Lithology segmentation using deep neural network

Summary This paper avoids the difficulties in using conventional methods in lithology segmentation task by putting the tasks in the frame of computer vision. First, we setup a lithology dataset which contains paired topology, satellite and lithology images; Second, two heated neural networks HyperNet and UNet are introduced and applied in lithology segmentation task. The experiments show that both HyperNet and UNet are efficient and promising for the application in lithology segmentation. % Neural networks can increase the predicted accuracy three times than random guess, that greatly reduce the workload of professional lithology geologist.