Vol. 11, issue 01, article # 15

Protasov K. T. Isolation of cloud fields in satellite pictures using the segmentation algorithm based on classification and pattern recognition. // Atmospheric and oceanic optics. 1998. V. 11. No. 01. P. 74-79.    PDF
Copy the reference to clipboard

Abstract:

We have developed a combined algorithm involving four-step procedure for making segmentation of multispectral pictures of clouds and underlying surface taken from a satellite. The procedure steps include (1) fragment-by-fragment, clusterization of video data into classes; (2) then the closest classes identified are united using Bhattacharia spacing; (3) thus formed ensembles of classes are used to teach the pattern recognition algorithm used; (4) and finally, this algorithm is used to make fragmentation of the entire image. This approach enables one to compromise between the necessity of using certain models of the patterns to be recognized, for instance, those based on the Johnson approximations we use here, and bulky and cumbersome arrays of the initial data. The algorithm performance is illustrated by an example of discriminating cloud fields from the patterns recorded with an Advanced Very High Resolution Radiometer (AVHRR) onboard NOAA satellite.