A Complex Permittivity Inversion Method Based on Free-space Method and BP Neural Network

Abstract

This paper presents a complex permittivity inversion method based on free-space method and BP neural network. The free-space method can operate the non-contact measurement of large area composite materials, and there are no strict process requirements for the measured materials. Based on the chamber measurement platform and the principle of free-space method, a complex permittivity measurement system is established. The complex permittivity of air and soil at frequencies from 22 GHz to 26 GHz were measured by this measurement system. Then, same samples are used for measurement in the coaxial waveguide system. Results show that relative error of the real part of complex permittivity measured by the two methods is less than 6.5%, which verifies the accuracy of the data measured in chamber. Next, the soil measurement data from 22 GHz to 26 GHz are used as the training data, and three BP neural networks are trained respectively. The difference is that one neural network outputs the real and imaginary parts of the complex permittivity at the same time, and the other two neural networks outputs the real and imaginary parts respectively. Finally, the three trained BP neural networks are used to invert the complex permittivity of soil at frequencies from 18 GHz to 21 GHz. The inversion results are compared with the measurement results of waveguide method again. After comparison, it is found that the relative error obtained by using two BP neural networks for inversion is smaller. The average error of complex permittivity is 6.64%, which verifies the reliability of the inversion method.