Automatic classification of sidescan sonar images for mapping marine mineral resources

Abstract

The Brazilian continental shelf is one of the widest in the world, but its sea floor is still very poor known in terms of shallow mineral resources (siliceous, carbonates and heavy minerals sediments). The main source of published information on that comes from the Projeto REMAC (1974–1979) and the Programa REVIZEE (1995–2000), where a bare distribution of bottom sediment samples information were plotted on large scale maps. Despite of geophysical acoustics techniques have being widely used worldwide since fifty years ago, they are not often used in Brazil for a regular mapping of shallow mineral resources on the continental shelf. The exception is some local areas of interest for the oil industry. In this context, one of the easiest and cheapest ways for mapping the sea floor is the use of sidescan sonar imagery. Nowadays there are a number of Brazilian institutions and companies owners of these equipments and a few software for processing and interpretation of its data are available. However, as any geophysical data, the results depend a lot on direct geological information besides the professional expertise. In particular the interpretation of sidescan sonar images is very time consuming due to a labor-intensive and qualitative way of delimiting the sediment textural boundaries or targets, which generates many uncertainties. Another source of ambiguity comes from the interplay among the geotechnical properties of sediments/rocks, its spatial variations in 3D and the non-linear effects of the sound backscattering itself. One way for avoiding these subjectivities is the use of automated classification algorithms based on spectral analysis of the image properties. There is some commercial software in the market that produces reasonable results after automatic imaging classification. However, most of them are efficient only when the target object is well delimited — as for mapping cells in molecular biology or in remote sensing soil classification — but they produce odd results when the image properties change slightly. This is the case of acoustic images of sea floor resources acquired by sidescan sonar, because its backscattering changes in a complex way due to the spatial anisotropy; although the same target at different depths will give a different response due to sound transmission losses. In this paper we discuss the use of automated classification algorithms and its efficiency on mapping marine mineral resources based on sidescan sonar images. For this we have selected a strategic area on the Brazilian continental shelf where geological information is available for validating the results. The conclusion is that after the appropriate image processing (gain equalization, class definition, spatial and spectral subsets and filtering) we can automatically generate thematic textural maps of sea floor mineral resources, despite of its spatial and mineralogical complexities. These techniques might contribute for fast and standard mapping of marine mineral resources on the Brazilian continental shelf.