| 基于卷积神经网络迁移学习模型的矿岩智能识别方法 |
| Received:May 01, 2023 Revised:August 11, 2023 点此下载全文 |
| 引用本文：ZHAO XingDong,WANG HongYu,BAI Ye.2023.Mineralized and barren rock intelligent identification method based on convolutional neural network transfer learning model[J].Mineral Deposits,42(5):1003~1010 |
| Hits: 1346 |
| Download times: 1097 |
| Author Name | Affiliation | | ZHAO XingDong | Laboratory for Safe Mining in Deep Metal Mine, Northeastern University, ShenYang 110004, Liaoning, China | | WANG HongYu | Laboratory for Safe Mining in Deep Metal Mine, Northeastern University, ShenYang 110004, Liaoning, China | | BAI Ye | Laboratory for Safe Mining in Deep Metal Mine, Northeastern University, ShenYang 110004, Liaoning, China |
|
| 基金项目:本文得到NSFC-山东联合基金项目（编号：U1806208）、国家自然科学基金重点项目（编号：52130403）和中央高校基本科研业务费项目（编号：N2001033）联合资助 |
|
| 中文摘要:文章基于Inception-v3卷积神经网络模型，通过对采集的金矿石、铜矿石、铁矿石、铅锌矿、花岗岩、片麻岩、大理岩和页岩，8种岩石453张图像进行特征提取和迁移学习，建立了岩性分类的迁移学习模型，实现了岩性的自动识别和分类。每种岩石图像随机抽取4张作为测试集进行测试，剩余421张图像作为训练集参加训练，经测试全部图像的岩性分类结果均正确，识别正确率超过80%的岩石图像占测试集图像总数的90%以上。识别正确率未达到80%的图像经过处理后重新训练并测试，其识别正确率均超过了80%，表明了该模型具有良好的岩性识别能力且鲁棒性较好，为岩性识别和自动分类提供了一种新的智能分析方法。 |
| 中文关键词:岩性识别 图像分类 迁移学习 卷积神经网络 智能识别 |
| |
| Mineralized and barren rock intelligent identification method based on convolutional neural network transfer learning model |
|
|
| Abstract:Based on the Inception-v3 convolutional neural network model, the article features 453 images of 8 kinds of rocks, including gold ore, copper ore, iron ore, lead-zinc ore, granite, gneiss, marble and shale. Extraction and transfer learning, the transfer learning model of lithology classification is established, and the automatic identification and classification of lithology is realized. Four images of each type of rock were randomly selected as the test set for testing, and the remaining 421 images were used as the training set to participate in the training. After testing, the lithology classification results of all images were correct, and the rock images with a recognition accuracy rate of more than 80% accounted for more than 90% of the total number of images in the test set. The images whose recognition accuracy rate did not reach 80% were retrained and tested after processing, and the recognition accuracy rate exceeded 80%, indicating that the model has good lithology recognition ability and good robustness, and is an important tool which provides a new intelligent analysis method for lithology recognition and automatic classification. |
| keywords:lithology identification image classification transfer learning convolutional neural network intelligent identification |
| View Full Text View/Add Comment Download reader |
|
|
|
