Next Article in Journal
Reductions and Exact Solutions of Nonlinear Wave-Type PDEs with Proportional and More Complex Delays
Next Article in Special Issue
Analysis of Microscopic Pore Characteristics and Macroscopic Energy Evolution of Rock Materials under Freeze-Thaw Cycle Conditions
Previous Article in Journal
A Novel Context-Aware Reliable Routing Protocol and SVM Implementation in Vehicular Area Networks
Previous Article in Special Issue
Applications of Two Neuro-Based Metaheuristic Techniques in Evaluating Ground Vibration Resulting from Tunnel Blasting
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Tailings Pond Classification Based on Satellite Images and Machine Learning: An Exploration of Microsoft ML.Net

Civil Engineering Discipline, School of Engineering, Malaysia Campus, Monash University, Bandar Sunway 47500, Malaysia
*
Author to whom correspondence should be addressed.
Mathematics 2023, 11(3), 517; https://doi.org/10.3390/math11030517
Submission received: 18 November 2022 / Revised: 4 January 2023 / Accepted: 17 January 2023 / Published: 18 January 2023

Abstract

:
Mine pollution from mining activities is often widely recognised as a serious threat to public health, with mine solid waste causing problems such as tailings pond accumulation, which is considered the biggest hidden danger. The construction of tailings ponds not only causes land occupation and vegetation damage but also brings about potential environmental pollution, such as water and dust pollution, posing a health risk to nearby residents. If remote sensing images and machine learning techniques could be used to determine whether a tailings pond might have potential pollution and safety hazards, mainly monitoring tailings ponds that may have potential hazards, it would save a lot of effort in tailings ponds monitoring. Therefore, based on this background, this paper proposes to classify tailings ponds into two categories according to whether they are potentially risky or generally safe and to classify tailings ponds with remote sensing satellite images of tailings ponds using the DDN + ResNet-50 machine learning model based on ML.Net developed by Microsoft. In the discussion section, the paper introduces the environmental hazards of mine pollution and proposes the concept of “Healthy Mine” to provide development directions for mining companies and solutions to mine pollution and public health crises. Finally, we claim this paper serves as a guide to begin a conversation and to encourage experts, researchers and scholars to engage in the research field of mine solid waste pollution monitoring, assessment and treatment.

1. Introduction

1.1. Research Background

With the increasing frequency of mining activities worldwide, mine discharge brings plenty of environmental problems. Among these, mine solid waste discharge is considered one of the most serious environmental problems, and as mine solid waste has a low reuse rate compared to other solid waste, tailings ponds generally need to be built to stockpile mine solid waste [1,2].
There is no doubt that the construction of tailings ponds, while allowing for the storage of mine solid waste, is not a good thing for the environment and human beings: the construction of tailings ponds takes up a lot of land and causes damage to vegetation cover, while the leachate from the ponds can have a serious negative impact on the environment and public health. There are many cases (as shown in Table 1) which confirm this.
In 2003, Agrawal, A. et al. [3] introduced the world to the environmental impact and damage caused by solid waste discharge from the non-ferrous metal industry in India, such as leachate pollution; their research showed that metal recycling of solid waste from the non-ferrous metals industry would be effective in mitigating environmental pollution, and Shengo’s [4] review endorsed this practice of recovering metal resources from solid waste. In 2016, Liu, Y. et al. [5] suggested that solid waste discharges can lead to damage to the surface landscape, for example, tailings pond stockpiles, which not only occupy surface space but also bring a major safety hazard that would result in serious human casualties at the mine site in the event of a tailings pond failure; in the same year, 2016, Asif, Z. and Chen, Z. [6] argued that the land occupation of tailings pond stockpiles is indeed a nuisance, and therefore they advocated the use of mine solid waste for land reclamation.
Table 1. Research cases on the mine solid waste pollution.
Table 1. Research cases on the mine solid waste pollution.
Pollution IssueResearch CasesResearch AreaResearch PurposeResearch Findings
Mine Solid Waste Pollution (Tailings Ponds)Agrawal, A. et al., 2004 [3]India: Non-ferrous metals IndustryTo study solid waste pollution and management in the non-ferrous metals industry in India.The results showed that solid waste polluted surface water as well as groundwater, primarily through leachate, thus affecting farmland, rivers and public health. Additionally, the authors advocated that mines should commit to metal recycling of non-ferrous solid waste, which would mitigate solid waste pollution.
Liu, Y. et al., 2016 [5]China: Mining IndustryTo study the pollution of industrial solid waste in general (mining solid waste in particular) and to make recommendations related to solid waste management based on the current state of the resource and environmental development in China.The authors suggested that the problem of land occupation by solid waste (tailings pond stockpiling) from mines is very serious, especially in China; at the same time, tailings ponds are a major safety hazard that would result in serious human casualties at the mine site in the event of a tailings pond failure.
Asif, Z. et al., 2016 [6]North America: Mining IndustryTo discuss the challenges of environmental management, particularly solid waste management, in the North American mining industry.The author highlighted the hazards of land occupation from tailings pond accumulation, and the author recommended the use of non-hazardous mine solid waste for land reclamation.
Shengo, L. M. 2021 [4]Democratic Republic of the Congo: Mining IndustryIn order to explore the environmental issues related to the management of mineral waste in the mining industry in the Democratic Republic of the Congo.The recycling and reuse of non-ferrous solid waste were very important, not only to mitigate the problem of solid waste pollution but also to bring potential resource value.
In addition to the potential environmental pollution and health risks associated with tailings ponds, they are also a potential source of danger and can lead to potential safety incidents. If a tailings pond were to fail, it would be a huge disaster for the environment and the people living in the vicinity of the mine. In Brazil, serious tailings pond failures occurred in 2015 and 2019 [7], causing massive damage to homes and vehicles. In China, a tailings pond failure accident occurred in 2008 in Xiangfen Country, Shanxi, resulting in a large number of casualties and environmental damage [8,9].
Since 2010, the safety and environmental pollution hazards of tailings ponds have received increasing attention from researchers [10,11]. Based on the Google Scholar database (https://scholar.google.com; accessed on 14 November 2022), using “tailings ponds and safety” and “tailings ponds and environment” as the keywords, the number of studies related to both keywords for each three-year period from 2010 to 2020 was found, as shown in Figure 1. The number of related literature results in the last 10 years clearly has an upward trend, showing that the safety and environmental pollution hazards of tailings ponds are receiving the public’s increasing attention.

