Thin section identification is the basis of various geological work such as research on sedimentation,diagenesis,and reservoir of carbonate rocks.Carbonate rocks have strong heterogeneity,various structural components and particle types.The artificial thin section identification is subjective,difficult,time-consuming and labor-intensive,and not easy to be widely popularized.In the big data and artificial intelligence(AI)background,it is promising to increase the efficiency by applying AI identification technology.This study summarized the research status and analyzed the existed problems in AI identification of carbonate thin sections.The main contents of AI identification of carbonate thin sections include:(1)Preparation of thin sections and image processing.Dyeing thin sections with no-cover glass are the basis of later recognition.The blue casting thin sections are significant for pore recognition.Photos should be captured under different optical property including PPL and XPL with different rotation degree.Image pre-processing and segmentation can help to increase the later identification.The establishment of carbonate thin section database is the basis of AI identification.(2)Based on the prior knowledge of carbonate professionals,the structural components,mineral components and pore types of the image are classified,label classification is established,and manual labeling is carried out by carbonate professionals.It is established that the classification chart of major component labels in carbonate thin sections.The establishment of label database can contribute to further machine learning.(3)The convolution neural network and deep learning are introduced into the labeled thin section images,which can learn and discriminate the morphology and internal structure of various components.The knowledge graph of the thin section image labels is established by combination of machine learning and manual correction,which can classify rock types,recognize sedimentary structures and grain types.(4)It is performed that intelligence recognition of structural components,mineral components and pore types and contents.The denomination specification for AI identification of carbonate thin sections is established.Automatically denomination would be achieved.There are still problems including label sample amount,indeterminate semantic object segmentation,diagenesis,etc,which need further research.The future development directions of AI carbonate identification include the identification of core-outcrop-microscopic image,geochemical image(CT,SEM,FL,etc.),interpretation of logging and geophysical data.

最新研究表明四川盆地川中北斜坡地区走滑断裂发育,具有断面陡直、断距小、断裂不明显的特征,由于目的层下古生界埋藏较深,地震资料信噪比低,难以对走滑断裂进行快速精确地识别。为此,采用了优化后的AI断裂识别技术：①对地震数据进行背景建模,求取局部背景能量数据与原始地震数据残差,增强地层背景反射之下微小断裂地震信息的显现;②进行构造导向滤波,提高地震数据的信噪比,使地震数据同相轴的连续性和断裂特征更加明显;③进行AI断裂识别。应用结果表明：不仅大大节省了人工解释的时间,还具有较强抗噪性,能较好地压制地震数据的冗余信息,提高信噪比,识别出常规地震属性难以识别的微小断裂,更好地预测走滑断裂的展布特征及交割关系。

The latest studies suggest strike-slip faults developed in northern slope of central Sichuan Basin. With steep plane and small throw, these faults are not apparent. It is hard to rapidly and accurately identify them running through targets of the deep-seated Lower Paleozoic due to low signal-to-noise ratio (SNR) of seismic data. So, an optimized AI technique has been used for fault identification. Firstly, background simulation is performed to obtain the residual error between local background energy data and original seismic data, thereby enhancing the signal manifest of minor faults beneath stratigraphic reflection. Secondly, the SNR is improved by means of structure-oriented filtering which makes both continuity of seismic event and fault features more pronounced. Finally, AI technique is used for this identification. Results demonstrate that this technique is not only greatly time saving in artificial interpretation, but exhibits strong noise resistance for redundant suppression with effect, improves the SNR, identifies the minor faults that are difficult to be detected using conventional attributes, and makes the prediction on strike-slip faults’ distribution and intersection even better.

为了提高断层识别的准确率, 提出改进Unet模型。为编码器部分设计一种多分支的并联结构M-block(Multi-branch block), 它可以捕获多尺度上下文信息, 并且多分支的并联结构会带来高性能收益。在解码器部分加入Self-Attention块和注意力门控机制。Self-Attention通过对输入特征上下文的加权平均操作, 不仅使注意力模块能够灵活地关注图像的不同区域, 而且弥补了CNN(Convolutional Neural Network)网络局部性的缺点, 为神经网络带来更多的可能性。通过合成数据和实际数据证实, 该模型将传统卷积中的权值共享优点和Self-Attention动态计算注意力权重的优点结合, 提高了断层识别的精度, 与Unet相比, 验证损失下降了33.68%。模型不仅准确识别出了断层特征, 且比目前流行的深度学习方法更准确。

In order to improve the accuracy of fault identification, an improved Unet model is proposed. A multi-branch parallel structure M-block (Multi-branch block) is designed for the encoder part. It can capture multi-scale context information, and multi branch parallel structure will bring high performance benefits. Self-Attention block and attention gating mechanism are added to the decoder. Self-Attention not only enables the attention module to flexibly focus on different areas of the image, but also makes up for the shortcomings of the local CNN (Convolutional Neural Network) and brings more possibilities to the neural network through the weighted average operation of the input feature context. It is verified by synthetic data and actual data that the model combines the advantages of weight sharing in traditional convolution with the advantages of Self Attention's dynamic calculation of Attention weight to improve the accuracy of fault identification. Compared with Unet, the test loss is reduced by 33.68%. The model not only identifies fault features accurately, but also is more accurate than the current popular depth learning method.