Hidden Object Recognition using Convolutional Neural Network

In this paper, the detection, and localization of a hidden object in the human body using deep neural networks have been studied. To build a model, an electromagnetic simulator is employed. The model consists of four layers (skin-fat-muscle-bone) each of these layers has different conductivity and relative permittivity. Spherical shrapnel of different sizes 5mm, 10mm, and 15mm is supposed to be at various places in the model. The signal is directed at the model using a monopole ultra-wideband antenna, which is also used to pick up signals that are reflected back. In order to determine whether shrapnel is present or not, its size, and where it is located, the collected signals are analyzed using a deep neural network. The acquired results utilizing the suggested method are encouraging, with 90% success in shrapnel identification, 88% success in shrapnel sizing, and 78% success in shrapnel depth. More antennae could be used to improve performance.