We have developed an improved adaptive iterative method based on the nonstationary iterative Tikhonov regularization method for performing a downward continuation of the potential-field data from a horizontal plane. Our method uses the Tikhonov regularization result as initial value and has an incremental geometric choice of the regularization parameter. We compared our method with previous methods (Tikhonov regularization, Landweber iteration, and integral-iteration method). The downward-continuation performance of these methods in spatial and wavenumber domains were compared with the aspects of their iterative schemes, filter functions, and downward-continuation operators. Applications to synthetic gravity and real aeromagnetic data showed that our iterative method yields a better downward continuation of the data than other methods. Our method shows fast computation times and a stable convergence. In addition, the L-curve criterion for choosing the regularization parameter is expressed here in the wavenumber domain and used to speed up computations and to adapt the wavenumber-domain iterative method.

Geomagnetic vector field is always measured by a strapdown triaxial magnetometer on a vehicle. The magnetometer measurements are polluted by the magnetometer inherent errors, misalignment error, and vehicle-generated magnetic inferences. With the aid of absolute attitude information from the gyro, the calibration accuracy has been proved to be greatly improved, especially for vector measurements. However, the performance of calibration methods is greatly affected by the gyro bias, which causes error accumulated over time in the absolute attitude information. In this work, we propose an improved calibration method robust to gyro bias based on relative attitude information from the gyro. Simulation results indicated that the proposed method can estimate the calibration parameters more accurately and stably than the traditional method when the gyro bias exists. The influence on the calibration accuracy of the different parts of the gyro bias is also analyzed by simulation. Experiment using a low-cost gyro proved that the proposed method can effectively improve the estimation accuracy of the calibration parameters, as the influence on the traditional method of the gyro bias is 2-3 orders of magnitude higher than that on the proposed method.

SUMMARY: The reliable classification of seismic events is crucial to precise seismic cataloging and robust hazard evaluation. Recent advances in deep learning have achieved great success in seismic event identification, leveraging their exceptional ability to automatically extract and recognize features. However, existing deep learning approaches to seismic classification rely exclusively on deterministic models, which cannot quantify epistemic uncertainty, preventing the estimation of prediction confidence that is critical for reliability evaluation. To address this issue, in this paper, we develop two uncertainty-aware deep learning models for earthquake (EQ) versus explosion (EP) classification using the DiTing 2.0 artificial intelligence training data set: a Bayesian convolutional neural network (BCNN) and a dropout-based CNN (DropCNN). We also implement a conventional deterministic CNN as a baseline model for comparative analysis. The experimental results demonstrate that both the BCNN and DropCNN can achieve a classification accuracy comparable to the conventional CNN, while providing additional uncertainty metrics for estimation confidence of prediction. Crucially, their uncertainty scores increase markedly in terms of encountering misclassifications or out-of-distribution samples compared to correct classifications, enabling automatic rejection of unreliable predictions based on the uncertainty threshold setting, triggering human verification or alternative discrimination methods. We then apply the trained models to analyse suspicious explosion events in the DiTing 2.0 data set. The BCNN and DropCNN results exhibits strong agreement, consistently identifying 79 EP and two EQ events and flagging the remaining samples as uncertain classifications needing further verification. Our findings demonstrate that deep learning methods incorporating uncertainty estimation not only maintain a high accuracy in seismic event discrimination but also provide uncertainty estimation. This capability significantly enhances the model’s reliability and decision-making value in practical applications.

Infrasound signal classification represents a critical challenge that demands immediate attention. Feature extraction stands as the core concept for enhancing classification accuracy in infrasound signal processing. However, existing feature extraction methodologies fail to meet the requirements for long-distance detection scenarios. To address these limitations, this study proposes a novel classification framework based on the spatiotemporal characteristics of infrasound signals. The proposed framework incorporates advanced signal processing techniques, signal enhancement algorithms, and deep learning architectures to achieve precise classification of infrasound signals. This paper designs three sets of comparative experiments, and the results demonstrate that the proposed method achieves a classification accuracy rate of 83.9% on chemical explosion and seismic infrasound datasets, outperforming eight other comparative classification methods. This substantiates the efficacy of the proposed approach.