Using relative channel impulse responses to train a machine learning algorithm for 2-D surface ship source localization

Matched field processing (MFP) is an established technique for source localization in known multipath acoustic environments. Unfortunately, in many situations, particularly those involving high frequency signals, imperfect knowledge of the actual propagation environment prevents accurate propagation modeling and source localization via MFP fails. For beamforming applications, this actual-to-model mismatch problem was mitigated through a frequency downshift, made possible by a nonlinear array-signal-processing technique called frequency difference beamforming [Abadi, Song, and Dowling (2012). J. Acoust. Soc. Am. 132, 3018-3029]. Here, this technique is extended to conventional (Bartlett) MFP using simulations and measurements from the 2011 Kauai Acoustic Communications MURI experiment (KAM11) to produce ambiguity surfaces at frequencies well below the signal bandwidth where the detrimental effects of mismatch are reduced. Both the simulation and experimental results suggest that frequency difference MFP can be more robust against environmental mismatch than conventional MFP. In particular, signals of frequency 11.2 kHz-32.8 kHz were broadcast 3 km through a 106-m-deep shallow ocean sound channel to a sparse 16-element vertical receiving array. Frequency difference MFP unambiguously localized the source in several experimental data sets with average peak-to-side-lobe ratio of 0.9 dB, average absolute-value range error of 170 m, and average absolute-value depth error of 10 m.

A technique is developed for range-dependent matched field processing (MFP) where the field is computed using a parabolic equation (PE) approximation to the acoustic wave equation. The two-dimensional PE field is computed at each sensor in the array for a set of possible source ranges, depths, and bearings to form an N*2-dimensional model for the acoustic field. Discrete estimates of the position of the source are obtained by applying MFP, in which the match is optimized as a function of source position. Using simulated data, source localization in a noisy range-dependent environment is illustrated for several types of array. For the cases studied in this report, a lower level of ambiguity in localization was observed for range-dependent MFP than for a related range-independent problem. >

It has been proposed that source localization in passive sonar be accomplished via some form of matched field processing (MFP). Fully coherent, minimum variance distortionless response (MVDR) MFP assumes complete and perfect characterization of the underwater multipath propagation channel and is known to be extremely sensitive to mismatch between model parameters and actual environmental parameters. A minimax, multirank signal variation of MVDR MFP, semicoherent MVDR MFP, is presented. Simulation results show semicoherent MVDR MFP to be both relatively robust to mismatch, with respect to relative amplitudes and phases among multipath arrivals, and comparable in performance to fully coherent MVDR MFP under no mismatch conditions. >

A feature-compressed multi-task learning U-Net for shallow-water source localization in the presence of internal waves

Matched field processing (MFP) is generalized beamforming method which utilizes the physics of wave propagation in the ocean waveguide to estimate the range, depth and azimuth of the sources or inversion the waveguide. Traditional MFP only uses the pressure information of the sound field. The vector sensor measures the pressure and three orthogonal components of the particle velocity simultaneously. A novel MFP method based on vertical vector-sensor array for location is proposed. Three Bartlett processors are put forward based on matching acoustic intensity flux. The performance of this algorithm is examined in simulation example. They can suppress side-lobes effectively with high resolution. And the fusion algorithm for the newly Bartlett processors is introduced in to take full advantage of the vector field. Furthermore, the combine processors have a more promising performance. Hence, the method base on matching acoustic intensity flux brings a new way for MFP.

Time-reversal processing (TRP) is an implementation of matched-field processing (MFP) where the ocean itself is used to construct the replica field. This paper introduces virtual time-reversal processing (VTRP) that is implemented electronically at a receiver array and simulates the kind of processing that would be done by an actual TRP during the reciprocal propagation stage. MFP is a forward propagation process, while VTRP is a back-propagation process, which exploits the properties of reciprocity and superposition and is realized by weighting the replica surface with the complex conjugate of the data received on the corresponding element, followed by summation of the processed received data. The number of parabolic equation computational grids of VTRP is much smaller than that of MFP in a range-dependent waveguide. As a result, the localization surface of VTRP can be formed faster than its MFP counterpart in a range-dependent waveguide. As the number of parabolic equation computational grids for VTRP is much smaller than that for MFP, VTRP proceeds about 100 times faster than MFP. The performance of VTRP for source localization is validated through numerical simulations and data from the Mediterranean Sea.

