Magnetic Anomaly Detection Using Multifeature Fusion-Based Neural Network

Magnetic anomaly detection (MAD) has been widely used for detecting the magnetic targets in many areas. In low signal-to-noise ratio (SNR) situations, the magnetic anomaly is usually buried in the magnetic noise, which leads to a decrease in the detection performance of traditional MAD methods. In order to improve the detection probability of MAD in low SNR, we propose an adaptive method of MAD using support vector machine (SVM). First, we design a detection model using SVM with radial basis function (RBF) kernel function. Meanwhile, we analyze the magnetic anomaly signal and determined the orthonormal basis function (OBF) energy and magnetic entropy as the features of the magnetic anomaly, which are as the input of the model. Then, the model is trained based on the data set with the magnetic signals. Finally, the trained model is applied to the raw magnetic signal to detect the anomaly. Compared to the traditional methods, the proposed method has a better performance and adaptive ability. The proposed method is validated by the real magnetic noise. The results show that the proposed method can reduce the case of false alarm and improve the detection performance in low SNR.

Differentiable homotopy methods for gradually reinforcing the training of fully connected neural networks

In the field of magnetic anomaly detection (MAD), the anomaly signal is easily submerged by ambient electromagnetic interference. Though the existing noise suppression methods can effectively improve the signal-to-noise ratio (SNR), there are still some intractable problems, such as signal distortion and boundary blur. To solve these problems, a novel MAD method based on structured low rank (SLR) and total variation (TV) regularization constraints is proposed in this letter. The noise suppression performance is improved by leveraging the structured low rankness of the signal. To preserve clean boundaries of the anomalies, an anisotropic TV regularization constraint is employed in the approach. Comparing the SLR-TV method with four state-of-the-art methods with extensive field tests, the results demonstrate that the proposed SLR-TV method achieves the greatest SNR improvement by about 63.24% and the best structural similarity (SSIM) improvement by about 53.02% over other methods in the range from −40 to 0 dB, showing the utility and high fidelity of the proposed framework in low SNR.

Magnetic anomaly detection is a passive and effective approach for detecting magnetic targets. During detection using one scalar magnetometer, temporal magnetic variations will affect the detection performance. In order to reduce the influence of magnetic variations on detection, we focus on the analysis of magnetic gradient signals and obtain four orthonormal basis functions (OBFs) of gradient signal. Based on the set of OBFs, we propose an improved method using the magnetic gradient signal. Meanwhile, we can also obtain the bearing of the target relative to the survey line in the method. The proposed method is validated by simulation and real-world magnetic noise. The results show that the proposed method has a good detection performance with low signal to noise ratio. Therefore, the proposed method can improve the detection probability of weak magnetic anomalies and also determine the bearing of targets relative to the survey line.

In certain scenarios, such as detection of unexploded ordnance (UXO), submarines, and intruders, magnetic anomaly detection (MAD) is an effective method due to the magnetic field advantages of small operating power, strong penetrability, and strong anti-interference ability. However, variability in the background geomagnetic field and low signal-to-noise ratio (SNR) in MAD cannot be avoided. In this article, we propose a MAD method employing an improved orthogonal basis function (OBF) for the magnetic tensor contraction decomposition. The orthogonal basis decomposition approach makes the best use of modeling the MAD output signal, whereas the magnetic tensor contraction aid in avoiding the background field problem. Furthermore, by increasing the baseline length of the magnetic gradiometer, the SNR can also be improved in some manner. A dual magnetic gradiometer framework is introduced to fulfill the demand of the magnetic anomaly localization. From our simulation results, the proposed method improves the SNR of the MAD raw signal and is capable of detecting and locating the magnetic anomaly. Moreover, field test has been carried out to evaluate the performance of the proposed method.

