The impact of ecological compensation systems and resource prediction for mineral resource development under deep learning and artificial intelligence

To address the persistent challenges of subjectivity in ecological compensation assessment, delayed system response, and uncertainty in resource reserve prediction during mineral resource development, this study introduces the Dynamic Ecological Compensation Prediction Network (DECP-Net). The framework integrates the dynamic evolution of mining-area ecosystems with predictive modeling of mineral resource distribution. Built upon an enhanced U-shaped encoder–decoder architecture and augmented with a spatiotemporal attention mechanism, DECP-Net processes multi-temporal remote sensing imagery from the MineNetCD dataset alongside heterogeneous ancillary data, including geological exploration records and environmental monitoring indicators. A shared encoder extracts hierarchical spatiotemporal representations of surface disturbances, while two task-specific decoders concurrently execute pixel-level segmentation of ecological disturbance zones and quantitative prediction of mineral resource potential. The model ultimately produces a dynamic compensation evaluation map that holistically integrates resource value and ecological degradation. Experimental results demonstrate that DECP-Net delivers consistently high performance across both tasks. For mineral resource potential prediction, the model achieves an accuracy of approximately 92% and an AUC of 0.963, indicating robust discriminative capacity. For ecological disturbance segmentation, the Dice coefficient reaches 0.920, enabling fine-grained delineation of disturbance boundaries. Overall, the proposed framework offers a forward-looking and rigorously quantitative tool for integrated decision-making in mineral development and environmental governance, providing substantial scientific and practical value for advancing sustainable resource management.