Recognition of Stratum Lithology of Seismic Facies Based on Deep Belief Network

The Deep Belief Network (DBN) is one of the major algorithms of deep learning. It simulates human brain to extract the features efficiently, so that the model has much strong learning ability. Because it is difficult to extract features from a variety of seismic data effectively, multiple sampling points of seismic data are used as inputs. Then we use DBN to extract the features from seismic data, which can be stacked by RBMs layer-by- layer. The model of lithological recognition can be constructed from previous step, further to recognize stratum lithology. By experiments and practical application, it is proved that partial strata information can be utilized effectively when multiple sampling points of seismic data are used as inputs. In this way, we can effectively recognize the stratum lithology based on DBN.