The number of channels in an image defines its spectral resolution and help to compose the spectral signatures of objects to be identified. The number of grayscale values of a particular channel in an image are called also digital numbers and define the radiometric resolution which if expressed in bit codes it is known as dynamic range i.e. 3-bit image has 8 (third power of two) gray scale values while 10-bit image has 1024 (tenth power of two) grayscale values.

Remote Sensing methods can be distinguished into four groups as follows:

(a) Image interpretation or photointerpretation.

(b) Image Classification

(c) Regression

(d) Expert systems

The image interpretation is based on the creation of a good quality enhanced image and extraction of information is performed directly by the human eye. The image interpreter looks at a single image or stereo image and based on experience and proper training draws polygon lines over Arial information i.e. land cover, or lines over linear features i.e. drainage features. Extracted information is then digitized and transferred into a GIS system. Photointerpretation can be performed on hard copy images or digital images over a TV screen. Features of digital images can be digitized directly by a mouse or line following algorithms. Image enhancement targets to the contrast increase between the information to be extracted and its background. For photographic analog images prints with increased contrast are prepared while for digital images radiometric enhancement, filtering and other processing techniques are used.

Image classification is performed on multichannel digital images. There are two basic procedures called supervised classification and unsupervised classification accordingly. Both procedures are performed by the system’s software and each pixel on the image is classified to one class out of a set of classes using certain criteria such as: maximum likelihood, minimum Euclidean distance, parallelepiped etc. Classes in supervised classification are specifically known and they are identified by ground samples over limited areas. Classes in unsupervised classification are not precisely known and they are defined by the system. The idea of multispectral classification is to obtain spectral signatures of a set of classes and then pick up one by one each pixel in the image, determine its signature and classify the pixel in the class category whose signature matches better. Signature matching algorithms are based on criteria, as mentioned above, and one widely known of such criteria is the maximum likelihood.

The method of regression can be performed on single channel or multichannel images and it is based on the correlation between grayscale values of spectral bands and the class to be identified. A good example is the estimation of chlorophyll in the sea environment which is correlated to the Landsat three visible spectral bands with a correlation model such as:

Y = a + bR1 + cR2 + dR3

Where Y is the chlorophyll concentration, a, b, c, are coefficients which are determined by the regression, R1, R2, R3 are gray scale values of one Landsat pixel in three visible spectral bands. The process requires ground samples on more than three sea stations at the same time the satellite is taking the image and in the same location of the corresponding Landsat pixel.

Expert systems are dealing with the development of knowledge of specific feature extraction. Such processes are based on the patterns of certain features, the interrelation among feature attributes and the interrelation between neighboring features. Expert systems make efficient utilization of all methods used to extract reliable information from images.

Most Remote Sensing results are tested to meet certain reliability criteria using ground truth samples and then are entered in raster or vector form into the GIS system.

© 1999 TRIANET, Program of the European Union Socrates-Comenius
Last update on 04.05.1999 by Markus Zapke-Gründemann