Electrocardiogram (ECG) signals are significantly distorted during recording by muscle artifact (MA), causing signal frequency overlap and making it difficult to interpret ECG data correctly. Deep learning (DL) methods for signal processing have shown promising results. However, there is a significant necessity in building proper DL models with appropriate datasets. We propose an enhanced hybrid deep learning framework called HRGB-Net based on residual neural network (ResNet), global channel attention block (GCAB), and bidirectional-long-short-term memory (Bi-LSTM) blocks for filtering the MA noise from ECG by using three distinctive MIT-BIH real-time datasets from the PhysioNet repository by creating suitable datasets for training. We use both raw ECG data and short-time Fourier-transformed (STFT) ECG data for comparative analysis with three neural network models: a convolutional neural Network (CNN), a fully connected neural network (FCNN), and a regression-based LSTM (Reg-LSTM-DNN) model to assess the proposed model. The signal-to-noise ratio (SNR) of noisy ECG signals is varied from - 7dB to 2dB to analyze the mean square error (MSE) and correlation coefficient (CC) performances after the denoising process. Our proposed method utilizes the regression ability to remove MA noise and generate a clean ECG signal with improved values of these signal parameters. The STFT-trained and tested ECG data shows better results than the raw ECG data for efficiently eliminating the MA with a 98.82% correlation coefficient and optimal MSE value of 0.053068. The results prove our proposed HRGB-Net model's remarkable ability to outperform the neural network models and other standard techniques.

This paper proposes the prediction model of wind speed and direction using convolutional neural network - long short-term memory (CNN-LSTM). The proposed prediction model combines CNN, LSTM, and fully connected neural networks (FCNN) which are useful for getting high prediction accuracy of wind speed and direction for wind power. Performances of the prediction models are evaluated by using root mean square error (RMSE) between actual measurement data and predicted data. To verify the effectiveness of the proposed prediction model in comparison with that using FCNN, CNN, or LSTM model. The usefulness of the proposed prediction model is evaluated from the improvement of prediction accuracy for each season. The proposed prediction model using CNN-LSTM can improve 27.95 - 42.16% for wind speed and 28.71 - 35.15% for wind direction depending on the season in comparison with using the FCNN that is a higher accuracy than CNN and LSTM models, and also it indicates the strongest prediction model.

In recent years, successful applications of deep learning (DL) have inspired scientists to explore the possibilities of applying DL approaches to modeling scientific problems. Existing studies have revealed that to bake the physics into the DL makes a good supplement to the traditional numerical methods (e.g., finite element, finite volume method) which primarily rely on partial differential equations (PDEs). While DL models are ordinarily trained in a purely data-driven manner, integrating physics into them for simulating scientific problems has several benefits such as (i) physics constraints could regularize the over-parameterized model and hence mitigate the overfitting issue commonly seen in DL; (ii) physics information could also effectively reduce the amount of data needed for training the model; (iii) the resultant physics-informed DL models feature better interpretability and generalizability compared with the conventional black-box model. Furthermore, the powerful expressiveness of the deep network, guaranteed by the universal approximation theorem, makes it a suitable approximator for the solution to a physical system. In this dissertation, we develop two different DL architectures (or approaches), one being continuous scheme-based while the other discrete scheme-based, that leverage physics knowledge for modeling scientific problems. Through comprehensive numerical experiments, we demonstrate the proposed models can be used in solving general PDEs, establishing predictive data-driven models for dynamical systems, identifying the parameters in governing PDEs or even discovering the entire governing PDEs of dynamical systems from scarce and noisy measurements. The continuous model roots on the physics-informed neural network (PINN) which uses a fully connected neural network (FCNN) to approximate the physical fields of a system globally. This model is mesh-free as the residual of the physics (e.g., PDEs, initial/boundary values) is evaluated on a set of collocation points within the physical domain. Several applications including the forward simulations, data-driven simulations and solving inverse problems are presented to exemplify the advantages of PINN over traditional numerical methods. However, the original PINN suffers from inaccurate initial/boundary values due to the weak enforcement of the initial/boundary conditions (I/BCs). To overcome this issue, we propose an improved PINN model by utilizing multiple deep neural networks (DNNs) to construct the solution. Through a DNN pre-trained to represent the initial/boundary values, the approximated solution would obey the given I/BCs forcibly. With several numerical examples, we show that the improved PINN is characterized with much better accuracy on the I/BCs. Though the PINN shows great promise in data-driven modeling and solving inverse problems, some inherent limitations of PINN still exist, such as (i) the solution might lacks fine-scale details due to the global approximation of FCNN; (ii) high computational expense caused by the FCNN it roots on; (iii) incapability to incorporate existing PDE terms (e.g., $\Delta u$) into the network architecture. To overcome these drawbacks, this dissertation also proposes a discrete model - Physics-encoded Recurrent Convolutional Neural Network (PeRCNN) which recurrently updates the solution (or state variable) for time marching. Specifically, it utilizes convolutional neural network (CNN) to capture the spatial patterns of the solution while the recurrent network mimics the forward Euler scheme (or Runge-Kutta scheme) in numerical methods. PeRCNN is a mesh-based and discrete model due to the discretization in time and spatial dimension. The local connectivity of CNN makes PeRCNN more computationally efficient. In addition, the coercive encoding mechanism of physics in PeRCNN, fundamentally different from the PINN relying on soft penalty, ensures the network to rigorously obey given physics. The proposed PeRCNN is successfully applied to solving general PDEs, the data-driven modeling of dynamical systems and the data-driven discovery of governing PDEs from scarce and noisy measurements. Comparisons with the state-of-the-art DL models demonstrate that the proposed PeRCNN possesses excellent computational efficiency, accuracy and generalizability. --Author's abstract