Machine learning has achieved remarkable success across diverse domains. Nevertheless, concerns about interpretability in black-box models, especially within Deep Neural Networks (DNNs), have become pronounced in safety-critical fields like healthcare and finance. Classical machine learning (ML) classifiers, known for their higher interpretability, are preferred in these domains. Similar to DNNs, classical ML classifiers can exhibit bugs that could lead to severe consequences in practice. Test input prioritization has emerged as a promising approach to ensure the quality of an ML system, which prioritizes potentially misclassified tests so that such tests can be identified earlier with limited manual labeling costs. However, when applying to classical ML classifiers, existing DNN test prioritization methods are constrained from three perspectives: 1) Coverage-based methods are inefficient and time-consuming; 2) Mutation-based methods cannot be adapted to classical ML models due to mismatched model mutation rules; 3) Confidence-based methods are restricted to a single dimension when applying to binary ML classifiers, solely depending on the model’s prediction probability for one class. To overcome the challenges, we propose MLPrior, a test prioritization approach specifically tailored for classical ML models. MLPrior leverages the characteristics of classical ML classifiers (i.e., interpretable models and carefully engineered attribute features) to prioritize test inputs. The foundational principles are: 1) tests more sensitive to mutations are more likely to be misclassified, and 2) tests closer to the model’s decision boundary are more likely to be misclassified. Building on the first concept, we design mutation rules to generate two types of mutation features (i.e., <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">model mutation features</b> and <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">input mutation features</b> ) for each test. Drawing from the second notion, MLPrior generates <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">attribute features</b> of each test based on its attribute values, which can indirectly reveal the proximity between the test and the decision boundary. For each test, MLPrior combines all three types of features of it into a final vector. Subsequently, MLPrior employs a pre-trained ranking model to predict the misclassification probability of each test based on its final vector and ranks tests accordingly. We conducted an extensive study to evaluate MLPrior based on 185 subjects, encompassing natural datasets, mixed noisy datasets, and fairness datasets. The results demonstrate that MLPrior outperforms all the compared test prioritization approaches, with an average improvement of 14.74%∼66.93% on natural datasets, 18.55%∼67.73% on mixed noisy datasets, and 15.34%∼62.72% on fairness datasets.

Acoustic waves are well-suited for remote sensing applications and structural health monitoring purposes because they convey information about their source and can be measured using non-contacting methods. Source localization is an important structural health monitoring task; however, traditional time-of-flight array signal processing techniques used to localize acoustic sources are ill-suited for many structural engineering applications due to the potential for complicated propagation paths, the dispersive propagation of acoustic waves in structures, and the coupling of the vibrating structure and the surrounding medium. Thus, source localization experiments were conducted using matched field processing (MFP) for a 0.9-m diameter round aluminum plate excited by the impact of a 1.3-cm stainless-steel ball bearing dropped from 7.6-cm. A 14-sensor linear remote acoustic array placed 8.9-cm above the plate measured the sound radiated by the 0.64-cm thick plate. MFP array signal processing localization techniques were used along with a physics-based finite element acoustic model to localize the excitation on the structure. Source localization results in both a quiet environment and environment with additive white Gaussian noise are discussed, and these results are compared to those from an acoustic model where plate edge reflections are neglected. [Work was sponsored by a SMART Scholarship and the NEEC.]

Matched-field processing (MFP) achieves underwater source localization by measuring the correlation between the array and replica signals, with traditional MFP being equivalent to estimating the Euclidean distance between the data cross-spectral density matrix (CSDM) and replica matrices. However, in practical applications, random inhomogeneities in the marine environment and inaccurate estimation of CSDM reduce MFP performance. The traditional minimum variance matched-field processor with environmental perturbation constraints perturbs a priori environment parameters to obtain linear constraints and yields the optimal weight vectors as the replica vectors. In this paper, within the framework of information geometry, the geometric properties of CSDMs as semi-positive definite and Hermitian enable CSDMs to be described as points in a Riemannian manifold. Source localization can be achieved by quantifying the similarity between the CSDMs as the geodesic distance between the points on the manifold. This paper introduces a constrained replica CSDM composed of perturbed replica vectors and proposes a robust matched-field processor based on two non-Euclidean distances: the Riemannian distance and the modified Jensen-Shannon distance. Simulations and experimental results demonstrate that the proposed processors are more robust against environmental and statistical mismatches than traditional processors and can also reduce sidelobe level and improve the resolution.

Time-reversal processing (TRP) is an implementation of matched-field processing (MFP) where the ocean itself is used to construct the replica field. This paper introduces virtual time-reversal processing (VTRP) that is implemented electronically at a receiver array and simulates the kind of processing that would be done by an actual TRP during the reciprocal propagation stage. MFP is a forward propagation process, while VTRP is a back-propagation process, which exploits the properties of reciprocity and superposition and is realized by weighting the replica surface with the complex conjugate of the data received on the corresponding element, followed by summation of the processed received data. The number of parabolic equation computational grids of VTRP is much smaller than that of MFP in a range-dependent waveguide. As a result, the localization surface of VTRP can be formed faster than its MFP counterpart in a range-dependent waveguide. The performance of VTRP for source localization is validated through numerical simulations and data from the Mediterranean Sea.

