Application of Concentration-Area fractal modeling and artificial neural network to identify Cu, Zn±Pb geochemical anomalies in Hashtjin area, NW of Iran
Identification of geochemical anomalies plays an essential role in mineral exploration. Recent research investigations have shown that Machine Learning (ML) algorithms can identify geochemical anomalies associated with mineralization that represent targets for mineral exploration. Machine Learning a...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | fas |
| Published: |
Ferdowsi University of Mashhad
2024-10-01
|
| Series: | Journal of Economic Geology |
| Subjects: | |
| Online Access: | https://econg.um.ac.ir/article_45781.html?lang=en |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Identification of geochemical anomalies plays an essential role in mineral exploration. Recent research investigations have shown that Machine Learning (ML) algorithms can identify geochemical anomalies associated with mineralization that represent targets for mineral exploration. Machine Learning algorithms are widely used in various fields due to their strong capability to extract and display high-level features of training samples. Autoencoder networks show a high ability to identify geochemical anomalies. In this study, the combined method of autoencoder network with the fractal concentration-area method was used to identify geochemical anomalies. First, using multivariate factor analysis, the elements barium, lead, zinc, copper, gold, iron, gold and arsenic were selected as indicators. Subsequently, the uni-element geochemical maps of these elements were prepared, and to standardize the maps in terms of minimum and maximum values, all maps were fuzzified and scaled. Using the fuzzy gamma operator, uni-element geochemical maps were combined. Then the resulting map applied the deep autoencoder method with eight layers of encoder and decoder were reconstructed. Finally, a mineral prospectively map was prepared for the potential area using the concentration-area fractal method. The mixed model proposed in this study introduces the region with high mineralization potential northeast of the studied area.
Introduction
Over the past few decades, the identification of geochemical anomalies has played an important role in mineral exploration (Coates et al., 2011; Lecun et al., 2015; Bergen et al., 2019) Various methods have been used to identify geochemical anomalies in the last few decades, including statistical analysis, geostatistical approaches (Nabavi, 1976), fractal modeling (Ziaii et al., 2009; Ziaii et al., 2012) and many other methods. Fractal/multifractal models are powerful tools that have been widely used to detect geochemical anomalies.
Recent developments in machine learning methods have led to significant advances in geoscience. ML-based approaches to mineral prospectivity mapping using geochemical data can more effectively identify statistical correlations between geochemical patterns than other non-ML methods. Recent research shows that machine-learning approaches enable the integration of geochemical data and the successful identification and separation of geochemical anomalies associated with mineralization that may be overlooked using non-machine learning methods (Tukey, 1977; Cheng, 2006; Cheng et al., 2010).
In this study, the combined method of autoencoder network with the fractal concentration-area method was applied to identify geochemical anomalies in the Hashtjin area (Ardabil province), NW Iran.
Materials and methods
The combined method of the autoencoder network with the fractal concentration-area method was used to identify the geochemical anomalies. The flowchart of this study is as follows:
1- First single-element geochemical maps of Pb, Zn, Cu, Au, As, Fe, and Ba elements were prepared, and to incorporate the minimum and maximum values, all of them became fuzzy to be in the range of (0 and 1).
2- Geochemical maps were combined using the fuzzy gamma operator.
3- The deep autoencoder method was implemented on the resulting map, and the reconstructed output was obtained.
4- Using the fractal concentration-area method, the final map of mineral prospectivity maps was prepared.
Discussion
An autoencoder is a type of artificial neural network used for learning efficient encoding of unlabeled data (unsupervised learning). Autoencoder consists of two functions: an encoder function that encodes the input data into a lower-dimensional hidden layer and a decoder function that reconstructs the encoded input data.
The encoder part of an autoencoder network attempts to reduce the dimensionality of the input data while preserving the majority of the information, and it encodes the input data in a blinded space. The decoder part attempts to capture the encoded data and reconstruct the original data with minimal error. The deep autoencoder network is an autoencoder network in which the neural network is designed profoundly, and the number of layers in it is greater. Each layer looks at the data as a new perspective. This neural network automatically and through unsupervised learning identifies patterns, complex structures, and high-level features of the input data.
Considering the evidence of the fractal nature of element distribution, the use of these methods in geochemical exploration for separating anomalous populations from the background with high confidence levels is one of the most potent known methods. The concentration-area fractal method is based on the amount of area occupied by a particular concentration in the study area.
Results
The study area exhibits complex geochemical features. In such a complex setting, it is essential to implement multiple methods in combination to separate anomalies from the background accurately. The combined method introduced in this study is a powerful tool for identifying geochemical anomalies in areas with complex geological history and diverse geochemical backgrounds. According to the results, all three methods used in this study represent a high favorability of ore mineralization in the northeast of the study area; Therefore, further investigations and investigation are recommended in the introduced area during later stages.
The results obtained from applying the fractal method to the output of the deep autoencoder method indicated a potentially favorable area for the mineralization with approximately 9,047,500 m²
In the predictive map obtained in the northeast of the studied area, an area with high mineralization potential is introduced, which is geologically represented by the Upper Jurassic (Lar Formation) and Cretaceous formation controlled by, fault boundaries and in the northeastern part of the Hashtjin.
|
|---|---|
| ISSN: | 2008-7306 |