CNN Based Fault Recognition with Multi-Scale Fusion Attention Mechanism. Issue 4 (April 2020)

• Record Type: Journal Article
• Authors: Wang, Bo; Ma, Jinwen
• Is Part Of: IOP conference series. Volume 474:Issue 4(2020)
• Publication Date: 2020-04
• Subjects: Earth sciences -- Periodicals; Environmental sciences -- Congresses; Environmental sciences -- Periodicals; 550.5
• DOI: 10.1088/1755-1315/474/4/042007
• Languages: English
• ISSNs: 1755-1307

Abstract

Abstract: In geological exploration, seismic data interpretation is very important and its key task is fault detection or recognition. In fact, the accurate recognition of fault shape and distribution plays an important role in oil and gas exploration. With the increasing demand of fault interpretation accuracy in oilfield, traditional manual fault interpretation methods are susceptible to subjective factors of interpreters. With the rapid development of deep learning technique, convolutional neural networks (CNNs) has successfully applied to the complicated tasks of image classification. This paper utilizes proposes a CNN based fault recognition framework with a multi-scale fusion attention mechanism. It is demonstrated by the experimental results that our proposed approach is superior to the traditional methods on both the fault interpretation accuracy and efficiency of 3D seismic images, greatly reducing the risk of oil and gas exploration and development. With a GeForce GTX 1080 GPU, the training processing takes about 4 hours and predicts the faults in a 33×33×33 seismic volume using only tens of milliseconds.