Abstract: A sonic tool is activated in a well having multiple casings and annuli surrounding the casing. Detected data is preprocessed using slowness time coherence (STC) processing to obtain STC data. The STC data is provided to a machine learning module which has been trained on labeled STC data. The machine learning module provides an answer product regarding the states of the borehole annuli which may be used to make decision regarding remedial action with respect to the borehole casings. The machine learning module may implement a convolutional neural network (CNN), a support vector machine (SVM), or an auto-encoder.

Abstract: Embodiments of the present disclosure disclose a method and a system for training a neural network for improving adversarial robustness. The method includes collecting a plurality of data samples comprising clean data samples and adversarial data samples. The training of the neural network includes training of a probabilistic encoder to encode the plurality of data samples into a probabilistic distribution over a latent space representation. In addition, the training of the neural network comprising training of a classifier to classify an instance of the latent space representation to produce a classification result. In addition, the method includes training shared parameters of a first instance of the neural network using the clean data samples and a second instance of the neural network using the adversarial data samples. Further, the method includes outputting the shared parameters of the first instance of the neural network and the second instance of the neural network.

Abstract: Embodiments of the present disclosure disclose a method and a system for training a neural network for generating universal adversarial perturbations. The method includes collecting a plurality of data samples. Each of the plurality of data samples is identified by a label from a finite set of labels. The method includes training a probabilistic neural network for transforming the plurality of data samples into a corresponding plurality of perturbed data samples having a bounded probability of deviation from the plurality of data samples by maximizing a conditional entropy of the finite set of labels of the plurality of data samples conditioned on the plurality of perturbed data samples. The conditional entropy is unknown. The probabilistic neural network is trained based on an iterative estimation of a gradient of the unknown conditional entropy of labels. The method further includes generating the universal adversarial perturbations based on the trained probabilistic neural network.

Abstract: An imaging system to reconstruct a reflectivity image of a scene including an object moving with the scene. A tracking system to track a deforming object to estimate an object deformation for each time step. Sensors acquire snapshots of the scene, each acquired snapshot of the object includes measurements in the object deformation for that time step, to produce a set of object measurements with deformed shapes over the time steps. Compute a correction to estimates of object deformation for each time step, with matching measurements of the corrected object deformation for each time step to measurements in the acquired snapshot of object for that time step. Select a corrected deformation over other corrected deformations for each time step, according to a distance between the corrected deformation and the estimate of the deformation, to obtain a final estimate of the deformation of the deformable object moving in the scene.

Abstract: A sonic tool is activated in a well having multiple casings and annuli surrounding the casing. Detected data is preprocessed using slowness time coherence (STC) processing to obtain STC data. The STC data is provided to a machine learning module which has been trained on labeled STC data. The machine learning module provides an answer product regarding the states of the borehole annuli which may be used to make decision regarding remedial action with respect to the borehole casings. The machine learning module may implement a convolutional neural network (CNN), a support vector machine (SVM), or an auto-encoder.

Abstract: An imaging system to reconstruct a reflectivity image of a scene including an object moving with the scene. A tracking system to track a deforming object to estimate an object deformation for each time step. Sensors acquire snapshots of the scene, each acquired snapshot of the object includes measurements in the object deformation for that time step, to produce a set of object measurements with deformed shapes over the time steps. Compute a correction to estimates of object deformation for each time step, with matching measurements of the corrected object deformation for each time step to measurements in the acquired snapshot of object for that time step. Select a corrected deformation over other corrected deformations for each time step, according to a distance between the corrected deformation and the estimate of the deformation, to obtain a final estimate of the deformation of the deformable object moving in the scene.

Abstract: A sonic tool is activated in a well having multiple casings and annuli surrounding the casing. Detected data is preprocessed using slowness time coherence (STC) processing to obtain STC data. The STC data is provided to a machine learning module which has been trained on labeled STC data. The machine learning module provides an answer product regarding the states of the borehole annuli which may be used to make decision regarding remedial action with respect to the borehole casings. The machine learning module may implement a convolutional neural network (CNN), a support vector machine (SVM), or an auto-encoder.

