A novel method for the identification of geochemical anomalies with emphasis on addressing the problem of elemental background variation using Pb deposits in Shaoshan, central China, as a case study

The EM-DAE model was established in this study by separating multiple populations and efficiently utilizing deep learning techniques to define geochemical anomalies. Hence, this novel method significantly improves the definition of geochemical anomalies due to the elimination of elemental background variation influence, mainly in the elimination of spurious anomalies in high-background areas and the identification of subtle anomalies in low-background areas. • The EM-DAE method significantly aids in the identification of geochemical anomalies. • This novel A.I.-backed method reduces geochemical background variation effectively. • This A.I.-backed method recognizes subtle anomalies and eliminates spurious ones. Most currently available methods for identifying geochemical anomalies using machine/deep learning algorithms ignore the issue of elemental background variation. This study exemplifies the identification of Pb anomalies in regional stream sediments from Shaoshan, central China, by utilizing a deep autoencoder (DAE). The focus is on applying this algorithm to detect geochemical anomalies in areas with varying geochemical background of elements. Firstly, we grouped the stream sediment samples into seven clusters using the Expectation-Maximization (EM) clustering algorithm, effectively minimizing the influence of elemental background variation. Subsequently, elements associated with Pb mineralization in groups one to seven were determined through robust principal component analysis (RPCA): Bi-Li-Sn-Pb, Li-Pb-MgO, As-Nb-Pb, Nb-Pb-Zn-Al 2 O 3 , Li-Pb-Al 2 O 3 -Fe 2 O 3 , Ag-Pb-CaO, and Ag-Bi-Li-Pb-Sb-SiO 2 . The elemental data for each group were then input into the DAE respectively to calculate the reconstruction error, with a threshold value of 0.24 established to delineate Pb anomalies. The identified anomalies corresponded to the known Pb deposits with an accuracy of 89%. In comparison to the DAE method, the combined approach offers a more effective means of identifying geochemical anomalies. This is primarily evident in its ability to eliminate false anomalies in areas with high background while also detecting weak anomalies in regions with low background. The integration of the EM clustering algorithm with machine/deep learning techniques for anomaly detection can significantly enhance the accuracy of geochemical anomaly identification.