Understanding the Effect of Different Prior Knowledge on CNN Fault Interpreter

As deep learning (DL) models have been successfully applied to various image processing tasks, DL models, particularly convolutional neural networks (CNN), introduced into the geosciences assist geologists in faster seismic interpretation. However, generalization of DL-based fault interpretation is a challenge. When data with different characteristics, their performance degrades significantly. Several recent studies proposed transfer techniques, which similar but source tasks are assumed benefit target task. it unclear datasets would be most beneficial for this particular task (i.e. interpretation). In paper, we first demonstrate through systematic literature review that synthetic popular area. Further, previous not compared them other types datasets. Then, experimentally choice dataset should influenced by amount annotation available dataset. addition, normalization appears an essential factor fine-tuning when interpreting faults. Finally, state-of-the-art was achieved on ThebeFault (0.903 AP, 0.849 OIS and 0.845 ODS). Our code publicly at: https://github.com/anyuzoey/pretrain.