Identification of regional geochemical anomalies based on VGAE-T

• Integrating Transformer and VGAE enhances geochemical spatial dependency modeling. • The self-supervised VGAE-T model effectively alleviates the scarcity of labeled data. • The model demonstrates strong generalization across regions. The identification of geochemical anomalies is crucial for mineral deposit targeting and metallogenic probability assessment in mineral exploration. This paper proposes a graph self-supervised learning framework named VGAE-T for regional geochemical anomaly identification. The VGAE-T framework aids mineral resource exploration. It first captures global spatial dependencies within geochemical data using a self-attention mechanism. The framework is then divided into unsupervised and supervised learning phases. In the unsupervised phase, a graph structure is built using element concentrations and spatial proximity. The model is optimized using reconstruction loss and KL divergence. In the supervised phase, features from the Transformer, the original adjacency matrix, and labels are fed into a GCN classifier for prediction. The study area is the stratabound Pb-Zn deposit in the Changba ore cluster, Gansu Province, formed during the Indosinian orogeny. Comparative analyses were conducted against the C-N fractal method, GCN, and VGAE-GCN models. The results demonstrate that the VGAE-T model exhibits significant advantages in identifying geochemical anomalies, achieving an AUC of 0.959. The anomalies it delineates show a higher degree of correspondence with the actual distribution of ore deposits. To validate the model’s generalization capability, the trained model was further applied to the northern Mila Mountain area in Tibet for cross-regional prediction. The results also achieved excellent performance. In summary, incorporating global spatial dependencies can effectively enhance the representation of geochemical anomalies. The VGAE-T model constructed in this study contributes to improving the accuracy of geochemical anomaly identification and provides a new perspective for mineral resource exploration.