1.2. Research Purpose and Significance

Monitoring and management of tailings ponds are particularly important in order to avoid environmental pollution and safety accidents in tailings ponds [12]. However, monitoring tailings ponds is often very time-consuming and labour-intensive [13]; if remote sensing images and machine learning could be used to determine whether a tailings pond might have potential pollution and safety hazards and then mainly monitoring tailings ponds that may have potential hazards, it would save a lot of effort in tailings ponds monitoring [14,15].
As a result, this paper divides tailings ponds into two categories according to whether they are potentially risky: 1# Tailings Pond, which has potential environmental and safety hazards, and 2# Tailings Pond which has no obvious potential environmental and safety hazards. Combining the remote sensing images (satellite maps) with the results of the field surveys (as shown in Figure 2): it defines that 1# Tailings Pond is an unclosed tailings pond that has significant surface water leaching, thus posing a potential contamination and safety hazard (as shown in Figure 2A); it defines that 2# Tailings Pond is generally a closed (almost closed) tailings pond or a dry stockpile pond with no significant surface water leaching, which may show signs of land reclamation and can generally be considered to have no obvious potential environmental and safety hazards (as shown in Figure 2B).
Based on the features of the two categories of tailings ponds, this paper planned to implement the image identification and classification function of tailings ponds by building a machine learning model via ML.Net developed by Microsoft [16]. At the same time, this paper planned to explore the accuracy of the ML.Net machine learning framework and its machine learning model in classifying and identifying the two types of tailings ponds with different characteristics, providing a starting point for future remote sensing techniques to monitor tailings pond risk and pollution.

2. Materials and Methods

2.1. Machine Learning Model

The training environment used was local training on a computer using a CPU (Intel Core i7-9750H; Memory: 16 GB). Additionally, the study was carried out on Visual Studio 2022 Professional, based on ML.Net developed by Microsoft [17]:
ML.Net is a machine learning framework developed by Microsoft for the new “.Net” platform and provides a low-code development tool called “Model Builder”, an intuitive graphical Visual Studio extension for generating, training and deploying custom machine learning models [18]. Therefore, for “.Net” platform developers, using the ML.Net machine learning framework is an excellent choice in terms of ease of use, performance and accuracy [19]. The ML.Net machine learning framework uses a DNN (Deep Neural Network) and Resnet-50 model (DNN + Resnet-50) to implement image classification functions so the study was based on DNN and the ResNet-50 model to categorise two types of tailings ponds with different features:
ResNet-50 is a residual network that uses a shortcut connection to connect the inputs directly to the outputs (as shown in Figure 3A), which effectively solves the problem of performance degradation due to the deepening of the network as the shortcut connection does not increase the amount of computation [20].
In essence, the idea of residual network learning can be understood as a block, which can be defined by Equation (1) [21], where y represents the output, F ( x ,   { W i } ) represents the residual component and x represents the sample:
y = F ( x ,   { W i } ) + x .
As a result, the ResNet-50 residual network is well suited for feature extraction of the data sets [22]. Additionally, regarding structure, the ResNet-50 network is divided into six parts, of which Stage 1 is the input module, consisting of Conv and Max Pool, Stage 2 to Stage 5 are the residual modules, containing both Conv Block and Identity Block, and Stage 6 is the output module [22,23]. The structure of ResNet-50 is shown in Figure 3B.

