Mineral prediction based on prototype learning

In the field of mineral resource prediction, acquiring labeled data and bearing high annotation costs pose significant challenges. Moreover, distinct characteristics are present in different types of data, including geophysical, geochemical, and geological data. However, conventional deep learning methods often treat all data uniformly, neglecting the specificities inherent in various domains of knowledge. To address these issues, we propose Geo -Meta, an optimized prototype learning model that combines prototype learning, metric learning, and meta -learning strategies to classify multi -source geological samples. Geo -Meta utilizes tailored network architectures for different data types to fuse features and compute prototypes in the metric space. Additionally, it employs a generative model to identify subtle anomaly features and incorporates a semi -supervised algorithm based on label propagation to refine the initial prototypes, enhancing their robustness and category -awareness. Furthermore, Geo -Meta incorporates a dynamic distance metric module to quantify the similarity between sample features and class prototypes, facilitating effective category partitioning. Extensive experimentation demonstrates that our proposed model achieves remarkable performance. It attains an impressive Area Under the Receiver Operator Curve (AUC) of 98% without data augmentation, using a significantly limited number of samples. Moreover, it accurately identifies approximately 93% of ore deposits within only 7% of the study area, surpassing similar models in both accuracy and efficiency. The mineral resource prediction map generated by Geo -Meta aligns with the spatial distribution of mineral resources and adheres to geological knowledge, providing valuable technical support for decision -making in target area exploration. Our proposed model enables the adaptation of deep learning algorithms to geological big data, offering substantial potential for cost and time reduction in mineral prediction. Thus, it holds great relevance and practicality for the mining industry.