To solve the problem of parameter diagnosis of microwave excited plasma, the equivalent relative complex permittivity of low-temperature plasma excited by microwave is measured by the two-port transmission reflection method. Firstly, the sample-test structure is designed based on the standard BJ22 waveguide structure with the finite element method. The structure can ensure that the permittivity and port parameters of |S11|,φS11,|S21| keep good monotonicity in a large range of permittivity, avoiding multi-value problems. Then, artificial neural network is utilized to train the S parameters which are corresponding to each permittivity obtained by simulation, and the training results achieve sufficient accuracy. Finally, through experiments, the measured S parameters are inverted by neural network to obtain the relative complex permittivity of the substance to be measured. This method can be applied to measure the permittivity of materials like plasma which has negative real part of complex permittivity.