Machine learning for analysis of GPR images and electromagnetic diagnostics

The aim of this work is to exploit Machine Learning (ML) for the analysis of Ground Penetrating Radar images. In particular, the objective is to apply a scaled-down version of DenseNet [1] architecture with a multiporse approach to extract from b-scan images of buried cylinders: the cylinder radius, the cylinder length, the depth with respect to the ground, and the relative permittivity of the cylinder and of the medium in which the cylinder is immersed. The cylinders have an infinite length or have a length much greater than the diameter. The main feature of the network chosen in order to extract those features is that each layer is connected to all subsequent layers, through the concatenation of the feature maps. Indeed, traditional convolutional networks, composed of Llayers, present L connections, one for each layer, while DenseNet presents L(L+1)/2 direct connections. The DenseNet network has many advantages: it reduces the problem of the evanescent gradient, strengthens the propagation of features, encourages the reuse of parameters and substantially reduces the number of parameters. The Georadar (or Ground Penetrating Radar, GPR) images are obtained through the GprMax[2] software simulation tool, combining the relative dielectric constant of the medium and of the cylinder, radius, the length and depth of the cylinder.