基于卷积神经网络深度学习的地质断层智能识别方法

Intelligent Recognition Method of Geological Fault Based on Deep Learning of Convolutional Neural Network

为解决传统机器学习方法识别断裂构造能力较差的问题,提出了基于卷积神经网络的UNet++网络结构模型用于识别地质断层。模型的建立过程中引入了不同注意力机制与损失函数,可以更好地实现语义深度学习与特征融合,并进行了相关性指标分析与图像分析。结果表明:WCE损失函数对应的预测图具有最清晰的输出效果,ECA+UNet++模型利用WCE损失函数的训练效果最佳,识别的准确率也更高。将采用WCE损失函数的ECA+UNet++模型在官渡河煤矿断层区域进行应用,可以对断层位置进行智能识别,并且对地下噪音的降噪处理较好;表明采用引入ECA注意力机制的UNet++网络结构模型能保证对断层识别的效率与精度。

In order to solve the problem that traditional machine learning methods have poor ability to identify fault structures,a UNet++network structure model based on convolutional neural network is pro⁃posed to identify geological faults.Different attention mechanisms and loss functions are introduced in the es⁃tablishment of the model,which can better realize semantic deep learning and feature fusion.Correlation in⁃dex analysis and image analysis are also carried out.The results show that the prediction graph correspond⁃ing to WCE loss function has the clearest output effect,and ECA+UN++model has the best training effect using WCE loss function,and the recognition accuracy is higher.ECA+UN++model with WCE loss func⁃tion is applied in the fault area of Guanduhe Coal Mine,which can identify the fault location intelligently and deal with the noise reduction of underground noise well.It shows that the UNet++network structure mod⁃el with ECA attention mechanism can effectively improve the efficiency and accuracy of fault recognition.