Mineral prospectivity prediction based on convolutional neural network and ensemble learning.

Area of Science:

• Geoscience
• Artificial Intelligence
• Data Science

Background:

• Deep learning models are crucial for mineral prospectivity prediction but exhibit prediction instability due to varying network structures.
• The correlation between geological big data patterns and ore deposit locations differs across algorithms, impacting prediction reliability.

Purpose of the Study:

• To enhance the stability and accuracy of mineral prospectivity prediction using ensemble learning.
• To synthesize convolutional neural network (CNN) and self-attention mechanism algorithms for improved geological big data analysis.

Main Methods:

• Selected 14 gold mineralization factors, including 10 geochemical and 4 geological data types.
• Utilized six CNN models (MobileNet V2, ResNet 50, VGG 16, AlexNet, LeNet, VIT) for feature extraction.
• Applied ensemble learning to combine model predictions for a final prospectivity map.

Main Results:

• Achieved over 94% accuracy in predicting mineral prospectivity using trained network models.
• Generated a prospectivity prediction map for the Bawanggou mine area, guiding gold exploration.
• Demonstrated the effectiveness of ensemble learning in leveraging diverse model strengths.

Conclusions:

• The proposed ensemble learning method offers a stable and extensible approach for mineral prospectivity prediction.
• This method effectively extracts deep-level mineralization relationships from geological big data.
• Future work can further enhance results by incorporating more mineralization factors and novel algorithms.