A New GPR Image Analysis Proposal Based on a Multi-agent Approach and Properties of Groups. Towards Automatic Interpretations

This paper proposes a new methodology to analyze images from subsoil surveys obtained by using ground penetrating radar (GPR) as a non-destructive method. The aim is to advance towards automatic GPR image interpretations. The main idea is to promote the inclusion of these nondestructive technologies as support in technical management in large-scale systems such as water distribution systems (WDSs). The methodology proposed in this paper uses pre-processing techniques based on a multi-agent approach and analysis of the properties of the obtained groups. In addition, a first classification (via perceptron) of the selected property in the groups is performed. The work is based on GPR studies performed on controlled laboratory conditions; two kinds of test conditions are raised. The first kind pursues capturing objects in GPR images to favor feature extraction thereof, since these tests are used to develop the proposed methodology. The second kind of tests seeks to assess the feasibility of the proposed methodology implementation, while the interaction of the captured objects with other objects and with the environment is evaluated. The results of this study are promising in regard to the selection of areas in GPR images via automatic interpretation that allows the extraction of relevant features for WDS networks. In addition, with the proposed methodology, the use of GPR is promoted by non-highly skilful and expert operators in GPR image interpretation.