基于改进UNet++的地震断层识别方法研究

Research on Seismic Fault Identification Methods Based on Improved UNet++

断层解释是油气勘探开发过程中的重要工作,但是随着勘探规模的增大,传统的人工解释断层的方法已经无法满足实际生产的需要。针对人工标注断层特征费时费力、传统断层识别结果连续性不足的局限,以及地震资料中断层与非断层样本分类不均衡的问题,提出基于CBAM-UNet++模型的地震断层识别方法。采用合成地震数据自动生成地震数据和断层标签,提高断层标注的效率。首先,将CBAM注意力模块引入UNet++,从通道和空间两个维度抑制地震振幅信号干扰,增强地震断层的检测能力,采用DropBlock模块抑制网络中产生的过拟合问题;其次,引入Dice Loss损失函数用于减小断层识别任务中数据不均衡问题对模型的影响;再次,对断层预测结果进行霍夫变换,提取骨架,使断层预测结果更好地应用于地质目标;最后,在合成地震数据集、北海地区F3区块真实地震数据上评估CBAM-UNet++模型,与UNet++、UNet、SegNet进行对比。结果表明,基于CBAM-UNet++的断层识别方法在准确率、断层连续性方面表现优异,可自动、有效地识别地震图像中的断层。

Fault interpretation is an important task in the exploration and development process of oil and gas.However,as the exploration scale increases,the traditional manual fault interpretation can no longer satisfy the needs of actual production.In light of the limitations of manual fault annotation,such as being time-consuming&laborious,insufficient continuity of traditional fault identification results,and the unbalanced classification of fault and non-fault samples in seismic data,a seismic fault identification method based on CBAM-UNet++model is proposed,which adopts synthesized seismic data to automatically generate seismic data and fault labels,so as to improve the efficiency of fault annotation.Firstly,we introduce the CBAM attention module into UNet++,which suppresses the signal interference of seismic amplitude from the two dimensions of channel and space to enhance the detectability of seismic fault,and suppresses the over fitting in the network with the use of DropBlock module.Secondly,introduce the Dice Loss function to reduce the impact of data imbalance on the model in fault identification task.Thirdly,perform Hough transform to the fault prediction results for skeleton extracting,so that the fault prediction results can be better applied to geological objectives.Finally,evaluate the CBAM-UNet++model based on the synthesized seismic data set and the real seismic data of F3 block in the North Sea,and compare with UNet++,UNet and SegNet.The results indicate that the fault identification method based on CBAM-UNet++model has excellent performance in terms of accuracy and fault continuity,and can identify faults in seismic images automatically and effectively.