A Deep Learning Estimation of the Earth Resistivity Model for the Airborne Transient Electromagnetic Observation

Because of the relatively weak useful signal and noises with complex characteristics, the data processing of airborne electromagnetic observation is very difficult. As the downstream of data processing, inversion generally cannot further distinguish whether there are residual noises in the processed data. That is, in the current processing flow, the signal-noise distinguishing is isolated from the signal-model mapping. The reliability of the estimated earth resistivity model will be seriously affected due to the probable disadvantages in the denoising process. To this end, we propose a manifold assumption, and so establish one "feedback" mechanism between signal-noise distinguishing and signal-model mapping. On this basis, we propose a deep learning method: through simultaneous optimal training the network parts for signal-noise distinguishing and earth resistivity model estimation respectively, the entire neural network can perform the denoising and inversion in the traditional sense at the same time, so as to obtain more objective and reliable estimation results of earth resistivity model. We use the Stacked Auto-encoder neural network structure to implement the proposed method, and test the network performance with simulation and measured data. The results show that the proposed method can obtain a more reliable earth resistivity model directly from the noisy data.