研究背景与问题Background & Problem

地质钻孔报告以非结构化文本形式存储于各地质调查局数据库中，岩性描述存在严重的命名不一致问题（同一岩性可能有数十种不同描述方式），阻碍了大规模三维地质建模的数据准备。 Geological drillhole reports are stored as unstructured text in geological survey databases. Lithological descriptions suffer from severe naming inconsistency (the same lithology may have dozens of different descriptions), hindering data preparation for large-scale 3D geological modeling.

核心方法Core Methodology

• 三段式数据模型：将钻孔数据标准化为孔口(collar)、测斜(survey)、岩性(lithology)三个核心间隔表 Three-part Data Model: Standardizes drillhole data into collar, survey, and lithology interval tables
• 词库驱动匹配：构建标准化岩性词库，将原始描述映射到标准术语 Lexicon-driven Matching: Builds standardized lithology lexicon to map raw descriptions to standard terms
• 模糊匹配算法：采用模糊字符串匹配(fuzzy string matching)处理拼写变体和同义词 Fuzzy Matching: Uses fuzzy string matching to handle spelling variants and synonyms
• 多级分类体系：支持从详细岩性到粗粒度类别的多级归并，适配不同尺度的三维建模需求 Multi-level Taxonomy: Supports aggregation from detailed lithology to coarse categories for different modeling scales

创新点Innovations

• 首次将完整的钻孔数据提取流程开源为可复用Python库 First open-source Python library providing a complete drillhole data extraction pipeline
• 提出词库+模糊匹配的两阶段岩性命名消歧方法 Proposes a two-stage lithology name disambiguation method: lexicon + fuzzy matching
• 多级岩性分类机制解决了从区域建模到局部建模的数据粒度适配问题 Multi-level classification mechanism solves data granularity adaptation from regional to local modeling

研究背景与问题Background & Problem

从钻孔岩性数据恢复地层序列是三维地质建模的关键步骤，但传统人工解释效率低且主观性强。如何自动从岩性描述中推断地层序列，并量化解释的不确定性，是核心科学问题。 Recovering stratigraphic sequences from drillhole lithology data is a key step in 3D geological modeling, but traditional manual interpretation is inefficient and subjective. Automatically inferring stratigraphic sequences from lithological descriptions while quantifying uncertainty is the core scientific problem.

核心方法Core Methodology

• 系统搜索算法：通过搜索算法系统探索所有地质合理的地层排序方案，而非单一确定性解 Systematic Search: Explores all geologically plausible stratigraphic orderings via search algorithms, not a single deterministic solution
• 地质合理性约束：引入地质规则约束搜索空间，如地层叠覆律、岩性过渡概率等 Geological Plausibility Constraints: Introduces geological rules (superposition law, lithology transition probabilities) to constrain search space
• 跨钻孔对比：结合多钻孔间地层单元的拓扑关系进行对比验证，增强解释可靠性 Cross-hole Correlation: Correlates stratigraphic units across multiple drillholes using topological relationships to enhance reliability
• 不确定性量化：输出多个地质一致性解及其概率分布，为建模提供风险依据 Uncertainty Quantification: Outputs multiple geologically consistent solutions with probability distributions, providing risk basis for modeling

创新点Innovations

• 将地层解释从确定性问题重新定义为概率推断问题，首次实现全自动不确定性量化 Redefines stratigraphic interpretation from a deterministic to a probabilistic inference problem, achieving fully automated uncertainty quantification
• 提出跨钻孔地层单元拓扑关系对比方法，将单孔解释的可靠性提升到多孔联合约束层面 Proposes cross-hole stratigraphic unit topological correlation, elevating single-hole interpretation reliability to multi-hole joint constraint level
• 开源算法litho2strat 1.0可直接集成到三维地质建模工作流中 Open-source algorithm litho2strat 1.0 can be directly integrated into 3D geological modeling workflows

研究背景与问题Background & Problem

钻孔柱状图图像的文字识别准确率受限于复杂布局（多栏、旋转文字、表格线干扰等），传统OCR直接应用效果不佳。 Text recognition accuracy in borehole log images is limited by complex layouts (multi-column, rotated text, table line interference, etc.), where traditional OCR performs poorly.

核心方法Core Methodology

• 多布局调整策略：对原始图像进行多种布局变换（如分栏切割、旋转校正、对比度增强等），生成多个候选输入 Multi-layout Adjustment: Applies various layout transformations (column splitting, rotation correction, contrast enhancement) to generate multiple candidate inputs
• 多OCR结果生成：每个候选布局独立运行OCR，产生多个识别结果 Multiple OCR Results: Each candidate layout runs OCR independently, producing multiple recognition results
• 投票机制整合：设计投票算法对多路OCR结果进行置信度加权整合，输出最优识别结果 Voting Integration: Designed voting algorithm performs confidence-weighted integration of multi-path OCR results to output optimal recognition

