Mineral Prospecting Prediction via Transfer Learning Based on Geological Big Data: A Case Study of Huayuan, Hunan, China
In the big data era, mineral explorations need to accommodate for the growth in spatial dimensions and data dimensions, as well as the data volume and the correlation between data. Aiming to overcome the problems of limited and scattered data sources, chaotic data types, questionable data quality, a...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-03-01
|
| Series: | Minerals |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2075-163X/13/4/504 |
| _version_ | 1797604149560344576 |
|---|---|
| author | Shi Li Chang Liu Jianping Chen |
| author_facet | Shi Li Chang Liu Jianping Chen |
| author_sort | Shi Li |
| collection | DOAJ |
| description | In the big data era, mineral explorations need to accommodate for the growth in spatial dimensions and data dimensions, as well as the data volume and the correlation between data. Aiming to overcome the problems of limited and scattered data sources, chaotic data types, questionable data quality, asymmetric data information, and small sample sizes in current mineral prospecting data, this paper improved traditional 3D prediction methods based on the characteristics and actual needs of relevant mineral prospecting data. First, for the regions with incomplete data, a new 3D prediction method based on transfer learning was proposed. Meanwhile, random noise was adopted to compensate for the limited sample size in mineral prediction. By taking the Huayuan Mn deposit in Hunan Province as the study area, 22 proposed ore-controlling variables were divided into six groups for comparative tests under different combinations, and each group was further divided into the 3D CNN prediction method and the transfer learning prediction method. After the similarities between the regional metallogenic backgrounds were proven, the convolution kernel of the Minle area was transferred to that of the Huayuan area with poor data. Then, both were used to train a 3D prediction model to realize the training and transfer of the spatial correlation between the spatial distribution of ore-controlling factors and the manganese ore. The results indicated that the accuracy of the transfer learning model in test 6 could reach 100%, with good stability of the transfer learning prediction model and a high convergence speed. By comparing the 3D-predicted targets before and after the transfer learning of tests 5 and 6, it was found that the 3D CNN model of test 5 still performed well, but the transfer learning model of test 6 was superior. In verifications based on superposition with the basin model and the growth fault model, the prediction results were consistent with the geological characteristics of the research area. To sum up, the 3D CNN prediction method has advantages in mineral prediction when big data are available, and transfer learning based on the 3D CNN algorithm helps to realize 3D deep mineral prospect prediction in the case of incomplete data. |
| first_indexed | 2024-03-11T04:42:17Z |
| format | Article |
| id | doaj.art-230b593c0d0c40c8b204de12432eb86a |
| institution | Directory Open Access Journal |
| issn | 2075-163X |
| language | English |
| last_indexed | 2024-03-11T04:42:17Z |
| publishDate | 2023-03-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Minerals |
| spelling | doaj.art-230b593c0d0c40c8b204de12432eb86a2023-11-17T20:35:37ZengMDPI AGMinerals2075-163X2023-03-0113450410.3390/min13040504Mineral Prospecting Prediction via Transfer Learning Based on Geological Big Data: A Case Study of Huayuan, Hunan, ChinaShi Li0Chang Liu1Jianping Chen2School of Information, Beijing Wuzi University, Beijing 101149, ChinaSchool of Earth Sciences and Resources, China University of Geosciences (Beijing), Beijing 100083, ChinaSchool of Earth Sciences and Resources, China University of Geosciences (Beijing), Beijing 100083, ChinaIn the big data era, mineral explorations need to accommodate for the growth in spatial dimensions and data dimensions, as well as the data volume and the correlation between data. Aiming to overcome the problems of limited and scattered data sources, chaotic data types, questionable data quality, asymmetric data information, and small sample sizes in current mineral prospecting data, this paper improved traditional 3D prediction methods based on the characteristics and actual needs of relevant mineral prospecting data. First, for the regions with incomplete data, a new 3D prediction method based on transfer learning was proposed. Meanwhile, random noise was adopted to compensate for the limited sample size in mineral prediction. By taking the Huayuan Mn deposit in Hunan Province as the study area, 22 proposed ore-controlling variables were divided into six groups for comparative tests under different combinations, and each group was further divided into the 3D CNN prediction method and the transfer learning prediction method. After the similarities between the regional metallogenic backgrounds were proven, the convolution kernel of the Minle area was transferred to that of the Huayuan area with poor data. Then, both were used to train a 3D prediction model to realize the training and transfer of the spatial correlation between the spatial distribution of ore-controlling factors and the manganese ore. The results indicated that the accuracy of the transfer learning model in test 6 could reach 100%, with good stability of the transfer learning prediction model and a high convergence speed. By comparing the 3D-predicted targets before and after the transfer learning of tests 5 and 6, it was found that the 3D CNN model of test 5 still performed well, but the transfer learning model of test 6 was superior. In verifications based on superposition with the basin model and the growth fault model, the prediction results were consistent with the geological characteristics of the research area. To sum up, the 3D CNN prediction method has advantages in mineral prediction when big data are available, and transfer learning based on the 3D CNN algorithm helps to realize 3D deep mineral prospect prediction in the case of incomplete data.https://www.mdpi.com/2075-163X/13/4/504big datamineral prospection mappingtransfer learning3D convolutional neural networksHuayuan Mn deposit |
| spellingShingle | Shi Li Chang Liu Jianping Chen Mineral Prospecting Prediction via Transfer Learning Based on Geological Big Data: A Case Study of Huayuan, Hunan, China Minerals big data mineral prospection mapping transfer learning 3D convolutional neural networks Huayuan Mn deposit |
| title | Mineral Prospecting Prediction via Transfer Learning Based on Geological Big Data: A Case Study of Huayuan, Hunan, China |
| title_full | Mineral Prospecting Prediction via Transfer Learning Based on Geological Big Data: A Case Study of Huayuan, Hunan, China |
| title_fullStr | Mineral Prospecting Prediction via Transfer Learning Based on Geological Big Data: A Case Study of Huayuan, Hunan, China |
| title_full_unstemmed | Mineral Prospecting Prediction via Transfer Learning Based on Geological Big Data: A Case Study of Huayuan, Hunan, China |
| title_short | Mineral Prospecting Prediction via Transfer Learning Based on Geological Big Data: A Case Study of Huayuan, Hunan, China |
| title_sort | mineral prospecting prediction via transfer learning based on geological big data a case study of huayuan hunan china |
| topic | big data mineral prospection mapping transfer learning 3D convolutional neural networks Huayuan Mn deposit |
| url | https://www.mdpi.com/2075-163X/13/4/504 |
| work_keys_str_mv | AT shili mineralprospectingpredictionviatransferlearningbasedongeologicalbigdataacasestudyofhuayuanhunanchina AT changliu mineralprospectingpredictionviatransferlearningbasedongeologicalbigdataacasestudyofhuayuanhunanchina AT jianpingchen mineralprospectingpredictionviatransferlearningbasedongeologicalbigdataacasestudyofhuayuanhunanchina |