Non-Magnetic Materials Assignment based on Artificial Neural Network

In this paper, signal recognition system is proposed for classification of soil types and buried object by using feature extraction and artificial neural networks. Using GprMax simulation program, 300 GPR A scan signals were obtained for different ground planes and materials with 206 Hz sampling frequency. Since the materials of buried objects are non-magnetic, these signals contain electric fields in the z direction (Ez) and magnetic fields in the x and y directions (Hx, HY). Features are extracted from them in time and frequency domain such as mean, standard deviation, skewness, kurtosis etc. Therefore, signal features are obtained for inputs of artificial neural networks. Feed forward backpropagation neural network is chosen as an artificial neural network model. The outputs of neural network are types of soil and materials. Also, classification accuracy is examined with changes in learning rates, iteration numbers, momentum constant and number of neurons in hidden layer. The obtained results show high accuracy rate for soil and material recognition.