Implementation of Deep Convolutional Neural Network to Reconstruct Subsurface Resistivity Using Gramian Angular Field for Airborne Electromagnetic System

The airborne transient electromagnetic (AEM) method offers advantages in deep exploration and complex terrain adaptability but faces challenges in efficient inversion due to large datasets and nonlinearity. We propose a novel data-driven approach integrating Gramian angular field (GAF) transformation and 2-D convolutional neural networks (CNNs) for rapid resistivity imaging. Our method converts TEM signals into two GAF representations (difference/summation fields) and incorporates flight altitude as a constant channel, forming a three-channel input for CNN training. A specialized 2-D-CNN architecture, trained on synthetic 1-D forward modeling data, learns nonlinear relationships between TEM responses and subsurface resistivity without iterative computations. Comparative tests demonstrate superior convergence and accuracy over traditional artificial neural networks, with robustness against noise contamination. Field validation in Alxa League mineral exploration shows close agreement with deterministic Occam inversion results. The framework enables instantaneous resistivity reconstruction after training, eliminating conventional time-consuming optimization processes. Synthetic and field experiments confirm the method's capability to resolve layered structures and conductive anomalies efficiently. This advancement provides a transformative tool for rapid AEM data interpretation, particularly beneficial for large-scale surveys and real-time imaging in challenging environments. The integration of signal encoding via GAF and deep learning establishes a new paradigm for solving nonlinear geophysical inverse problems, significantly enhancing interpretation efficiency while maintaining geological reliability. Our approach addresses critical bottlenecks in AEM applications, enabling high-resolution subsurface characterization with reduced computational overhead.