中国科技核心期刊
美国化学文摘社（CAS）数据库
美国EBSCO学术数据库
日本科学技术振兴机构数据库(JST)
岩性智能识别技术发展现状与实践
- 作者：
黄麟淇，魏云峰，李夕兵
- 作者单位：
- （中南大学资源与安全工程学院）
基金项目：
- 国家自然科学基金项目（52174098）
详细信息：
- 作者简介：
- 黄麟淇（1987—），女，教授，博士，从事矿山微震监测和安全预警等研究工作；E‑mail：huanglinqi@csu. edu. cn
- PDF下载
Current development and practice of smart lithology identification technology
- English Author：
Huang Linqi, Wei Yunfeng, Li Xibing
- Unit：
- (School of Resources and Safety Engineering, Central South University)
- 摘要
- 在线预览
- 参考文献
摘要:
岩性识别技术是分辨目标岩石类型，从而掌握地质构造特征和矿产分布的有力手段。传统岩性识别技术（如岩石手标本/薄片显微鉴定、遥感法和测井法等）存在主观性强、效率低、抗干扰性弱、工艺复杂等局限。基于传统机器学习的岩性自动化识别方法虽一定程度上突破了传统岩性识别技术的瓶颈，但丰富的岩性特征信息难以被充分利用。基于深度学习的岩性智能识别技术逐渐表现出优异性能。基于现有深度学习模型，详细介绍VGG-16卷积神经网络通过迁移学习来识别岩性的技术流程，该技术不但可以降低对岩性标注数据量的依赖，还能通过自动捕捉岩性的高维特征信息来增加识别的泛化能力。此外，岩性智能识别技术将会在以下方向进行创新发展：构建包含多矿种、多尺度的标准岩石图像数据库；开发现场实时岩性识别系统，轻量化处理识别模型；融合岩石视觉、光谱、物理属性等多维度的数据信息，通过协同分析方法来提升智能识别模型的可靠性；构建矿山岩性动态感知系统，推动精准采矿向全流程智能化升级。
关键词:
矿山勘探；岩性识别；人工智能；深度学习；VGG-16；机器学习
Abstract:
Lithology identification technology is a powerful means to distinguish the type of target rock, therebymastering the geological structural characteristics and mineral distribution. Traditional lithology identification technolo‑gies（such as rock hand specimen/thin section microscopic identification, remote sensing, and logging methods）havelimitations such as strong subjectivity, low efficiency, weak anti‑interference ability, and complex processes. Althoughrock lithology automatic identification methods based on traditional machine learning have broken through the bottle‑necks of traditional lithology identification technology to some extent, the rich lithology feature information is not fullyutilized. Smart lithology identification technology based on deep learning has gradually shown excellent performance.Based on existing deep learning models, this paper introduces the technical process of using the VGG-16 convolutionalneural network to recognize lithology through transfer learning, which not only reduces the dependency on the amountof lithology annotation data but also increases the generalization ability of identification by automatically capturing thehigh‑dimensional feature information of lithology. In addition, the development and innovation of smart lithology identi‑fication technology will be carried out in the following directions: constructing a standard rock image database containingmultiple ore types and multiple scales; developing an on‑site real‑time lithology identification system with lightweightprocessing of identification models; integrating multi‑dimensional data information such as rock vision, spectral, andphysical properties, and improving the reliability of intelligent identification models through collaborative analysismethods; and constructing a mine lithology dynamic sensing system to promote the smart upgrading of the wholeprocess of precision mining.
Keywords:
mine exploration; lithology identification; artificial intelligence; deep learning; VGG-16; machinelearning13
