Research On The Application Of Deep Belief Network In Image Processing Of Rock Slices

The upper palaeozoic gas reservoir of PCOC(China petroleum Changqing Oilfield Company)is one of the main exploration area in the Erdos basin,the study of reservoir pore structure characteristics in Erdos basin will contribute to predict the geological reserves and to point out the development direction of oil and gas.Analysis on microscopic images for the rock thin section is one of the most common used methods for investigations on reservoir microstructures,which plays an important role in the petroleum and natural gas exploration and development.Rock image classification is the research basis of rock characteristics,it occupies a pivotal position in geological analysis research.The goal of this thesis is improve the accuracy of classification of massive rock data with high resolution.Deep belief network(DBN)in deep learning can extract features from lower level to higher level,which combines the advantages of unsupervised and supervised learning,and has better classification performance for high-dimensional data.In this thesis,deep learning is applied to the field of complex microscopic rock image,and a rock image classification model based on deep belief network is established.In the early stage of the study,the image acquisition and image mosaic of the cast sheet in Ordos Basin were completed,to prepare for the subsequent image classification.This thesis selects the appropriate feature extraction method,and applies it to the classification of rock slice images.The experimental results show that the classification accuracy is 98.75%,which is better than other traditional classifiers.Researches in this thesis provide an appropriate deep learning method to classification high resolution rock microscopic images.Related results have high reference significance to the acquisition of wide view and high resolution rock images.Exploration works on petroleum and gas exploration and development may be promoted.