Abstract: Methods and apparatus for waveform processing are disclosed. An example method includes determining shrinkage estimators in a Discrete Radon transform domain based on semblance of waveform data and de-noising the waveform data using a processor and the shrinkage estimators to enable the identification of weak signals in the waveform data.

Abstract: Methods and apparatus for waveform processing are disclosed. An example method includes representing waveform data using space time propagators in the Discrete Radon Transform Domain. The method also includes identifying signals within the represented waveform data using a Sparisty Penalized Transform.

Abstract: Slowness dispersion characteristics of multiple possibly interfering signals in broadband acoustic waves as received by an array of two or more sensors are extracted without using a physical model. The problem of dispersion extraction is mapped to the problem of reconstructing signals having a sparse representation in an appropriately chosen over-complete dictionary of basis elements. A sparsity penalized signal reconstruction algorithm is described where the sparsity constraints are implemented by imposing a l1 norm type penalty. The candidate modes that are extracted are consolidated by means of a clustering algorithm to extract phase and group slowness estimates at a number of frequencies which are then used to reconstruct the desired dispersion curves. These estimates can be further refined by building time domain propagators when signals are known to be time compact, such as by using the continuous wavelet transform.

Abstract: Methods and apparatus for waveform processing are disclosed. An example method includes determining shrinkage estimators in a Discrete Radon transform domain based on semblance of waveform data and de-noising the waveform data using a processor and the shrinkage estimators to enable the identification of weak signals in the waveform data.

Abstract: Slowness dispersion characteristics of multiple possibly interfering signals in broadband acoustic waves as received by an array of two or more sensors are extracted without using a physical model. The problem of dispersion extraction is mapped to the problem of reconstructing signals having a sparse representation in an appropriately chosen over-complete dictionary of basis elements. A sparsity penalized signal reconstruction algorithm is described where the sparsity constraints are implemented by imposing a l1 norm type penalty. The candidate modes that are extracted are consolidated by means of a clustering algorithm to extract phase and group slowness estimates at a number of frequencies which are then used to reconstruct the desired dispersion curves. These estimates can be further refined by building time domain propagators when signals are known to be time compact, such as by using the continuous wavelet transform.

Abstract: Slowness dispersion characteristics of multiple possibly interfering signals in broadband acoustic waves as received by an array of two or more sensors are extracted without using a physical model. The problem of dispersion extraction is mapped to the problem of reconstructing signals having a sparse representation in an appropriately chosen over-complete dictionary of basis elements. A sparsity penalized signal reconstruction algorithm is described where the sparsity constraints are implemented by imposing a l1 norm type penalty. The candidate modes that are extracted are consolidated by means of a clustering algorithm to extract phase and group slowness estimates at a number of frequencies which are then used to reconstruct the desired dispersion curves. These estimates can be further refined by building time domain propagators when signals are known to be time compact, such as by using the continuous wavelet transform.

Abstract: This invention pertains to the extraction of the slowness dispersion characteristics of acoustic waves received by an array of two or more sensors by the application of a continuous wavelet transform on the received array waveforms (data). This produces a time-frequency map of the data for each sensor that facilitates the separation of the propagating components thereon. Two different methods are described to achieve the dispersion extraction by exploiting the time frequency localization of the propagating mode and the continuity of the dispersion curve as a function of frequency. The first method uses some features on the modulus map such as the peak to determine the time locus of the energy of each mode as a function of frequency. The second method uses a new modified Radon transform applied to the coefficients of the time frequency representation of the waveform traces received by the aforementioned sensors.

Abstract: This invention pertains to the extraction of the slowness dispersion characteristics of acoustic waves received by an array of two or more sensors by the application of a continuous wavelet transform on the received array waveforms (data). This produces a time-frequency map of the data for each sensor that facilitates the separation of the propagating components thereon. Two different methods are described to achieve the dispersion extraction by exploiting the time frequency localization of the propagating mode and the continuity of the dispersion curve as a function of frequency. The first method uses some features on the modulus map such as the peak to determine the time locus of the energy of each mode as a function of frequency. The second method uses a new modified Radon transform applied to the coefficients of the time frequency representation of the waveform traces received by the aforementioned sensors.