Research On Airborne Electromagnetic Inversion Method Based On Deep Learning

Airborne electromagnetic(AEM)inversion is a process that utilizes electromagnetic data obtained through airborne surveying to infer underground geological structures,mineral deposits,groundwater resources,and other important geological features.This paper proposes a deep learning-based AEM inversion method that addresses traditional inversion issues such as reliance on prior information,dependence on initial models,and long inversion times.Two neural networks,a convolutional neural network(CNN)and Res Net INV network based on the original residual network(Res Net),are used to compare the impact of different network structures on AEM inversion results.Firstly,this paper designs an appropriate observation system for AEM and constructs different shaped,positioned,and resistivity-valued electrical models.The finite volume method is used for forward simulation to obtain induction electromotive force(EMF)data,which constitutes the data set and undergoes preprocessing.Then,the CNN and the Res Net INV network are trained using the training set in the above data set.After the training is completed,the validation set is used to evaluate the network’s learning ability,and the network parameters are continuously adjusted until the expected effect is achieved.Finally,the test set data and noise data are input into the CNN and the Res Net INV network to invert their corresponding electrical models.The inversion results are analyzed to evaluate the network’s generalization ability and noise resistance.Experimental results show that both the CNN network and the Res Net INV network can complete airborne electromagnetics inversion.However,the Res Net INV network with a more complex structure and deeper depth is superior to the CNN network in terms of inversion accuracy,detection depth,and longitudinal resolution.Overall,the Res Net INV network has strong learning ability,stability,and certain generalization ability and noise resistance,which can effectively solve the high-dimensional inversion problem of airborne electromagnetics.