2.2. Training Set and Test Set

The data chosen for the study were satellite images of tailings ponds within China from Geovis (http://www.geovis.com.cn/ (accessed on 10 November 2022) and Tianditu (https://www.tianditu.gov.cn/ (accessed on 10 November 2022), with a total of 30 sets of both the 1# tailings pond (15 sets) and the 2# tailings pond (15 sets). The two different categories of data in the training set have their own distinctive features: the data in the category 1# Tailings Pond are all unclosed tailings ponds, with significant surface water leaching on the satellite images; the data in the category 2# Tailings Pond are generally closed (or almost closed) tailings ponds or dry storage tailings ponds, with no significant surface water leaching on the satellite images and signs of land reclamation. For data set details, please refer to http://dx.doi.org/10.13140/RG.2.2.26494.87367 (accessed on 10 November 2022).

2.3. Validation Methods

To further validate the accuracy of the image recognition and classification function of the ML.NET machine learning framework [24,25], the cross-validation method was chosen to randomly disrupt the data from training sets and the test sets, re-train the new training sets with DDN + ResNet-50 machine learning framework, test with the new test sets and repeat another 19 times (total 20 times) to find the mean value of the accuracy as an estimate of the accuracy [26]. The entire study process is shown in Figure 4.
After training, the accuracy was tested with the test sets in the intuitive graphical Visual Studio extension module of ML.Net. If the model determines that a satellite image of a tailings pond has a greater than 50% probability of belonging to its original category, then the model is considered to have correctly identified and categorised the tailings pond for this time (as shown in Figure 5).

3. Results and Discussion

3.1. Test Accuracy

According to Figure 4, each dataset of the 1# Tailings Pond and 2# Tailings Pond was randomly divided into a training set (10 sets of data) and a test set (5 sets of data), respectively, and each training set was trained by the built DDN + ResNet-50 machine learning model. The model was then tested through the intuitive graphical Visual Studio extension module of ML.Net using the test set according to Figure 5. The whole process was repeated a total of 20 times.
After 20 times cross-validation, the DDN + ResNet-50 network model was found to perform well for the identification and classification of satellite images of tailings ponds, with an average test accuracy of 83.5%: 84% for the 1# Tailings Pond and 83% for the 2# Tailings Pond. The test accuracy data for the 20 times cross-validation are shown in Figure 6.

3.2. Analysis

The results show that the test accuracy of identification and classification of satellite images of tailings ponds based on the DDN + ResNet-50 machine learning model can reach 83.5%; however, in the cross-validation, the identification accuracy of test sets under different training sets has a relatively large difference. For example, as shown in Figure 6: in the 1st, 2nd, 7th, 10th, 13th, 14th, 16th and 17th time of the cross-validation, the identification accuracy of both categories reached 100%; however, in the 4th, 8th, 11th and 20th time of the cross-validation, the identification accuracy for both categories was lower, with a low identification rate of 40% for the 1# Tailings Pond and a low identification rate of 60% for the 2# Tailings Pond.
This may occur because of the presence of data with insignificant features in the dataset, resulting in insufficient generalisation of the model [27]. For example, in Figure 7, Tailings Pond A below has no significant surface water leaching compared to Tailings Pond B, although it belongs to the category of the 1# Tailings Pond. However, cross-validation solved this problem well; as the number of cross-validation times increased, the test accuracy reached closer to the true value.
Therefore, if further validation and improvement of the accuracy of machine learning models are required, the following measures are worth considering.
  • Using the cross-validation method, the total data set is split and combined into different training and testing sets, with the training set being used to train the model and the testing set being used to evaluate how well the model identifies and categorises, which further reflects the accuracy of the model [28]. S-fold cross-validation is a common form of cross-validation in which the total data set is randomly divided into S mutually exclusive subsets of equal size, and each time S-1 copies are randomly selected as the training set and the remaining 1 copy as the test set [29]. When the round is completed, S-1 copies are randomly selected again to train the data [30].
  • Expanding the dataset to allow the model to be more aware of the features of the data in the training set can improve the accuracy of the model. Among the ways to expand the dataset may be finding more relevant data, as well as data augmentation [31,32].

