Denoising of semi-airborne transient electromagnetic data based on transformer network

Semi-airborne transient electromagnetic exploration is a key component in geophysical exploration, but it often faces challenges due to the interference from multiple noise sources, particularly impacting the quality of late-stage signals. To enhance data quality and interpretative accuracy, this study introduces a data denoising method utilizing deep learning technology. This method leverages the strengths of Transformer networks in handling spatiotemporal features and processing long-duration data, specifically applied to semi-airborne electrical data denoising. Through training the network to recognize the spatiotemporal characteristics of geoelectric field signals, the approach effectively differentiates and suppresses various noise types, thereby boosting the signal-to-noise ratio. The experimental outcomes demonstrate that employing Transformer networks significantly improves data quality and preserves crucial information. This research contributes a cutting-edge deep learning strategy for semi-airborne electrical data processing, offering novel insights and techniques for future data processing in geophysical exploration.