【摘要】 地震勘探是探测地下油气资源的重要方法之一，利用人工地震剖面数据多种特征识别储层结构、岩性、物性等是地球物理勘探的关键问题。随着数据量的增大，对地震数据进行降维和信号特征编码成为快速处理和分析地震资料的重要方向。由于手工标记拾取地震数据特征耗时费力、效率不高，而且不同人员往往会产生不同的解释结果，为此，基于地震剖面数据与人类指纹之间存在的相似性特征，提出了地震数据指纹特征点提取和自动标签化方法。该方法通过识别地震剖面数据间断、分叉等特征点，从海量地震剖面数据中提取存储量小、信息丰富的指纹特征点，并实现指纹特征点阵列(dactylogram minutiae array, DMA)编码算法。实际三维地震数据体处理结果表明，指纹特征点和DMA编码可以将数据存储量降低2个数量级，同时，编码本身包含特征点位置、方位角等信息，能够将编码从一维字符串恢复为二维特征点矩阵，并重构出地震剖面的特征点分布。勘探区域地下结构和含油气情况决定了地震数据指纹特征点的分布，因此特征点具有唯一性特点。利用指纹特征点匹配算法可以在三维数据体中自动实现相似特征的标签化，为深度学习提供大量训练数据集合。基于指纹特征点识别和自动标签化算法具有大规模数据降维和标准化编码能力，适合于快速生成人工智能算法的训练数据，能够进一步挖掘现有数据中的信息，为地震数据处理提供新的数据资料特征。更多还原

【Abstract】 Seismic exploration is one of the most important methods for detecting underground oil and gas resources.The main challenge it faces is the identification of the reservoir structure, lithology, physical properties, and other parameters by using various features of artificial seismic profile data.Traditional data acquisition and storage methods record time-domain signals on all data tracks.With the increase in the data collection area, data storage and computational capabilities become an inevitable limitation.Therefore, dimension reduction and signal feature coding have become essential to achieve the fast processing and analysis of seismic data.The feature extraction method based on manual marking and picking up is time-consuming and inefficient, and the results are affected by human factors.On the other hand, the parameterized modeling method is usually only effective for some special datasets, which reduces the performance of large-scale data processing.Based on the similarity between two-dimensional seismic profile data and human fingerprints, this paper proposes a method for extracting dactylogram minutiae(DM) from seismic data and automatic labeling techniques.By identifying discontinuous and branching features in seismic profile data, the method extracts DM from massive seismic profile data with less storage capacity requirements and richer information.The distribution of seismic DM is determined by the underground structure and the oil and gas contents in the exploration area.Therefore, the feature points are unique and can reflect the reservoir structure and petrophysical characteristics.The algorithm based on DM recognition and automatic labeling is characterized by large-scale data reduction and standardized coding, which is suitable for the rapid generation of training data for artificial intelligence algorithms.The algorithm can also mine useful information from existing data and provide additional data features for seismic data processing.更多还原