Deep Convolutional Auto-encoder Based Lithologic Classification and Recognition

‍ ‍Lithologic classification and recognition is the most basic problem for geological exploration and seismic signal processing. However， it is rather complicated with a variety of factors so that statistical and conventional machine learning models cannot achieve a satisfactory accuracy of lithologic classification， and therefore the obtained classifiers cannot be effectively applied to the lithologic recognition in practical applications. In order to overcome this difficulty， this paper proposes a deep convolutional auto-encoder neural network as well as its parameter learning algorithm for lithologic classification. By adopting the run length smoothing algorithm （RLSA） to remove the isolated points in the classification result of the network on the input data series， our proposed deep learning model can make the lithologic classification more effectively. It is demonstrated by the experiments that the accuracy of lithologic classification of our proposed model is remarkably increased to a practically applicable level， even if there is only a small number of tag lithologic samples in well logging.