3.3. Optimisation

In order to further validate the accuracy of the image recognition and classification functions of the ML.NET machine learning framework and to optimise the original method of cross-validation, in this part, the three-fold cross-validation method was chosen to be used by randomly dividing the total data set into three equally sized sets, randomly selecting two each time as the training set and the remaining one as the test set, and the cycle was repeated three times to determine the accuracy mean value as the accuracy estimate. The three-fold cross-validation method was also repeated three times by randomly disrupting the data inside the A/B/C/D/E/F sets three times, as shown in Figure 8.
After three times three-fold cross-validation, the DDN + ResNet-50 network model was still found to perform well for the identification and classification of satellite images of tailings ponds, with an average test accuracy of 87.8%: 88.9% for the 1# Tailings Pond and 86.7% for the 2# Tailings Pond. The test accuracy data for the three times three-fold cross-validation are shown in Figure 9.
We then explored further and improved the accuracy of the ML.NET machine learning framework and its DNN + Resnet-50 model for the identification and classification of tailings ponds by expanding the dataset (training set and test set). The data for the new dataset were satellite images of tailings ponds within China, Australia and Malaysia from Geovis (http://www.geovis.com.cn/ (accessed on 10 November 2022), Tianditu (https://www.tianditu.gov.cn/ (accessed on 10 November 2022) and Google Earth (https://earth.google.com/ (accessed on 10 November 2022), with a total of 42 sets of both the 1# tailings pond (21 sets) and 2# tailings pond (21 sets). For data set details, please refer to http://dx.doi.org/10.13140/RG.2.2.27124.01928 (accessed on 10 November 2022).
The new dataset likewise underwent three times three-fold cross-validation (as shown in Figure 8). The DDN + ResNet-50 network model was still found to perform well for the identification and classification of satellite images of tailings ponds, with an average test accuracy of 87.3%: 90.5% for the 1# Tailings Pond and 84.1% for the 2# Tailings Pond. The test accuracy data for the three times three-fold cross-validation are shown in Figure 10.
The results showed that the ML.Net machine learning framework and its DDN + ResNet-50 machine learning model performed very well in the recognition and classification of satellite images of tailings ponds, with accuracies above 80% for all three validations (including the validation after expanding the database). However, the identification accuracy for the 2# Tailings Ponds was slightly lower than that for the 1# Tailings Ponds in all three validations. This may be due to the fact that the 2# Tailings Pond is not well characterised, which is exactly the case: some 2# Tailings Ponds, which are about to be closed or have just been closed, are not very different from 1# Tailings Ponds; while some 2# Tailings Ponds, which has been closed for some time, generally already show signs of extensive land reclamation, which are all different. This problem may need to be solved in the future by other methods, but there is no doubt that ML.Net has done an excellent job of identifying and classifying tailings ponds.

4. Discussion: Research Implications and Other Types of Mine Pollution

The monitoring and management of tailings ponds are particularly important in order to avoid environmental pollution and safety accidents in tailings ponds. However, monitoring tailings ponds is often very time-consuming and labour-intensive. This paper explored the accuracy of the ML.Net machine learning framework and its machine learning model in classifying and identifying the two types of tailings ponds with different characteristics, providing a starting point for future remote sensing techniques to monitor tailings pond risk and pollution [33,34].
It is also important to introduce the public to the severity of the current worldwide mine pollution and its hazards to the environment and public health because, in addition to mine solid waste pollution, mine wastewater and mine dust are also serious threats to the environment and the health of residents [35,36]: Mine wastewater pollution causes serious environmental problems (e.g., heavy metal pollution) to rivers, agricultural soils, the surrounding environment and drinking water for people living nearby; mine dust pollution can affect the safety of mining production and can also have a negative impact on the health of miners, for example by causing them to suffer from occupational diseases such as pneumoconiosis [37].
With the introduction of “Sustainable Development” [38] and “One Health” [39], issues related to mine pollution, environmental damage and public health are receiving increasingly widespread attention worldwide that more and more people are becoming aware of the negative health effects of mine pollution and they are trying to take precautions, while researchers are noticing the ecological and public health risks posed by mine pollution, so more and more research related to mine pollution, environmental damage and public health is being carried out. Additionally, many experts in the field of environmental engineering and public health have proposed measures based on their research expertise to address the problems associated with environmental pollution and health crises in mining; they have mostly focused their research on their own single area of study. For example, Li, S. et al. [40,41] have been working on Green Mine Construction and the elimination of mine pollution, but their research has remained focused on improving mining methods and thus mitigating mine pollution, without taking into account emissions pollution and the impact of emissions on public health; furthermore, Sahu, K et al. [42,43] were among the first researchers to propose the reuse of metal mine solid waste for metal resource recovery as well as to mitigate mine solid waste pollution, but their research was limited to chemical recovery processes, and no further research or discussion of mine pollution or public health threats was undertaken [44,45].
Consequently, in the discussion, we propose the concept of a “Healthy Mine” to provide a direction for development and solutions to the mine pollution and public health crises for mining companies to follow and to raise public awareness of mine pollution.
We define a “Healthy Mine” as a mine that actively addresses and mitigates the impact of mine environmental pollution from the mine discharge (water, solid and dust) on the ecological environment, residents’ health and the occupational health of miners through company management, pollution treatment technologies and employee education in the process of resource development. We advocate all existing mines in the world today should be moving in this direction so that environmental pollution problems and public health crises can be well alleviated.
By definition, a mine is considered a “Healthy Mine” if it meets the following basic conditions: (A) Wastewater and Leachate Treatment: the wastewater and leachate generating from mine solid waste should be treated, so the mine should actively introduce wastewater treatment technology, and the quality of discharged wastewater and leachate should meet the emission standard; (B) “Healthcare”: the mine should ensure that the surrounding population is not affected by pollution from the mine wastewater pollution and the leachate pollution; (C) Solid Waste Management: the mine should have strict management of solid waste discharge sites, and actively implement the land reclamation; (D) Solid Waste Recycle: the mine should be active in the reuse of mine solids, for example in the preparation of construction (or backfill) materials; (E) Dust Control: the mine should actively introduce dust control measure, such as spraying covering agents on the surface of dusty materials; (F) Company Management and Employee Education: the mine should make regulations to strictly manage pollution and discharge control during the mining process, and also strengthen health education for mine employees, for example by strictly requiring them to wear dust filtering masks during mining operations. Thus, the concept diagram of the “Healthy Mine” is as follows in Figure 11:

