基于双注意力U-Net网络的提高地震分辨率方法

Seismic resolution improvement method based on dual-attention U-Net network

提高地震数据分辨率的传统方法,如反褶积、Q补偿等,受到子波为最小相位、反射系数为白噪声等假定条件的限制且需要求取复杂参数,不便于实际应用。深度学习方法使用数据驱动的方式可以自适应地刻画输入与目标间的关系,具备良好的自主学习能力,但目前基于深度学习提高地震数据分辨率的方法对注意力信息的利用不够全面。因此,提出一种基于双注意力U-Net网络的提高地震数据分辨率方法。首先,在原始U-Net网络中加入改进的通道注意力模块、空间注意力模块和级联残差模块,不仅可以快速学习高、低分辨率数据间的映射关系,还能够合理分配不同通道和空间的权重、充分利用数据间的相关性;然后,使用L1损失和多尺度结构相似性指数损失的组合作为损失函数,提高模型对局部信息变化的敏感度,便于恢复细节信息。模拟数据和实际数据的测试结果表明,该方法提升了地震数据的主频,增加了频带宽度,同相轴变得更清晰,细节纹理信息更丰富,有效提高了地震数据的分辨率。

Traditional methods to improve the resolution of seismic data,such as deconvolution and Qcompensation,are limited by the assumption that the wavelet is the minimum phase,and the reflection coefficient is white noise,and they need to calculate complex parameters,which brings inconvenience to practical application.The deep learning method uses the data-driven method to adaptively depict the relationship between input and target and has excellent self-learning ability.However,the current method of improving seismic data resolution based on deep learning does not fully utilize attention information.Therefore,a method of improving seismic data resolution based on dual attention U-Net network is proposed.First of all,the improved channel attention module,spatial attention module,and cascade residual module are added to the original U-Net network,which can not only quickly learn the mapping relationship between high-and low-resolution data but also reasonably allocate the weight of different channels and spaces and make full use of the correlation between data;then,the combination of L1loss and loss of multi-scale-structural similarity index measurement is used as the loss function to improve the sensitivity of the model to local information changes and facilitate the recovery of detail information.The test results of simulated data and actual data show that this method improves the main frequency and frequency band width of seismic data,makes the event axis clearer,enriches the detailed texture information,and effectively improves the resolution of seismic data.