Artificial Intelligence for Mineral Exploration Imagery: A Systematic Mapping Review of Data, Tasks, and Methods

Abstract

Artificial intelligence (AI) is increasingly applied to image-based and image-formatted geoscientific evidence across mineral exploration, from regional and surface surveys to drilling/core analysis and laboratory characterization. However, the literature remains distributed across different research communities, making it difficult to compare how visual evidence, AI tasks, methodological choices, and geological outputs relate across exploration contexts. This systematic mapping review searched the Web of Science Core Collection, Scopus, and IEEE Xplore for English-language journal articles published between 2016 and 2026. Of 397 identified records, 91 studies were retained after deduplication, screening, and full-text assessment. The studies were mapped across exploration contexts, visual-data domains, AI visual tasks, methodological paradigms, and geological outputs. Regional and surface survey data dominate the evidence base, while drilling/core and laboratory imaging remain less represented. Classification is the most frequent visual task, and CNN-based models remain the dominant methodological paradigm, with hybrid and emerging architectures forming the next major group. The evidence does not indicate a single architecture that is uniformly suitable across mineral-exploration settings; model choice depends on input structure, required geological output, labeled-data availability, and spatial scale. Multisource integration can combine complementary evidence, but differences in spatial support, annotation, sensing conditions, and validation design continue to limit direct comparison and cross-region generalization. Future progress requires more traceable public resources, geographically independent validation, and more systematic integration of complementary geological evidence.