Seismic facies recognition based on prestack data using two-dimensional convolutional auto-encoder and cluster analysis

In the exploration and development of underground sedimentary minerals such as oil, natural gas and coal, prestack seismic data can help us find more complex atypical reservoirs, so we are increasingly inclined to use prestack seismic data for seismic facies recognition. However, prestack data contains excessive redundancy, so it is critical to effectively extract features from prestack seismic data. In this paper, the seismic facies recognition of prestack seismic data is considered as an image clustering problem in computer vision (CV). Each prestack data gather is treated as a picture, using a 2D convolutional auto-encoder network (2DCAE) to extract the features of the pre-stack data and eliminate the redundancy. The optimal loss function and the optimal regularization term are also explored in this model. Then unsupervised pattern recognition methods are used to cluster the extracted features to realize seismic facies recognition. Experiments show that the huber function is used as the loss function, and the addition of the huber regularization term works best.