Abstract: Azimuth electromagnetic wave resistivity logging can provide abundant subsurface information and help to determine reservoir location and boundary detection. However, the common iterative inversion method based on physical equation has low computational efficiency and is limited in real-time geosteering. Therefore, an intelligent retrieval method of azimuth electromagnetic logging data based on depth residual network (ResNet) is proposed. The method replaces the convolution and pooling layers in the residual block with fully connected layers, and uses a Multi-head Attention mechanism to understand the relevance of the input data to solve the nonlinear regression problem. By evaluating the depth and width of the model and using Bayesian optimization tuning algorithm to find the optimal hyperparameters of the electromagnetic wave resistivity inversion method, the performance of the inversion model is improved. The method shows good accuracy in model experiments, with an average accuracy of 98.5%. In the actual logging data, the average accuracy rate is 97.2%, and the single point inversion time is about 0.01 s, which can quickly and accurately invert the electromagnetic wave resistivity logging data.