5. Conclusions

As the construction of tailings ponds is a potential environmental and safety hazard, their monitoring is very necessary. Against this background, this paper proposes to classify tailings ponds into two categories according to whether they are potentially risky or generally safe and to classify tailings ponds’ remote sensing satellite images using DDN + ResNet-50 machine learning model based on ML.Net developed by Microsoft. Meanwhile, this paper also explored the accuracy of the ML.Net machine learning framework and its machine learning model in classifying tailings pond types according to the different characteristics of the 1# Tailings Pond and 2# Tailings Pond.
The conclusions we have drawn are as follows:
  • ResNet-50 is a residual network that uses a shortcut connection to connect the inputs directly to the outputs. Its classification is more accurate, solves the problem of deep network degradation and is well suited to studying the identification and classification of tailings ponds’ satellite images.
  • DDN + ResNet-50 was found to perform well in the identification and classification of satellite images of tailings ponds. The ML.Net machine learning framework and its model achieved an accuracy of 83.5% for the identification and classification of tailings ponds in the case of 20 times cross-validation, achieved an accuracy of 87.8% for the identification and classification of tailings ponds in the case of three-fold cross-validation and achieved an accuracy of 87.3% for the identification and classification of tailings ponds in the case of three-fold cross-validation after expanding the dataset.
  • In this study, the identification accuracy of the 2# Tailings Ponds was slightly lower than that of the 1# Tailings Ponds. This may be due to the fact that the characteristics of 2# Tailings Ponds are not obvious on the satellite maps: some 2# Tailings Ponds that are about to be closed or have just been closed do not differ much from 1# Tailings Ponds on the satellite maps, while some 2# Tailings Ponds that have been closed for some time generally already show signs of extensive land reclamation on the satellite maps, which are different from each other.
In a nutshell, we claim that this research serves as a guide to starting a conversation, and we hope more and more experts, researchers and scholars will be interested and engage in research in this field of mine pollution assessment using remote sensing technologies and machine learning models.

Author Contributions

Conceptualisation: H.Y.; methodology: H.Y.; writing—original draft preparation: H.Y.; writing—review and editing: H.Y. and I.Z.; supervision: I.Z.; project administration: I.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

Data will be made available on request.

