融合AlexNet与FPN的磁异常反演方法研究

Research on Magnetic Anomaly Inversion Method Based on the Integration of AlexNet and FPN

摘要: 针对传统磁异常二维反演方法精度与稳定性不足的问题，本研究提出一种融合深度卷积神经网络(CNN)与特征金字塔网络(FPN)的混合反演方法。通过对AlexNet进行改进，增强其特征提取能力，并与FPN结合构建AlexNet-FPN网络，实现对复杂地下地质结构的精准识别。研究基于多种地质体模型构建合成数据集开展实验，结果显示，在简单模型、复杂构造模型及含噪声干扰条件下，该方法反演效果均优于传统方法，提高了稳定性与抗噪性。实际勘探数据验证进一步证实了方法的适用性。结果表明，该方法具备优异的泛化能力，为不同地质条件下的磁异常反演提供了创新技术路径。

Abstract: Aiming at the problems of insufficient accuracy and stability of traditional two-dimensional magnetic anomaly inversion methods, this study proposes a hybrid inversion method that integrates deep convolutional neural networks (CNN) and feature pyramid networks (FPN). By improving AlexNet to enhance its feature extraction ability and combining it with FPN to construct the AlexNet-FPN network, accurate identification of complex underground geological structures is achieved. The study constructs synthetic datasets based on various geological models and conducts experiments. The results show that under simple models, complex structural models, and conditions with noise interference, the inversion performance of this method is superior to traditional methods, significantly improving stability and noise resistance. Verification using actual exploration data further confirms the applicability of the method. The results indicate that this method has excellent generalization ability, providing an innovative technical approach for magnetic anomaly inversion under different geological conditions.