General Electric Company

General Dynamics Mission Systems, Inc.

Deep Learning Analytics, LLC

Exxonmobil Upstream Research Company

Ge Medical Systems Global Technology Company, LLC

John Patrick Kaufhold

Jeffrey Wayne Eberhard

Bernhard Erich Hermann Claus

Gopal Biligeri Avinash

John Michael Sabol

Kadri Nizar Jabri

John Eric Tkaczyk

Dinko Eduardo Gonzalez Trotter

Frederick Wilson Wheeler

Mark Lewis Grabb

Matthias Imhof

Stephen Wayne Metz

Pavel Dimitrov

Abdalmajeid Musa Alyassin

Cynthia Elizabeth Landberg

Abdulrahman Abdallah Al-Khalidy

Jerry A Thomas

Gulnar Siddiqi

Michael Jeremy Trammell

Bernhard E H Claus

Jennifer Alexander Sleeman

Bernard E H Claus

The present invention is directed to systems and methods for detecting objects in a radar image stream. Embodiments of the invention can receive a data stream from radar sensors and use a deep neural network to convert the received data stream into a set of semantic labels, where each semantic label corresponds to an object in the radar data stream that the deep neural network has identified. Processing units running the deep neural network may be collocated onboard an airborne vehicle along with the radar sensor(s). The processing units can be configured with powerful, high-speed graphics processing units or field-programmable gate arrays that are low in size, weight, and power requirements. Embodiments of the invention are also directed to providing innovative advances to object recognition training systems that utilize a detector and an object recognition cascade to analyze radar image streams in real time. The object recognition cascade can comprise at least one recognizer that receives a non-background stream of image patches from a detector and automatically assigns one or more semantic labels to each non-background image patch. In some embodiments, a separate recognizer for the background analysis of patches may also be incorporated. There may be multiple detectors and multiple recognizers, depending on the design of the cascade. Embodiments of the invention also include novel methods to tailor deep neural network algorithms to successfully process radar imagery, utilizing techniques such as normalization, sampling, data augmentation, foveation, cascade architectures, and label harmonization.

Systems and methods for recognizing objects in radar imagery

Embodiments of the present invention relate to systems and methods for improving the training of machine learning systems to recognize certain objects within a given image by supplementing an existing sparse set of real-world training images with a comparatively dense set of realistic training images. Embodiments may create such a dense set of realistic training images by training a machine learning translator with a convolutional autoencoder to translate a dense set of synthetic images of an object into more realistic training images. Embodiments may also create a dense set of realistic training images by training a generative adversarial network ('GAN') to create realistic training images from a combination of the existing sparse set of real-world training images and either Gaussian noise, translated images, or synthetic images. The created dense set of realistic training images may then be used to more effectively train a machine learning object recognizer to recognize a target object in a newly presented digital image.

Systems and methods for deep model translation generation

Embodiments of the present invention are directed to providing new systems and methods for using deep learning techniques to generate embeddings for high dimensional data objects that can both simulate prior art embedding algorithms and also provide superior performance compared to the prior art methods. Deep learning techniques used by embodiments of the present invention to embed high dimensional data objects may comprise the following steps: (1) generating an initial formal embedding of selected high-dimensional data objects using any of the traditional formal embedding techniques; (2a) designing a deep embedding architecture, which includes choosing the types and numbers of inputs and outputs, types and number of layers, types of units/nonlinearities, and types of pooling, for example, among other design choices, typically in a convolutional neural network; (2b) designing a training strategy; (2c) tuning the parameters of a deep embedding architecture to reproduce, as reliably as possible, the generated embedding for each training sample; (3) optionally deploying the trained deep embedding architecture to convert new high dimensional data objects into approximately the same embedded space as found in step (1); and optionally (4) feeding the computed embeddings of high dimensional objects to an application in a deployed embodiment.

Systems and methods for fast and repeatable embedding of high-dimensional data objects using deep learning with power efficient GPU and FPGA-based processing platforms

Method for representing seismic data as a spatially varying, second-order tensor field (). The spatial relationships expressed in these tensors are exploited to link, classify, or separate neighborhoods (); or to infer global or relational properties among them, thereby suggesting geobodies despite a noisy background. The tensors may be decomposed into their fundamentals that may either be used directly as derivative datasets or attributes, or may be used to facilitate linkage, classification, or separation of neighborhoods or analysis of linkage patterns. Decomposition may be by eigenvalues, with the eigenvalues used to define attributes called ballness, plateness and stickness (). Alternatively, connections may be made between points in the data where the tensor has been computed, called tokens (), based on tensor voting and polarity. Or, the connections may be based on a distance measure between tokens.

