Lithology Recognition Of Drilling Cuttings Based On Machine Learning

Lithology identification is the key link of oil and gas exploration and the important content of geological investigation.At present,the most direct and reliable lithology identification techniques are thin section identification and imaging logging,but both of them can not be popularized because of long cycle and high cost.In order to realize efficient lithology identification,a lithology identification method based on improved VGG16 and Swin Transformer is proposed in this paper.The details are as follows:Firstly,a lithology identification method based on improved VGG16 is proposed.In the pre-processing of rock image data,the data are expanded by image segmentation,normalization and data enhancement.Then,VGG16 is used as the backbone of the network to build the lithology recognition model.The problem of over-fitting and training is solved by introducing Dropout,Batch Normalization and residual modules.Because convolutional neural network are good at extracting local features,it is difficult to get global information.By adding SE attention module in the network,it can capture long-distance features of features.The experimental results show that the improved VGG16 network model has an accuracy of 90.32%,which is 3.65% higher than the original VGG16 network model.Secondly,applying Transformer network to lithology identification field,the lithology identification model based on Swin Transformer is established.In order to improve the convergence rate of the recognition model training,the Transformer network is trained on the Image Net data set to obtain the pre-training initial weights,then the weight is embedded into the model to train the experimental rock data set.The experimental results show that the prediction accuracy of the model reaches 88.27%,and the model can also be accurately predicted by randomly drawing pictures to verify the model.Therefore,Swin Transformer network also has a good prediction effect in the field of lithology identification,which realizes the research goal of introducing Transformer network into the field of lithology identification.