Source localization is an important structural health monitoring task, yet traditional localization techniques struggle due to complex geometries, dispersive wave propagation, and structure-medium coupling. This study applied matched field processing (MFP), a source localization technique developed for underwater acoustics, to localize impact sources on metal plates using remote acoustic measurements of airborne sound in conjunction with a physics-based acoustic-wave propagation model. A linear array of 14 microphones recorded radiated sound from a stainless-steel ball bearing dropped onto a 0.64 cm-thick, 91.4 cm-diameter aluminum plate in the nominal 5-20 kHz bandwidth. Physics-based finite element models were developed for both infinite and finite plates. The infinite plate model emphasized generic sound radiation with proper time-windowing, while the finite plate model included edge reflections specific to the plate studied. Both models achieved localization errors within 0.5 cm when data were temporally trimmed to accommodate model constraints. In environments with additive Gaussian noise, the finite plate model maintained greater than 80% localization accuracy down to a signal-to-noise ratio of -7.5 dB. Results further showed that MFP is robust to moderate mismatches in source characterization, but deviations in sensor location approaching a half-wavelength and deviations in plate thickness approaching 10% can reduce localization accuracy.

Matched field processing (MFP) has been widely used in source localization in shallow waters, whose performance is strongly correlated with the knowledge of environmental properties and proper selection of sound propagation model. This paper presents a neural network based approach for source ranging of moving target, which does not need heavy sound field calculation and no requirement of environmental information as a known prior. This neural network is designed to determine source range by observing multiple-frequency sound fields as excited by the source and recorded on a single vector receiver. In synthetic tests, this neural network is first trained on training data set as to adaptively select the optimal features through error back propagation, and then its localization performance is validated on testing data set. This approach is then tested on ship noise data as collected by a single vector sensor, and the relative error is smaller than 0.1 through comparing to GPS calculations. The promising results suggest the proposed approach can be further developed for source ranging applications with light system.

Background: Early identification of probable post-traumatic stress disorder (PTSD) can lead to early intervention and treatment.Aims: This study aimed to evaluate supervised machine learning (ML) classifiers for the identification of probable PTSD in those who are serving, or have recently served in the United Kingdom (UK) Armed Forces.Methods: Supervised ML classification techniques were applied to a military cohort of 13,690 serving and ex-serving UK Armed Forces personnel to identify probable PTSD based on self-reported service exposures and a range of validated self-report measures. Data were collected between 2004 and 2009.Results: The predictive performance of supervised ML classifiers to detect cases of probable PTSD were encouraging when compared to a validated measure, demonstrating a capability of supervised ML to detect the cases of probable PTSD. It was possible to identify which variables contributed to the performance, including alcohol misuse, gender and deployment status. A satisfactory sensitivity was obtained across a range of supervised ML classifiers, but sensitivity was low, indicating a potential for false negative diagnoses.Conclusions: Detection of probable PTSD based on self-reported measurement data is feasible, may greatly reduce the burden on public health and improve operational efficiencies by enabling early intervention, before manifestation of symptoms.

Magnetic resonance imaging (MRI) is playing an important role in the classification of breast tumors. MRI can be used to obtain multiparametric (mp) information, such as structural, hemodynamic, and physiological information. Quantitative analysis of mp-MRI data has shown potential in improving the accuracy of breast tumor classification. In general, a large set of quantitative and texture features can be generated depending upon the type of methodology used. A suitable combination of selected quantitative and texture features can further improve the accuracy of tumor classification. Machine learning (ML) classifiers based upon features derived from MRI data have shown potential in tumor classification. There is a need for further research studies on selecting an appropriate combination of features and evaluating the performance of different ML classifiers for accurate classification of breast tumors. The objective of the current study was to develop and optimize an ML framework based upon mp-MRI features for the characterization of breast tumors (malignant vs. benign and low- vs. high-grade). This study included the breast mp-MRI data of 60 female patients with histopathology results. A total of 128 features were extracted from the mp-MRI tumor data followed by features selection. Five ML classifiers were evaluated for tumor classification using 10-fold crossvalidation with 10 repetitions. The support vector machine (SVM) classifier based on optimum features selected using a wrapper method with an adaptive boosting (AdaBoost) technique provided the highest sensitivity (0.96 ± 0.03), specificity (0.92 ± 0.09), and accuracy (94% ± 2.91%) in the classification of malignant versus benign tumors. This method also provided the highest sensitivity (0.94 ± 0.07), specificity (0.80 ± 0.05), and accuracy (90% ± 5.48%) in the classification of low- versus high-grade tumors. These findings suggest that the SVM classifier outperformed other ML methods in the binary classification of breast tumors.

Deep transfer learning-based damage detection of composite structures by fusing monitoring data with physical mechanism