创新点Innovations

• 提出”多布局调整+投票”的新范式，突破了单一路径OCR的性能瓶颈 Proposes a “multi-layout adjustment + voting” paradigm, breaking through the performance bottleneck of single-path OCR
• 在大尹格庄和丁家山两个真实地质数据集上验证，F1值分别达到97.96%和94.36% Validated on two real geological datasets (Dayingezhuang and Dingjiashan), achieving F1 scores of 97.96% and 94.36% respectively
• 方法通用性强，可推广到其他复杂表格文档的OCR场景 Method is generalizable and can be extended to other complex table document OCR scenarios

研究背景与问题Background & Problem

区域地质调查产生大量钻孔数据，但缺乏系统化的数据库建设方法，数据质量参差不齐，难以支撑后续的三维地质建模和资源共享。 Regional geological surveys produce large amounts of drillhole data, but systematic database construction methods are lacking, data quality is inconsistent, and supporting 3D geological modeling and resource sharing is difficult.

核心方法Core Methodology

• 多层数据模型：建立1个钻孔基本信息图层 + 5个钻孔编录数据库 + 5个钻孔设计数据库 + 5张钻孔柱状图的分层组织结构 Multi-layer Data Model: Establishes a hierarchical structure: 1 basic borehole info layer + 5 drilling log databases + 5 drilling design databases + 5 borehole columnar diagrams
• 数据质量控制体系：从数据来源、处理方法、属性结构三个维度建立质量控制流程 Data Quality Control System: Establishes QC processes from three dimensions: data source, processing method, and attribute structure
• 标准化属性结构：统一钻孔编号、坐标、孔口标高、钻机类型、开终孔日期等字段规范 Standardized Attribute Structure: Unifies field specifications for borehole ID, coordinates, collar elevation, rig type, start/end dates, etc.

创新点Innovations

• 首次系统阐述了1:5万比例尺第四系钻孔数据库的完整建设方法 First systematic description of complete construction methods for 1:50,000 scale Quaternary borehole databases
• 提出的分层数据组织结构兼顾了原始编录数据、设计数据和成果图件的管理需求 The proposed hierarchical data organization accommodates management needs for original log data, design data, and output maps

研究背景与问题Background & Problem

传统钻孔柱状图与特定CAD平台强绑定，不同单位使用不同软件导致图件无法互通，数据复用困难。 Traditional borehole columnar diagrams are tightly bound to specific CAD platforms. Different organizations using different software prevents interoperability and data reuse.

核心方法Core Methodology

• 通用数据模型：设计独立于任何平台的钻孔通用数据模型，将岩性、深度、描述等要素抽象为平台无关的数据结构 Universal Data Model: Designs a platform-independent universal borehole data model, abstracting lithology, depth, descriptions into platform-agnostic data structures
• 岩性花纹自动填充算法：根据岩性编码自动匹配标准花纹库，按深度间隔自动填充岩性柱 Automatic Lithology Pattern Fill: Automatically matches standard pattern libraries by lithology code and fills lithology columns by depth interval
• 矢量图元生成：生成独立于平台的矢量图元数据集，可被任何支持矢量的软件渲染 Vector Primitive Generation: Generates platform-independent vector primitive datasets renderable by any vector-supporting software

创新点Innovations

• 提出”一次生成，到处使用”的跨平台理念，突破了CAD平台锁定 Proposes the “generate once, use everywhere” cross-platform concept, breaking through CAD platform lock-in
• 岩性花纹自动填充算法解决了大量钻孔批量成图的关键技术瓶颈 The automatic lithology pattern fill algorithm solves the key technical bottleneck for batch plotting of large numbers of boreholes

研究背景与问题Background & Problem

历史钻孔记录大量使用打字机填写，这些纸质/扫描件表格的数字化面临独特挑战，与印刷体或手写体文档有本质不同。 Historical drilling logs were largely typewritten. Digitizing these paper/scanned forms faces unique challenges fundamentally different from printed or handwritten documents.

核心方法与技术路径Core Methods & Technical Paths

论文分析了三大技术路径： The paper analyzes three technical paths:

1. 表格识别路径Table Recognition Path
   识别表格线结构 → 定位单元格 → 提取单元格内文字 Recognize table line structure → Locate cells → Extract text within cells
2. 语义关联路径Semantic Association Path
   识别字段标签 → 关联邻近数值 → 构建键值对 Recognize field labels → Associate nearby values → Build key-value pairs
3. 数值提取路径Numerical Extraction Path
   针对深度、厚度、百分比等数值型字段的特殊提取策略 Special extraction strategies for numerical fields like depth, thickness, percentages

关键挑战Key Challenges

• 打字机字体等宽特性导致的列对齐歧义 Column alignment ambiguity caused by monospace typewriter fonts
• 复写纸/碳带造成的字符粘连和模糊 Character adhesion and blurring from carbon paper/ribbon
• 表格线缺失或断裂导致的结构识别困难 Structural recognition difficulty due to missing or broken table lines
• 语义关联错误（如深度值与岩性描述错配） Semantic association errors (e.g., depth values mismatched with lithology descriptions)