Research and application of fault identification method based on four-channel deep supervised network

Manual fault interpretation is time-consuming and the experience of the interpreters tends to increase the uncertainty of the fault interpretation. With the development of computers and artificial intelligence, varieties of algorithms based on convolutional neural networks are more widely used in fault recognition. In order to further improve the accuracy of fault identification, a multi attribute convolutional neural network is proposed to identify faults. First, the residual block is introduced to replace the convolutional block in 3D UNet to simplify the learning goal and reduce the difficulty of training. Then the dilated convolution module and multi-layer deep supervision mechanism are introduced to take advantage of multi-scale information more effectively and further improve the accuracy of fault recognition. Finally, the four-channel network is constructed with multiple seismic attributes to learn various response characteristics of faults. The synthetic model test and the application of real seismic data show that this method can effectively identify the fault location, reduce the missed and wrong identification of small faults, and significantly improve the accuracy of fault identification.