Landsat image analysis : application to hydrographic mapping

Abstract

Problems of applying Landsat imagery to hydrographic surveying are analyzed and an approach which solves these in a practical way is proposed. An experimental analysis system has been developed to facilitate this study and has also been used in several other remote sensing projects. In addition to its utility as a research tool, experience with the system - both in its development and use - has given a number of insights into requirements to be met by the design of similar interactive analysis systems. The central problem, hydrographic surveying, is introduced by an analysis of the physical principles involved followed by two preliminary investigations - one based on the NASA/Cousteau experiment, the other on results derived from the Torres Strait scene. Two major requirements for the successful utilization of Landsat in this application are identified - thorough radiometric correction, and an appropriate digital image analysis system. Radiometric defects in Landsat data are aggravated by the limited range of useful signal for marine applications, and many correction techniques which are successful for land-oriented studies fail in this context where radiometric precision is critical. An alternative approach is developed to adequately handle marine data. Several phases of the mapping process involve close interaction of a human image interpreter with a computer analysis system. Most importantly, the interpreter draws on a number of special purpose numeric and graphic aids, as well as his own inspection of the image and background expertise, to segment the image into regions for which consistent optical models can be determined. These concepts are tested in a series of surveys which reflect practical survey conditions. The possibility of using data from multiple spectral bands and multiple overpasses of the satellite is explored. Simple comparison of depth maps derived independently from two images can detect a large proportion of likely anomalies but, to allow anomaly detection earlier in the processing, a modified cluster analysis technique has been devized. In its present state the cluster technique must be used interactively but there is potential for further automation. Finally, a number of implementational considerations and possible future developments are discussed. It is concluded that semi-production operation is now required to allow further evaluation of the technique and to develop the necessary expertise. Show more