Acknowledgments

First, the authors would like to thank the Graduate Research Excellence Scholarship (GRES) from Monash University. Second, the authors express their gratitude to the experts in the research field of mine pollution and public health for their effort.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Small, C.C.; Cho, S.; Hashisho, Z.; Ulrich, A.C. Emissions from oil sands tailings ponds: Review of tailings pond parameters and emission estimates. J. Pet. Sci. Eng. 2015, 127, 490–501. [Google Scholar] [CrossRef]
  2. Liu, J.; Liu, R.; Zhang, Z.; Cai, Y.; Zhang, L. A Bayesian Network-based risk dynamic simulation model for accidental water pollution discharge of mine tailings ponds at watershed-scale. J. Environ. Manag. 2019, 246, 821–831. [Google Scholar] [CrossRef]
  3. Agrawal, A.; Sahu, K.K.; Pandey, B.D. Solid waste management in non-ferrous industries in India. Resour. Conserv. Recycl. 2004, 42, 99–120. [Google Scholar] [CrossRef]
  4. Shengo, L.M. Review of practices in the managements of mineral wastes: The case of waste rocks and mine tailings. Water Air Soil Pollut. 2021, 232, 273. [Google Scholar] [CrossRef]
  5. Liu, Y.; Guo, D.; Dong, L.; Xu, Y.; Liu, J. Pollution status and environmental sound management (ESM) trends on typical general industrial solid waste. Procedia Environ. Sci. 2016, 31, 615–620. [Google Scholar] [CrossRef] [Green Version]
  6. Asif, Z.; Chen, Z. Environmental management in North American mining sector. Environ. Sci. Pollut. Res. 2016, 23, 167–179. [Google Scholar] [CrossRef] [PubMed]
  7. Porsani, J.L.; Jesus, F.A.N.D.; Stangari, M.C. GPR survey on an iron mining area after the collapse of the tailings dam I at the Córrego do Feijão mine in Brumadinho-MG, Brazil. Remote Sens. 2019, 11, 860. [Google Scholar] [CrossRef] [Green Version]
  8. Shen, L.; Luo, S.; Zeng, X.; Wang, H. Review on anti-seepage technology development of tailings pond in China. Procedia Eng. 2011, 26, 1803–1809. [Google Scholar] [CrossRef] [Green Version]
  9. Wei, Z.; Yin, G.; Wang, J.G.; Wan, L.; Li, G. Design, construction and management of tailings storage facilities for surface disposal in China: Case studies of failures. Waste Manag. Res. 2013, 31, 106–112. [Google Scholar] [CrossRef]
  10. Wang, T.; Zhou, Y.; Lv, Q.; Zhu, Y.; Jiang, C. A safety assessment of the new Xiangyun phosphogypsum tailings pond. Miner. Eng. 2011, 24, 1084–1090. [Google Scholar] [CrossRef]
  11. Fennell, J.; Arciszewski, T.J. Current knowledge of seepage from oil sands tailings ponds and its environmental influence in northeastern Alberta. Sci. Total Environ. 2019, 686, 968–985. [Google Scholar] [CrossRef] [PubMed]
  12. Che, D.; Liang, A.; Li, X.; Ma, B. Remote sensing assessment of safety risk of iron tailings pond based on runoff coefficient. Sensors 2018, 18, 4373. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  13. Zhang, L.; Huang, Y.; Wu, X.; Skibniewski, M.J. Risk-based estimate for operational safety in complex projects under uncertainty. Appl. Soft Comput. 2017, 54, 108–120. [Google Scholar] [CrossRef]
  14. Lyu, J.; Hu, Y.; Ren, S.; Yao, Y.; Ding, D.; Guan, Q.; Tao, L. Extracting the tailings ponds from high spatial resolution remote sensing images by integrating a deep learning-based model. Remote Sens. 2021, 13, 743. [Google Scholar] [CrossRef]
  15. Yan, D.; Zhang, H.; Li, G.; Li, X.; Lei, H.; Lu, K.; Zhang, L.; Zhu, F. Improved Method to Detect the Tailings Ponds from Multispectral Remote Sensing Images Based on Faster R-CNN and Transfer Learning. Remote Sens. 2021, 14, 103. [Google Scholar] [CrossRef]
  16. Ahmed, Z.; Amizadeh, S.; Bilenko, M.; Carr, R.; Chin, W.S.; Dekel, Y.; Dupre, X.; Eksarevskiy, V.; Filipi, S.; Finley, T.; et al. Machine learning at Microsoft with ML. NET. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, Anchorage, AK, USA, 4–8 August 2019; Association for Computing Machinery: New York, NY, USA, 2019; pp. 2448–2458. [Google Scholar]
  17. Capellman, J. Hands-On Machine Learning with ML. NET: Getting Started with Microsoft ML; NET to implement popular machine learning algorithms in C; Packt Publishing Ltd.: Birmingham, UK, 2020. [Google Scholar]
  18. Magdin, M.; Benc, J.; Koprda, Š.; Balogh, Z.; Tuček, D. Comparison of Multilayer Neural Network Models in Terms of Success of Classifications Based on EmguCV, ML. NET and Tensorflow. Net. Appl. Sci. 2022, 12, 3730. [Google Scholar] [CrossRef]
