工程科学学报.第42卷，第12期：1597-1604.2020年12月 Chinese Journal of Engineering,Vol.42,No.12:1597-1604,December 2020 https://doi.org/10.13374/j.issn2095-9389.2020.01.02.001;http://cje.ustb.edu.cn 卷积神经网络在矿区预测中的研究与应用 袁传新12，贾东宁12)四，周生辉) 1)中国海洋大学信息科学与工程学院，青岛2660002)青岛海洋科学与技术试点国家实验室高性能科学计算与系统仿真平台，青岛 266000 ☒通信作者，E-mail:jiadn@ouc.edu.cn 摘要在研究富钴结壳高产区地形特征基础上，以富钴结壳站点地理坐标为中心，获得了一平方公里的海拔高度数值矩阵 作为地形特征.使用卷积神经网络的分析方法对数值矩阵进行训练，学习坡度和平整度等区域特征，将富钴结壳站点地形和 其他海底地形进行区分.依据训练后获得的模型，对富钴结壳高产区进行预测，取得了较好的预测效果，结合其他因素的影 响，可以提高结壳靶区选取的精准度 关键词富钴结壳：海底地形：数值矩阵：卷积神经网络：矿区预测 分类号P744.3:TP183 Research and application of convolutional neural network in mining area prediction YUAN Chuan-xin2),JIA Dong-ning2,ZHOU Sheng-hu 1)College of Information Science and Engineering,Ocean University of China,Qingdao 266000,China 2)High-performance Scientific Computing and System Simulation Platform,Pilot National Laboratory for Marine Science and Technology (QingDao), Qingdao 266000,China Corresponding author,E-mail:jiadn@ouc.edu.cn ABSTRACT Cobalt-rich crusted deposits are found all over the world's oceans,and their distribution is closely related to the submarine topography.The determination of crusting area is the basic work for the exploration and mining of these deposits.Many factors affect the accumulation of crusts,and topography is a crucial factor.Mineralization forecast requires comprehensive consideration of geological background and experts'views and opinions,the prior knowledge of prospectors is the biggest factor affecting the results.In the course of ocean research,especially with the rapid development of space information technology,a huge amount of ocean data that cover about 70%of the total surface area have been accumulated rapidly;how to extract valuable information from large,fast,complex,and multisource data has become a hot topic in current ocean research.Machine learning-and deep learning- related research methods can read feature signs from mineral data to obtain existing mineral knowledge to further serve mine prediction work.Based on the study of terrain features of cobalt-rich crust in high-producing areas,the numerical matrix of altitude of 1 km2ocean surface was obtained,with the geographical coordinates of cobalt-rich crust sites as the center.Using the analysis method of convolutional neural network,the numerical matrix is trained to learn regional features such as slope and flatness and to distinguish the cobalt-rich crust-crust site topography from other submarine topography.According to the training model,the high-producing cobalt- rich crusting area was predicted and better forecasting value is obtained.Meanwhile,the accuracy of the selection of crusting target area was improved by combining the influence of other factors. KEY WORDS cobalt-rich crust;seafloor terrain;numerical matrix;convolutional neural network;mining area forecast 收稿日期：2020-01-02 基金项目：海洋大数据中心资助项目(2018SDPT01)卷积神经网络在矿区预测中的研究与应用 袁传新1,2)，贾东宁1,2) 苣，周生辉2) 1) 中国海洋大学信息科学与工程学院，青岛 266000    2) 青岛海洋科学与技术试点国家实验室高性能科学计算与系统仿真平台，青岛 266000 苣通信作者，E-mail：jiadn@ouc.edu.cn 摘    要    在研究富钴结壳高产区地形特征基础上，以富钴结壳站点地理坐标为中心，获得了一平方公里的海拔高度数值矩阵 作为地形特征. 使用卷积神经网络的分析方法对数值矩阵进行训练，学习坡度和平整度等区域特征，将富钴结壳站点地形和 其他海底地形进行区分. 依据训练后获得的模型，对富钴结壳高产区进行预测，取得了较好的预测效果，结合其他因素的影 响，可以提高结壳靶区选取的精准度. 关键词    富钴结壳；海底地形；数值矩阵；卷积神经网络；矿区预测 分类号    P744.3; TP183 Research and application of convolutional neural network in mining area prediction YUAN Chuan-xin1,2) ，JIA Dong-ning1,2) 苣 ，ZHOU Sheng-hui2) 1) College of Information Science and Engineering, Ocean University of China, Qingdao 266000, China 2) High-performance Scientific Computing and System Simulation Platform, Pilot National Laboratory for Marine Science and Technology (QingDao), Qingdao 266000, China 苣 Corresponding author, E-mail: jiadn@ouc.edu.cn ABSTRACT    Cobalt-rich  crusted  deposits  are  found  all  over  the  world ’s  oceans,  and  their  distribution  is  closely  related  to  the submarine  topography.  The  determination  of  crusting  area  is  the  basic  work  for  the  exploration  and  mining  of  these  deposits.  Many factors  affect  the  accumulation  of  crusts,  and  topography  is  a  crucial  factor.  Mineralization  forecast  requires  comprehensive consideration  of  geological  background  and  experts ’  views  and  opinions,  the  prior  knowledge  of  prospectors  is  the  biggest  factor affecting the results. In the course of ocean research, especially with the rapid development of space information technology, a huge amount of ocean data that cover about 70% of the total surface area have been accumulated rapidly; how to extract valuable information from large, fast, complex, and multisource data has become a hot topic in current ocean research. Machine learning- and deep learning￾related research methods can read feature signs from mineral data to obtain existing mineral knowledge to further serve mine prediction work. Based on the study of terrain features of cobalt-rich crust in high-producing areas, the numerical matrix of altitude of 1 km2 ocean surface  was  obtained,  with  the  geographical  coordinates  of  cobalt-rich  crust  sites  as  the  center.  Using  the  analysis  method  of convolutional neural network, the numerical matrix is trained to learn regional features such as slope and flatness and to distinguish the cobalt-rich crust–crust site topography from other submarine topography. According to the training model, the high-producing cobalt￾rich crusting area was predicted and better forecasting value is obtained. Meanwhile, the accuracy of the selection of crusting target area was improved by combining the influence of other factors. KEY WORDS    cobalt-rich crust；seafloor terrain；numerical matrix；convolutional neural network；mining area forecast 收稿日期: 2020−01−02 基金项目: 海洋大数据中心资助项目（2018SDPT01） 工程科学学报，第 42 卷，第 12 期：1597−1604，2020 年 12 月 Chinese Journal of Engineering, Vol. 42, No. 12: 1597−1604, December 2020 https://doi.org/10.13374/j.issn2095-9389.2020.01.02.001; http://cje.ustb.edu.cn