Standard Article

Processing and Classification of Satellite Images

Remote Sensing

  1. Professor Graeme G. Wilkinson

Published Online: 15 SEP 2006

DOI: 10.1002/9780470027318.a2318

Encyclopedia of Analytical Chemistry

Encyclopedia of Analytical Chemistry

How to Cite

Wilkinson, G. G. 2006. Processing and Classification of Satellite Images. Encyclopedia of Analytical Chemistry. .

Author Information

  1. Kingston University, Kingston Upon Thames, UK

Publication History

  1. Published Online: 15 SEP 2006


Earth-observing satellites primarily gather information about the environment as digital images acquired by multispectral imaging sensors. These images consist of large arrays of picture elements (pixels). Images are usually recorded in 3–7 spectral bands, though new generation sensors providing 18–36 spectral bands are in development. The spectral bands are located in the visible, infrared, millimeter, and microwave parts of the electromagnetic spectrum. The spatial resolutions of images at the subsatellite point at ground level (dictated by the sensor's instantaneous field of view, IFOV) are principally in the range 10 m–1 km depending on the satellite. Very-high-resolution satellite-borne sensors providing imagery with resolution as high as 1 m will soon become routinely available. Images are primarily used for mapping Earth resources and monitoring the state of the environment (terrestrial, atmospheric, and oceanic). Pixels consist of digitized radiances which are usually quantized on a scale of 0–255. Interpretation of surface features requires the images to be processed and classified by computer. The processing involves several stages, typically: (1) geometrical rectification and adjustment to a suitable cartographic coordinate system, (2) correction and calibration for atmospheric or system effects on detected radiances, (3) classification or product generation. Classification involves the transformation of the detected multispectral radiances in the image into meaningful descriptors of the surface – usually as thematic classes which can be displayed as a map. The mathematical conversion process is usually based on training an appropriate classifier algorithm by use of example spectral radiances from known surface objects (ground truth data). Classifier algorithms may either be nonparametric or parametric (e.g. based on a statistical model). Neural network algorithms have found useful application as classifiers for satellite image data. Products may also be generated by computing environmentally meaningful indices from spectral radiances such as the normalized difference vegetation index (NDVI).