A Fault Recognition Method Based on Convolutional Neural Network

Fault recognition is an important part of seismic interpretation, but the existing methods’ accuracy is not high enough. Convolutional neural networks have achieved high accuracy in handwriting recognition. Based on the similarity between fault and handwriting shape features, this paper proposes a fault recognition method based on the classical convolutional neural network. Firstly, a neural network model suitable for fault recognition is designed based on the successful LeNet5 model, which has been used for handwriting recognition. The output layer of the neural network is designed with two neurons to judge whether the seismic sample points belong to faults. Additionally, the Softmax Regression model is used to replace the original European radial basis function. This modification allows the network to output not only whether the seismic sample points belong to faults, but also the probability of belonging to or not belonging to a fault. Then, a 3D sample set and test set are established, using the most accurate manual fault recognition method, to train the neural network. Finally, the proposed method is tested on actual seismic data, and the experimental results confirm the effectiveness and progressiveness of the proposed approach. © (2024). All Rights Reserved.