基于深度学习的螺旋选矿机矿物分带图像特征信息提取方法研究  

Research on Feature Extraction Technology of Mineral Zoning Image of Spiral Concentrator Based on Deep Learning

作　　者：刘惠中[1,2] 宁剑 邹起华 闻成钰 LIU Huizhong;NING Jian;ZOU Qihua;WEN Chengyu(School of Mechanical and Electrical Engineering,Jiangxi University of Science and Technology,Ganzhou 341000,China;Jiangxi Province Engineering Research Center for Mechanical and Electrical of Mining and Metallurgy,Ganzhou 341000,China)

机构地区：[1]江西理工大学机电工程学院,江西赣州341000 [2]江西省矿冶机电工程技术研究中心,江西赣州341000

出　　处：《有色金属工程》2022年第12期91-99,共9页Nonferrous Metals Engineering

基　　金：国家自然科学基金资助项目(52164019);江西省2021年度研究生创新专项资金项目(YC2021-S575);江西省“双千计划”引进高层次创新人才项目(jxsq2018101046)。

摘　　要：螺旋选矿机是一种流膜重力选矿设备,目前其产品的截取是通过工人观察矿物分带并根据精矿与尾矿的分割位置,调节截取器的分矿块到相应的位置以实现对精矿的准确截取,获得合格的精矿产品。由于每个工人的经验和技术水平不一样,难以保证获取的矿带位置信息和操作的准确性,而容易造成选矿指标的波动。为了准确、快速、自动地获取螺旋选矿机矿物分带的位置信息,针对螺旋选矿机矿带分界模糊、识别难度大等难题,提出了一种优化的Canny边缘检测算法和基于深度学习的HED(Holistically-Nested Edge Detection)边缘检测算法,并分别对螺旋选矿机矿物分带图像进行了矿带分割位置提取试验。试验结果表明,基于深度学习的HED算法优于传统的边缘检测算法,其识别的精度可以满足生产中对螺旋选矿机矿带分割特征信息识别的要求。Spiral concentrator is a kind of flow film gravity beneficiation equipment.At present,its products are intercepted by workers observing the mineral zoning and adjusting the ore block of interceptor to the corresponding position according to the separation position of concentrate and tailings,so as to achieve accurate interception of concentrate and obtain qualified concentrate products.Due to the different experience and technical level of each worker,it is difficult to ensure the accuracy of the obtained ore belt location information and operation,which is easy to cause fluctuations in beneficiation indicators.In order to obtain the position information of mineral zoning of spiral concentrator accurately,quickly and automatically,aiming at the problems such as fuzzy ore zone boundary and difficult recognition of spiral concentrator,this paper proposes an optimized Canny edge detection algorithm and HED(Holistically-Nested Edge Detection)edge detection algorithm based on deep learning,and carries out the ore zone segmentation position extraction experiments on the mineral zoning images of spiral concentrator.The experimental results show that the HED algorithm based on deep learning is superior to the traditional edge detection algorithm,and its recognition accuracy can meet the requirements of spiral concentrator ore belt segmentation feature information recognition in production.

关 键 词：螺旋选矿机 重力选矿 图像识别技术 边缘检测算法 

分 类 号：TD455[矿业工程—矿山机电] TD922