Abstract: A signal generator outputs a reference signal corresponding to at least one wireless signal according to the predefined signal encoding to a channel emulator processor. The channel emulator processor is programmed to use at least one synthesized channel parameter and the reference signal to produce and store a perturbed signal as data for training machine learning and artificial intelligence systems. The synthesized channel parameter is synthesized using a channel synthesizer processor programmed to: ingest map elevation data, reference a transmitter and a receiver to the map elevation data, and perform ray tracing of a representative signal between the transmitter and the receiver, while applying at least one predetermined perturbation property to synthesize at least one channel parameter.

Abstract: 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.

Abstract: 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.

Abstract: 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.

Abstract: 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.

Abstract: 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.

Abstract: A cable containing an optical fiber is used to transmit data between an underwater remotely operated vehicle (ROV) and a support vessel floating on the surface of the water. The ROV stores the cable on a spool and releases the cable into the water as the ROV dives away from the support vessel. The ROV detects the tension in the cable and the rate that the cable is released from the ROV is proportional to the detected tension in the cable. After the ROV has completed the dive and retrieved by the support vessel, the cable can be retrieved from the water and rewound onto the spool in the ROV.

Abstract: 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.

Abstract: Methods and systems are provided for detecting a defect in a solar panel. The method includes initially imaging, via an infrared camera, a group of solar panels. Then, identifying, via a computer system configured for solar panel defect detection, the individual solar panels in the group of solar panels. Finally, identifying, via evaluation of an infrared image obtained by the infrared camera, a defect in at least one of the group of solar panels.

Abstract: A flow-through card rail module is provided in a circuit module chassis assembly for an embedded computing system. A set of elongated guide rails are formed on a base plate and define a card channel for receiving a circuit card. Each guide rail has a cooling passage extending from a fluid inlet to a fluid outlet. A corrugated structure is formed on an opposite side of the base plate and includes a set of elongated cells. Each elongated cell has a cooling passage formed therein extending from the fluid inlet to the fluid outlet. Internal walls subdivide the cooling passages formed in the guide rails and the elongated cells to form a honeycomb structure. The flow-through card rail module including the base plate, the guide rails and the corrugated structure may be formed as a monolithic component.

Abstract: A method is provided for sharing access to graphics processing unit (GPU) hardware between multiple client virtual machines, wherein each of the client virtual machines has a high-level application programming interface (API) associated therewith for communicating with the GPU hardware. The method includes virtualizing the GPU by intercepting GPU-specific commands from the plurality of client virtual machines, wherein the commands specific to the GPU are at a lower level than that of the high-level API, and providing the intercepted commands to the GPU hardware.

Abstract: Autonomous, unmanned submersible can turn upside down in order to use instrumentation placed on one side only. Batteries (106) and instrumentation housing (104) are installed on frame rails (112,14,108,110) and can change their relative positions, thus inverting the relative positions between center of buoyancy and center of gravity and subsequently inverting the submersible.

Abstract: A super-heterodyne receiver includes a plurality of input filters configured to receive an input signal and divide up a tuning range of the input signal into a plurality of respective bands, wherein the input filters are divided into a plurality of filter sets, each having a respective filter set output, and each of the input filters has a center frequency and a bandwidth. The receiver further includes a first amplifier having an input coupled to a first filter set output and an output coupled to a first mixer; a second amplifier having an input coupled to a second filter set output and an output coupled to a second mixer. The first filter set comprises a first plurality of input filters having respective center frequencies ranging from approximately 1250 MHz to 6250 MHz, a second plurality of input filters having respective center frequencies ranging from approximately 8250 MHz to 18500 MHz, and a third plurality of center frequencies ranging from approximately 24000 MHz to 38000 MHz.