Prospectivity Mapping of Tungsten Mineralization in Southern Jiangxi Province Using Few-Shot Learning

The development of mineral prospectivity mapping (MPM), which aims to outline and prioritize exploration targets, has been spurred by advances in data-driven machine learning algorithms. Supervised MPM is a typical few-shot task, suffering from scarcity labeled data, the over-fitting models an uncertainty predictions. main objective this contribution propose robust framework (FSL), combining data augmentation transfer enable generation with excellent predictive efficiency low uncertainty. systems approach was used conceptual system into mappable criteria. Synthetic minority over-sampling technique (SMOTE) employed augment balance dataset, allowing for model pre-training large synthetic training dataset source domain. knowledge derived pre-trained then transferred target domain fine-tuning, generated light assessments. proposed FSL applied tungsten southern Jiangxi Province. results indicated that SMOTE-ed balanced boosted classification accuracy process. yielded arch-shaped prediction point pattern favorable focusing potential targets high probability achieved performance (test AUC = 0.9172) lowest quantitative value compared benchmark algorithms random forest support vector machine. Four levels targeting zones, considering both uncertainty, were extracted resulting map. final high-potential low-risk only cover 4.27% area, but capture 41.53% known deposits, thus achieving superior performance. This study highlights capability control generate high-confidence