Taking the secondary field response of layered earth excited by magnetic dipole field source as the basic theory, an improved damping least squares method is adopted for frequency domain electromagnetic data inversion to solve the parameters such as thickness and conductivity of the media in each layer of underground space. Due to the significant difference in numerical values between conductivity parameters and thickness parameters, in the inversion calculation process, the order of magnitude of each element in the Jacobian matrix is adjusted to quickly converge the parameter vector to near the true value. This article first validates the proposed improved algorithm using an ideal three-layer geological model, and the results show that the traditional damping least squares method still cannot converge to the correct result after 100 iterations, while the improved algorithm can match the model parameters well after 20 iterations. Then a complex underground tunnel model was designed and the inversion results of model parameters were compared at signal-to-noise ratios of 15 and 20 dB. Research has found that when the signal-to-noise ratio is high, the inverted image is smoother. The inversion accuracy is high, with a maximum relative error of less than 6% in the inversion results of conductivity parameters, and a relative error of about 1.4% in the depth of the model body.