The generalized mutual information (GMI) of bit-interleaved coded modulation (BICM) systems, sometimes called the BICM capacity, is investigated at low signal-to-noise ratio (SNR). The combinations of input alphabet, input distribution, and binary labeling that achieve the Shannon limit <formula formulatype="inline" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex Notation="TeX">${- 1.59}\;{\rm dB}$</tex></formula> are completely characterized. The main conclusion is that a BICM system with probabilistic shaping achieves the Shannon limit at low SNR if and only if it can be represented as a zero-mean linear projection of a hypercube. Hence, probabilistic shaping offers no extra degrees of freedom to optimize the low-SNR BICM-GMI, in addition to what is provided by geometrical shaping. The analytical conclusions are confirmed by numerical results, which also show that for a fixed input alphabet, probabilistic shaping can improve the BICM-GMI in the low and medium SNR range.

To suppress the mutual coupling between closely-spaced patches, we propose a two-step decoupling design approach for a microstrip antenna array with the dimensions of 44.9 × 30.495 mm2. The first step is designing a decoupling unit on the basis of waveguided complementary split-ring resonators (WCSRRs) to improve isolation. The second step is presenting an optimization method by using a fully connected neural network (FCNN) to enhance design efficiency. By inserting WCSRRs structure between two patches with the edge-to-edge distance of 0.24λ0 (port-to-port distance with 0.66λ0), where λ0 is the wavelength of free space at a resonant frequency of 9 GHz, the mutual coupling can be reduced to −29.38 dB, which is confirmed by the EM simulation. We use the simulation results of the array geometries and EM performances to train FCNN at first and then use the trained FCNN to predict the array geometric parameters for the optimal EM response when the edge-to-edge distance between patches is further reduced to 0.21λ0 and 0.18λ0 (port-to-port distance reduced to 0.60λ0 and 0.56λ0). The measured isolation of the predicted microstrip antenna array is increased to 41.05 dB and 52.33 dB, respectively. Compared to the EM simulation, our FCNN approach has a higher accuracy and lower computational complexity. Therefore, the proposed array predicted by FCNN has a better decoupling performance. All the simulated and predicted results are validated via the measurement to demonstrate the effectiveness of our design scheme. The advantages of our work include the good decoupling performance of the proposed WCSRRs and the convenient optimization of the FCNN from physical response to optimal antenna array geometry design.

