基于目标检测的地质异常信息提取——以湖南香花岭地区为例  

Extraction of geological anomaly information based on target detection:A case study of Xianghualing area in Hunan Province

作　　者：李苍柏 李楠[1,2] 宋相龙[1,2] Li Cangbai;Li Nan;Song Xianglong(Institute of Mineral Resources,Chinese Academy of Geological Sciences,Beijing 100037,China;China University of Geosciences（Beijing）,Beijing 100083,China)

机构地区：[1]中国地质科学院矿产资源研究所,北京100037 [2]中国地质大学(北京),北京100083

出　　处：《地质学刊》2018年第3期434-439,共6页Journal of Geology

基　　金：国家自然科学基金项目"基于地质先验模型的区域大比例尺三维地质建模关键技术研究"(41672330);国土资源部公益性行业科研专项"地质大数据技术研究与应用试点"(201511079-04)

摘　　要：卷积神经网络在图像识别领域处于领先地位,在目标检测方面的应用也越来越广泛,其特点是可以提取点与点之间的相关关系。二维地质图中,点与点之间往往存在特定的空间关系,将卷积神经网络技术应用于地质异常信息的提取有其重要性。在讨论卷积神经网络以及基于该技术目标检测算法(YOLO)的基础上,以湖南香花岭地区为例,提取与锡矿成矿相关的构造信息并进行分析,结果该方法能够覆盖原有矿点,有效地定义点与点之间的相关关系,描述点与点之间的空间相关性,可靠地提取与成矿有关的构造线密度信息,在花岗岩体与构造复杂地区圈出地质异常信息。Convolutional Neural Network （CNN） has achieved a leading position in image recognition. It is widely used in object detection thanks to the fact that it can extract the correlation between points. In mapping 2D geological maps, CNN can be used in extracting geological anomalies of the geological body which has specific spatial relationship among points. Based on CNN and object detectionalgorithm （YOLO）, this paper extracts the structure information related to the tin mineralization in the area of Xianghualing, HunanProvince. The result shows that this method can cover all the known mineral occurrences, effectively define spatial relationship amongpoints, describe their spatial correlation, accurately extract information of structural line intensity related to mineralization, and delineate geological anomaly in granitic intrusion and complicated structural areas.

关 键 词：卷积神经网络 目标检测 YOLO算法 构造信息 湖南香花岭 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]