  19. Alexan, A.; Alexan, A.; Stefan, O. Soc based iot sensor network hub for activity recognition using ml. net framework. In 2020 IEEE 26th International Symposium for Design and Technology in Electronic Packaging (SIITME); IEEE: Pitesti, Romania, 2020; pp. 184–187. [Google Scholar]
  20. Wen, L.; Li, X.; Gao, L. A transfer convolutional neural network for fault diagnosis based on ResNet-50. Neural Comput. Appl. 2020, 32, 6111–6124. [Google Scholar] [CrossRef]
  21. Wu, Z.; Nagarajan, T.; Kumar, A.; Rennie, S.; Davis, L.S.; Grauman, K.; Feris, R. Blockdrop: Dynamic inference paths in residual networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, 18–23 June 2018; pp. 8817–8826. [Google Scholar]
  22. Bin, L.; Lima, D. Facial expression recognition via ResNet-50. Int. J. Cogn. Comput. Eng. 2021, 2, 57–64. [Google Scholar]
  23. Yu, H.; Zhao, C.; Li, S.; Wang, Z.; Zhang, Y. Pre-Work for the Birth of Driver-Less Scraper (LHD) in the Underground Mine: The Path Tracking Control Based on an LQR Controller and Algorithms Comparison. Sensors 2021, 21, 7839. [Google Scholar] [CrossRef]
  24. Li, X.X.; Li, D.; Ren, W.X.; Zhang, J.S. Loosening Identification of Multi-Bolt Connections Based on Wavelet Transform and ResNet-50 Convolutional Neural Network. Sensors 2020, 22, 6825. [Google Scholar] [CrossRef]
  25. Alexan, A.; Alexan, A.; Oniga, Ș. Smartwatch activity recognition feature comparison using ML. net. In Proceedings of the 2022 IEEE International Conference on Automation, Quality and Testing, Robotics (AQTR), Cluj-Napoca, Romania, 19–21 May 2022; pp. 1–6. [Google Scholar]
  26. Rajpal, S.; Lakhyani, N.; Singh, A.K.; Kohli, R.; Kumar, N. Using handpicked features in conjunction with ResNet-50 for improved detection of COVID-19 from chest X.-ray images. Chaos Solitons Fractals 2021, 145, 110749. [Google Scholar] [CrossRef] [PubMed]
  27. Du, J.; Chen, Q.; Peng, Y.; Xiang, Y.; Tao, C.; Lu, Z. ML-Net: Multi-label classification of biomedical texts with deep neural networks. J. Am. Med. Inform. Assoc. 2019, 26, 1279–1285. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  28. Ramezan, C.A.; Warner, T.A.; Maxwell, A.E. Evaluation of sampling and cross-validation tuning strategies for regional-scale machine learning classification. Remote Sens. 2019, 11, 185. [Google Scholar] [CrossRef] [Green Version]
  29. Ping, X.; Yang, F.; Zhang, H.; Zhang, J.; Zhang, W.; Song, G. Introducing machine learning and hybrid algorithm for prediction and optimization of multistage centrifugal pump in an ORC system. Energy 2021, 222, 120007. [Google Scholar] [CrossRef]
  30. Vu, H.L.; Ng, K.T.W.; Richter, A.; An, C. Analysis of input set characteristics and variances on k-fold cross validation for a Recurrent Neural Network model on waste disposal rate estimation. J. Environ. Manag. 2022, 311, 114869. [Google Scholar] [CrossRef]
  31. Shorten, C.; Khoshgoftaar, T.M. A survey on image data augmentation for deep learning. J. Big Data 2019, 6, 1–48. [Google Scholar] [CrossRef]
  32. Zhong, Z.; Zheng, L.; Kang, G.; Li, S.; Yang, Y. Random erasing data augmentation. In Proceedings of the AAAI Conference on Artificial Intelligence, New York, NY, USA, 7–12 February 2020; Volume 34, pp. 13001–13008. [Google Scholar]
  33. Ghanizadeh, A.R.; Delaram, A.; Fakharian, P.; Armaghani, D.J. Developing Predictive Models of Collapse Settlement and Coefficient of Stress Release of Sandy-Gravel Soil via Evolutionary Polynomial Regression. Appl. Sci. 2022, 12, 9986. [Google Scholar] [CrossRef]
  34. Wang, Y.; Zhao, Y.; Xu, S. Application of VNIR and machine learning technologies to predict heavy metals in soil and pollution indices in mining areas. J. Soils Sediments 2022, 22, 2777–2791. [Google Scholar] [CrossRef]
  35. Skentou, A.D.; Bardhan, A.; Mamou, A.; Lemonis, M.E.; Kumar, G.; Samui, P.; Armaghani, D.J.; Asteris, P.G. Closed-Form Equation for Estimating Unconfined Compressive Strength of Granite from Three Non-destructive Tests Using Soft Computing Models. Rock Mech. Rock Eng. 2022, 1–28. [Google Scholar] [CrossRef]
  36. Indraratna, B.; Armaghani, D.J.; Correia, A.G.; Hunt, H.; Ngo, T. Prediction of resilient modulus of ballast under cyclic loading using machine learning techniques. Transp. Geotech. 2022, 38, 100895. [Google Scholar] [CrossRef]
  37. Cavaleri, L.; Barkhordari, M.S.; Repapis, C.C.; Armaghani, D.J.; Ulrikh, D.V.; Asteris, P.G. Convolution-based ensemble learning algorithms to estimate the bond strength of the corroded reinforced concrete. Constr. Build. Mater. 2022, 359, 129504. [Google Scholar] [CrossRef]
  38. Pindór, T.; Preisner, L. Coal Sector Restructuring due to Sustainable Development. People 2000, 1990, 2005–2006. [Google Scholar]
