Automatic detection of three-dimensional geological features from remotely sensed imagery and digital terrain models

Traditionally, the role of remote sensing in geology has been to provide the geologist with visually enhanced imagery offering synoptic coverage of an area for manual interpretation. More recently, automated techniques have been introduced in an attempt to remove the subjectivity of human interpretation. The need for these techniques will grow as a result of the increase in data volumes and the expectations afforded by sensors on board the Polar Platforms, due to be launched in the 1990s. This paper describes a number of automated techniques designed to extract geological information from remotely-sensed imagery and digital terrain models (DTMs). In particular, an edge detection algorithm is applied to the imagery and the DTM to identify litho-logical boundaries and faults. Various rules and conditions are identified which enable separation of ‘geological’ edges from other types of edge (e.g. man-made features). The edge images are then thresholded and thinned to produced line entities. The resulting two-dimensional data are combined with the elevation data to determine local dip and strike measurements. A least squares approach is employed to ‘fit’ these three-dimensional data to a planar surface. The results are compared with measurements of dip and strike obtained in the field and derived from a 1:50,000 scale geology map of the area. In many parts of the study area the results are accurate to within ±10°. Other areas require improved/additional image processing techniques, knowledge-based rules and a structural model if accurate measurements are to be extracted successfully.