Tensor-based method for representation, analysis, and reconstruction of seismic data

A technique is provided for generating quantitative projection images from projection images. The pixels of the quantitative projection images correspond to quantitative composition estimates of two or more materials. The quantitative projection images are reconstructed to generate a quantitative volume in which each voxel value corresponds quantitatively to the two or more materials or a mixture of the two or more materials.

Quantitative image reconstruction method and system

At least one metric for a mammographic image is computed by a workstation, and provided on a workstation display along with the mammographic image. A clinician can select one or more different metrics to be computed for an image, as well as where they are to be shown and the manner in which they are to be shown on the display. Speech (may also be an audible sound which is not speech) may also be used in the workstation to audibly provide metrics information to the clinician.

Method and apparatus for providing mammographic image metrics to a clinician

A technique is provided for comparative image analysis and/or change detection using computer assisted detection and/or diagnosis (CAD) algorithms. The technique includes registering two or more images, comparing the images with one another to generate a change map, and detecting anomalies in the images based on the change map.

Method and system for volumetric comparative image analysis and diagnosis

Some configurations of method for reconstructing a volumetric image of an object include obtaining a tomosynthesis projection dataset of an object. The method also includes utilizing the tomosynthesis projection dataset and additional information about the object to minimize a selected energy function or functions to satisfy a selected set of constraints. Alternatively, constraints are applied to a reconstructed volumetric image in order to obtain an updated volumetric image. A 3D volume representative of the imaged object is thereby obtained in which each voxel is reconstructed and a correspondence indicated to a single one of the component material classes.

Methods and apparatus for reconstruction of volume data from projection data

Briefly in accordance with one embodiment, the present technique provides a multi-energy tomosynthesis imaging system. The system includes an X-ray source configured to emit X-rays from multiple locations within a limited angular range relative to an imaging volume. The imaging system also includes a digital detector with an array of detector elements to generate images in response to the emitted X-rays. The imaging system further includes a detector acquisition circuitry to acquire the images from the digital detector. The imaging system may also include a processing circuitry configured to decompose plurality of images based on energy characteristics and to reconstruct the plurality of images to generate a three-dimensional multi-energy tomosynthesis image.

Method and system for multi-energy tomosynthesis

A method of creating and displaying images resulting from digital tomosynthesis performed on a subject using a flat panel detector is disclosed. The method includes the step of acquiring a series of x-ray images of the subject, where each x-ray image is acquired at different angles relative to the subject. The method also includes the steps of applying a first set of corrective measures to the series of images, reconstructing the series of images into a series of slices through the subject, and applying a second set of corrective measures to the slices. The method further includes the step of displaying the images or slices according to at least one of a plurality of display options.

Imaging chain for digital tomosynthesis on a flat panel detector

An imaging system for scanning a volume of interest in an object, the system includes a radiation source configured to traverse in a plurality of focal spot positions yielding a plurality of focal spot trajectories. Each focal spot trajectory defines a two dimensional focal spot projection in an image acquisition plane. Each of the focal spot positions comprise at least two positions at unequal distances from the image acquisition plane.

Method and system for three dimensional tomosynthesis imaging

A method for facilitating an enhancement of a visibility of an object in an x-ray image includes generating an x-ray image including at least one object, generating an estimate of a background surrounding the at least one object, subtracting the background estimate from the x-ray image to generate an estimate of pixel intensities due the object, mapping the estimate of pixel intensities due to the object, and combining the mapped estimate of pixel intensities due to the object and the x-ray image to generate an enhanced image.

Method and apparatus for enhancing an image

A method for estimating a material composition of an imaged object using an imaging system. The imaging system includes a radiation source and a digital detector. The method also includes scanning a plurality of calibration phantoms with varying material composition to acquire a plurality of reference calibration images, estimating an attenuation coefficient thickness product for each pixel in the reference calibration images, and estimating a material composition of a region of interest using the estimated pixelwise coefficient thickness product.

Growing community of inventors

Arlington, VA, United States of America

John Patrick Kaufhold