Recognition of patterns from geological structures in radar signals with the neuronal network simulator JNNS

Aim of GPR measurements is not only to detect the spatial position of reflecting structures, but also to classify its geological reason. The reflection coefficient of a radar signal depends on the reflecting material. Numerous propeties of the recorded signal can be used to assign a reflection with geology. Necessary is the knowledge about unique parameters for a classification and the validation of the results in a known environment. Data from well explored salt mines allows this validation. Neuronal networks are a good tool to optimize parameter sets or patterns regarding to its relevance for an assignment. A neuronal network consists of simple units which transmit information weighted and directed to the next level. The connections between the units are characterized by activation functions which vary during learning processes. By validating a parameter set it is possible to connect an input level with an output level and creating an associative pattern without knowing the deterministic rules between them. Which parameters of a radar signal are relevant to get a result with this method is the topic of this paper. The neuronal network simulator JNNS used here was developed at the university Stuttgart.