Due to the huge amount of data generated by time-domain airborne electromagnetic (AEM) systems, conductivity depth imaging methods are widely used to help in the interpretation of these data because they can be generated quickly and easily. We have introduced a new imaging method generated using a deep neural network. The network structure combines four convolutional neural networks with a long short-term memory technique and adopts an error back-propagation scheme to update the parameters. A deep hierarchical structure is used to extract and store the complex nonlinear relationship between the model and electromagnetic (EM) responses, creating results close to 1D inversions. To check the effectiveness, we have examined our algorithm on synthetic and survey AEM data. The imaging result shows that this method, when using reasonable network parameters, can not only image well at high speed, but the method also is not very sensitive to noise. Another advantage of this method is that once the training is completed on a well-distributed training set, the network can be used without any changes to process other data sets from an identically configured AEM system.

We investigate an algorithm for 3D time-domain airborne electromagnetic (AEM) inversion based on the finite-volume (FV) method and direct Gauss-Newton optimization, where we obtain high efficiency by constraining the modeling volume to the AEM volume of influence (VOI) of a 3D source within the earth, rather than using the larger VOI of the AEM system. A half-space or layered earth is used to model the background field in the time domain, taking into account the transmitter waveform through convolution. Assuming that the 3D source of any secondary field detected at a survey point lies within the moving VOI of the airborne system, we conduct time-domain forward modeling and Jacobian calculation using an FV method within the 3D source VOI that requires a small number of cells for discretization. A local mesh and direct solver are shown to further speed up the computation. A synthetic isolated synclinal conductor inversion shows good agreement with the model geometry and provides a good fit to the data contaminated with noise. A synthetic multiple-body model inversion was also quite successful, showing that our algorithm is effective and about four times faster than inversion using the total-field method. Finally, we inverted GEOTEM data over the Lisheen deposit, where our inversion result was consistent with the published geology.

We have developed a goal-oriented adaptive unstructured finite-element method based on the scattered field for 3D frequency-domain airborne electromagnetic (AEM) modeling. To guarantee the EM field divergence free within each element and the continuity conditions at electrical material interfaces, we have used the edge-based shape functions to approximate the electrical field. The posterior error for finite-element adaptive meshing procedure is estimated from the continuity of the normal component of the current density, whereas the influence functions are estimated by solving a dual forward problem. Because the imaginary part of the scattered current is discontinuous and the real part is continuous, we use the latter to estimate the posterior error. For the multisources and multifrequencies problem in AEM, we calculate the weighted posterior error for each element by considering only those transmitter-receiver pairs that do not adhere to our convergence criteria. Finally, we add a minimum volume constraint to improve the stability of the adaptive procedure. To check the accuracy, we compared our adaptive results with the semianalytical solutions for AEM systems over a half-space model. To test the effectiveness of our algorithm for multiple sources and multiple frequencies of AEM, we analyzed meshes for separate frequencies and for combined frequencies. Finally, we calculated the AEM responses over a hill model with and without embedded abnormal bodies to prove the feasibility of our algorithm for AEM variable topography modeling.

In this paper, we propose a new wavelet-based 3D inversion method for frequency-domain airborne electromagnetic (FDAEM) data. Instead of inverting the model in the space domain using a smoothing constraint, this new method recovers the model in the wavelet domain based on a sparsity constraint. In the wavelet domain, the model is represented by two types of coefficients, which contain both large-scale and fine-scale information of the model, meaning the wavelet domain inversion has inherent multiresolution. In order to accomplish a sparsity constraint, we minimize an L1-norm measure in the wavelet domain that mostly gives a sparse solution. The final inversion system is solved by an iteratively reweighted least squares method. We investigate different orders of Daubechies wavelets to accomplish our inversion algorithm, and test them on synthetic frequency-domain AEM dataset. The results show that higher order wavelets having larger vanishing moments and regularity can deliver a more stable inversion process and give better local resolution, while the lower order wavelets are simpler and less smooth, and thus capable of recovering sharp discontinuities if the model is simple. At last, we test this new inversion algorithm on a frequency-domain helicopter EM (HEM) field dataset acquired in Byneset, Norway. Wavelet based 3D inversion of HEM data is compared to L₂-norm based 3D inversion's result to further investigate the features of the new method.