Heavy metal pollution poses a significant threat to aquatic ecosystems and drinking water sources, necessitating comprehensive environmental assessments. This study investigated the spatial distribution, potential ecological risks, and sources of heavy metals in the surface sediments of Wanfeng Lake to inform effective pollution management strategies. Twelve sediment samples were collected and analyzed for eight heavy metals (Hg, As, Cu, Pb, Cd, Cr, Ni, Zn) using inductively coupled plasma mass spectrometry. The geo-accumulation index (I_geo) and potential ecological risk index (RI) were applied to assess contamination levels and ecological risks. Cluster analysis and Kriging interpolation were used to identify potential pollution sources and spatial patterns. Results revealed that heavy metal concentrations decreased in the order Zn > Pb > Cu > Cr > Ni > As > Cd > Hg, with Hg concentrations exceeding the national average for Chinese lake sediments. Ecological risk assessments identified Hg (0.06 μg/g) and Cd (0.10 μg/g) as the priority pollutants. The Hg posed a moderate ecological risk, particularly in upstream areas, while Cd pollution was concentrated downstream. Atmospheric deposition was identified as the primary source of Hg, whereas agricultural activities were determined to be the main driver of Cd contamination. These findings provide a scientific basis for developing targeted pollution control measures in Wanfeng Lake.

This paper provides a unified view to explain different adversarial attacks and defense methods, \emph{i.e.} the view of multi-order interactions between input variables of DNNs. Based on the multi-order interaction, we discover that adversarial attacks mainly affect high-order interactions to fool the DNN. Furthermore, we find that the robustness of adversarially trained DNNs comes from category-specific low-order interactions. Our findings provide a potential method to unify adversarial perturbations and robustness, which can explain the existing defense methods in a principle way. Besides, our findings also make a revision of previous inaccurate understanding of the shape bias of adversarially learned features.