Analysis technology for land remote sensing

In this paper an overview is provided of the development to the present time of the ability to analyze vertical views of the earth's surface by machine. The paper begins with a brief review of the several phases of effort which have led to today's operational capabilities, starting with the research phase of the 1960's which defined the basic fundamentals, and continuing with the developmental phase of the 1970's in which large numbers of scientists and professionals became acquainted with the new technology and many of the needed application details were worked out. Three application examples are next discussed in some detail in order to illustrate the user environment. In the second half of the paper current and future analysis algorithms are discussed. As a result of data characteristics which may be exploited, this is divided into algorithms utilizing spectral, spatial, and temporal characteristics and the use of ancillary data. The most common means for utilizing multispectral observations in pattern classification algorithms, multispectral pixel classifiers, are discussed first. There follows a review of the use of quantitative texture estimation schemes, a scheme to partition the scene into statistically homogeneous objects and means for the direct use of spatial context. Multitemporal schemes using concatenated multispectral vectors and temporal partitioning or cascading are described. Ancillary data has traditionally been incorporated by interactive means, and this is discussed briefly, as is the use of relexation methods to utilize one or more ancillary parameters in quantitative form. The paper is concluded with a discussion of other trends in the field, such as those related to data base technology and new sensor systems.