  39. Frank, D. One world, one health, one medicine. Can. Vet. J. 2008, 49, 1063. Available online: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2572090/ (accessed on 10 November 2022). [PubMed]
  40. Li, S.; Yu, L.; Jiang, W.; Yu, H.; Wang, X. The Recent Progress China Has Made in Green Mine Construction, Part I: Mining Groundwater Pollution and Sustainable Mining. Int. J. Environ. Res. Public Health 2022, 19, 5673. [Google Scholar] [CrossRef]
  41. Yu, H.; Li, S.; Yu, L.; Wang, X. The Recent Progress China Has Made in Green Mine Construction, Part II: Typical Examples of Green Mines. Int. J. Environ. Res. Public Health 2022, 19, 8166. [Google Scholar] [CrossRef]
  42. Agrawal, A.; Kumari, S.; Sahu, K.K. Iron and copper recovery/removal from industrial wastes: A review. Ind. Eng. Chem. Res. 2009, 48, 6145–6161. [Google Scholar] [CrossRef]
  43. Agrawal, A.; Sahu, K.K. Problems, prospects and current trends of copper recycling in India: An overview. Resour. Conserv. Recycl. 2010, 54, 401–416. [Google Scholar] [CrossRef]
  44. Haoxuan, Y.; Zahidi, I. Environmental hazards posed by mine dust, and monitoring method of mine dust pollution using remote sensing technologies: An overview. Sci. Total Environ. 2022, 161135. [Google Scholar] [CrossRef]
  45. Haoxuan, Y.; Zahidi, I. Spatial and temporal variation of vegetation cover in the main mining area of Qibaoshan Town, China: Potential impacts from mining damage, solid waste discharge and land reclamation. Sci. Total Environ. 2023, 859, 160392. [Google Scholar]
Figure 1. Literature results change of keywords “tailings ponds and safety” and “tailings ponds and environment”.
Figure 1. Literature results change of keywords “tailings ponds and safety” and “tailings ponds and environment”.
Mathematics 11 00517 g001
Figure 2. Examples: (A) 1# Tailings Pond; (B) 2# Tailings Pond. Satellite images from tianditu.gov.cn (accessed on 10 November 2022).
Figure 2. Examples: (A) 1# Tailings Pond; (B) 2# Tailings Pond. Satellite images from tianditu.gov.cn (accessed on 10 November 2022).
Mathematics 11 00517 g002
Figure 3. (A) The structure of the shortcut connection; (B) the structure of the ResNet-50 network.
Figure 3. (A) The structure of the shortcut connection; (B) the structure of the ResNet-50 network.
Mathematics 11 00517 g003
Figure 4. The cross-validation of the machine learning model.
Figure 4. The cross-validation of the machine learning model.
Mathematics 11 00517 g004
Figure 5. The intuitive graphical Visual Studio extension module of ML.Net. Satellite images from tianditu.gov.cn (accessed on 10 November 2022).
Figure 5. The intuitive graphical Visual Studio extension module of ML.Net. Satellite images from tianditu.gov.cn (accessed on 10 November 2022).
Mathematics 11 00517 g005
Figure 6. The results of the cross-validation.
Figure 6. The results of the cross-validation.
Mathematics 11 00517 g006
Figure 7. Tailings ponds in the datset. (A): Source from tianditu.gov.cn; (B) source from geovis.com.cn (accessed on 10 November 2022).
Figure 7. Tailings ponds in the datset. (A): Source from tianditu.gov.cn; (B) source from geovis.com.cn (accessed on 10 November 2022).
Mathematics 11 00517 g007
Figure 8. The 3-fold cross-validation schematic diagram.
Figure 8. The 3-fold cross-validation schematic diagram.
Mathematics 11 00517 g008
Figure 9. The results of the 3-fold cross-validation.
Figure 9. The results of the 3-fold cross-validation.
Mathematics 11 00517 g009
Figure 10. The results of the 3-fold cross-validation for the expanded dataset.
Figure 10. The results of the 3-fold cross-validation for the expanded dataset.
Mathematics 11 00517 g010
Figure 11. The concept diagram of the “Healthy Mine”.
Figure 11. The concept diagram of the “Healthy Mine”.
Mathematics 11 00517 g011
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Yu, H.; Zahidi, I. Tailings Pond Classification Based on Satellite Images and Machine Learning: An Exploration of Microsoft ML.Net. Mathematics 2023, 11, 517. https://doi.org/10.3390/math11030517

AMA Style

Yu H, Zahidi I. Tailings Pond Classification Based on Satellite Images and Machine Learning: An Exploration of Microsoft ML.Net. Mathematics. 2023; 11(3):517. https://doi.org/10.3390/math11030517

Chicago/Turabian Style

Yu, Haoxuan, and Izni Zahidi. 2023. "Tailings Pond Classification Based on Satellite Images and Machine Learning: An Exploration of Microsoft ML.Net" Mathematics 11, no. 3: 517. https://doi.org/10.3390/math11030517

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop