MRIPCN: A Cascade Network for Multisource Residual Interference in UAV Aeromagnetic Data Postprocessing

Addressing the issue of significant strip and random interference persisting in aeromagnetic data acquired by multirotor unmanned aerial vehicle (UAVs) during terrain-following flights even after conventional processing, this article proposes a cascaded deep learning network multisource residual interference post-processing cascade network (MRIPCN). The method integrates supervised and self-supervised learning within a two-stage cascade architecture. Stage I incorporates a supervised strip interference suppression module, which utilizes multilayer convolutions to identify the distribution and amplitude of banded interference and employs a U-Net structure to effectively separate such interference from valid signals. Stage II consists of a self-supervised multiscale random interference removal module. By fusing multiview information through a multidimensional attention mechanism, this module achieves deep suppression of residual multiscale random interference while improving model generalization. Synthetic model experiments show that, compared with conventional filtering postprocessing algorithms, MRIPCN can adapt to anomaly processing requirements under different magnetization conditions. After processing, the residual interference exhibits lower energy and a 40% reduction in dispersion, with an interference suppression ratio reaching 97.69%. The method also demonstrates superior performance in preserving features and overall robustness. In postprocessing and interpretation of UAV aeromagnetic survey data from a mining area, the approach effectively suppresses multisource interference with minimal loss of geological field-source signals, significantly enhancing anomaly boundary clarity and interpretation reliability, confirming its robustness and applicability in practical scenarios. This study provides an efficient and reliable solution for multisource interference suppression in aeromagnetic data from small rotor UAVs, substantially improving both data quality and the reliability of geological interpretation.