Signal and image classification in georadargrams

The given tasks in our joint project are the detection and location of hyperbolas generated by pipes and cables in the subsurface when illuminated with a common pulse radar, and to find classification indications of the object's nature. Classification not only includes informations about the material and cross-section of an object, but also its three-dimensional extension in the subsurface. Thus, the processing of radar data is divided into several successive steps. First, the hyperbola has to be detected in the radargram and the approximate location of the pixels belonging to the hyperbola has to be detected. We perform this by means of the Hough transform algorithm. The result of the Hough transform and its degree of automation depend greatly on the preprocessing of the data transformed. To apply the correct preprocessing and processing strategy, classification of the radargrams by texture arialysis is investigated. After the detection of the hyperbola, an analysis of the reflected wavelets can provide some clues about the object's nature. Although absolute classification, which can be derived straightforwardly from simulations, will be successful only in a few exceptional cases, signal analysis is very useful in three-dimensional tracking of objects. We were able to demonstrate this by using different statistical correlation methods to identify whether hyberbolas in parallel transects were generated from the target or not. A simple but often successful method of detecting and following long objects is the presentation of parallel transects as a time slice.