With consideration of advantages of airborne magnetotelluric method in flexibility and high efficiency, it is suitable for rapid reconnaissance in remote environments with rugged terrain. At present, the method of solving partial derivatives of the objective function was adopted in most inversion studies of two-dimensional airborne magnetotelluric data, which made the inversion results highly dependent on the initial model and easily fell into the local minimum value. In order to solve the above problems, firstly, this paper deduced the formula for calculating the tipper response from Maxwell's frequency domain equations. Secondly, the tipper response of geoelectric model of the anomalous body which were calculated by finite element method, combined with the label of the underground resistivity value and the deep learning sample data set was constructed. Finally, the method of traditional electromagnetic and deep learning inversion were used, to carry out the inversion research on the geoelectric models of two-dimensional abnormal bodies with different buried depths. The inversion results of different inversion methods were compared and analyzed. The final results show that the deep learning inversion method is faster and more accurate for two-dimensional airborne electromagnetic data inversion, compared with the traditional electromagnetic inversion method.