Future electron-positron colliders will play a leading role in the precision measurement of Higgs boson couplings which is one of the central interests in particle physics. Aiming at maximizing the performance to measure the Higgs couplings to the bottom, charm and strange quarks, we develop machine learning methods to improve the selection of events with a Higgs decaying to dijets. Our methods are based on the Boosted Decision Tree (BDT), Fully-Connected Neural Network (FCNN) and Convolutional Neural Network (CNN). We find that the BDT and FCNN algorithms outperform the conventional cut-based method. With our improved selection of Higgs decaying to dijet events using the FCNN, the charm quark signal strength is measured with a 16% error, which is roughly a factor of two better than the 34% precision obtained by the cut-based analysis. Also, the strange quark signal strength is constrained as μss≲35 at the 95% C.L. with the FCNN, which is to be compared with μss≲70 obtained by the cut-based method.

Four-quadrant detectors (4QDs), characterized by high efficiency and precision, play crucial roles across diverse technological fields. The traditional 4QD spot positioning method approximates the actual spot center of mass position by outputting an offset value. When the spot center of mass is close to the detector center, the offset value is linear to the actual position, but as the distance increases, the offset value gradually becomes nonlinear, which leads to a decrease in measurement accuracy. To address this problem, this paper proposes a spot localization algorithm based on fully connected neural network (FCNN). First, the fully connected neural network architecture is constructed. Then, a large amount of simulation data is generated through the physical model and the neural network is trained segmentally using the metalearning method. It is verified that the method can effectively improve the localization accuracy of the system while using a small number of real samples.

The research difficulties of speech detection in the present focus on the cases that the signal-to-noise ratio(SNR) is low and the background noise changes dramatically. For the problem of speech detection under low SNR environments, based on the sparsity of speech in frequency domain and the sparse representation ability in frequency domain of the over-complete Fourier basis, the speech signal is reconstructed with Matching Pursuit algorithm, and we propose a low SNR speech detection method which uses the short time energy of the reconstructed signal as a detection feature. The experimental results show that this algorithm exhibits higher robustness in the low SNR white noise environments.

To overcome the limitations of Orthonormal Basis Function (OBF) methods in magnetic anomaly detection, including high false alarm rates and ambiguous target localization due to background noise, this paper introduces a high-confidence detection algorithm based on hierarchical clustering with an optimal cut height. The core of our approach is a theoretically derived optimal cut height, which is calculated from a physical model of the magnetic dipole's vertical gradient field. This model establishes the implicit functional relationship between the effective detection range and key parameters, including magnetic moment orientation, geomagnetic inclination, and sensor height. The calculated optimal cut height serves as the critical criterion in a complete-linkage hierarchical clustering algorithm, which processes the alarm point clouds generated by a preliminary Greatest-of Cell-Averaging Constant False Alarm Rate (GOCA-CFAR) detector. This effectively suppresses isolated false alarms caused by background fluctuations while preserving spatially coherent alarm clusters within the target's effective detection range, thereby significantly enhancing detection confidence. Results from both simulations and field experiments validate the efficacy of the proposed algorithm, demonstrating its superior capability to reliably discriminate genuine targets from false alarms compared to traditional one-dimensional CFAR detection.

P174 ATTENUATING MULTIPLES FOR LOW SIGNAL- TO-NOISE RATIO DATA USING OPTIMAL BEAMFORMING 1 Abstract This paper presents the theory of optimal beamforming and its application in attenuating multiples for low signalto-noise (S/N) ratio data. The key point of the theory is updating the constraint in optimal beamformer design by using adaptive beamformer with the quiescent weighted vector. Real data example shows that the newly developed optimal beamformer has improved the minimum variance unbiased (MVU) beamformer significantly in attenuating multiples for low folding and/or low S/N ratio data without much extra cost. Introduction The objective of seismic data processing is to