A method for diagnosing an open-circuit fault of a switching transistor of a single-phase half-bridge five-level inverter is provided. It includes the following steps. A semi-physical experiment platform with a DSP controller and an RT-LAB real-time simulator as its core constructed, and an output side voltage is selected as a fault signal variable. Empirical mode decomposition is used to extract a fault feature vector, and then a HHT time-frequency diagram of the fault feature vector is extracted, a voltage signal is converted into spectrum data, and time-frequency diagram fuzzy sets corresponding to different fault types are obtained. Fusion of the time-frequency diagram fuzzy sets of the same fault type is performed to obtain a fusion image that contains more fault features. The fusion images corresponding to all fault types are inputted into the deep convolutional neural network for training and testing, and a fault diagnosis result is obtained.

This invention presents a system that can be used for object identification and classification by training multiple neural networks on large quantities of images efficiently, The first in a series of convolutional neural networks is trained on a low resolution version of the image; in each successive stage of the series a model is trained on a smaller and more specific subregion of the original image. GradCAM is used to identify an area of focus in the image which later models will classify. The models are strung together into a single mega-classifier. The training time of this approach is significantly less, as smaller and lower resolution images are easier to manipulate and the implementation of GradCAM presented is much faster than standard library implementations The effectiveness of the proposed approach is demonstrated by applying it on the task of Intracranial Hemorrhage detection and classification.

Intracranial hemorrhage is a critical brain injury characterized by bleeding and swelling in the tissue surrounding a broken artery. Hemorrhages often cause strokes which are the 5th leading cause of death in the U.S. Current diagnostic procedures need a highly trained radiologist with specialized training in identifying brain hemorrhage. As a result, diagnosis is expensive, and in remote areas where radiologists are hard to find, diagnosis is difficult and often inaccurate. My research develops a computer aid to radiologists that can screen brain scans to cut costs and accelerate diagnosis. Through image windowing, data augmentation, and Convolutional Neural Networks (CNNs), the system I present achieves high accuracies in detecting hemorrhage and 5-way subtype classification. The system consists of a two-model ensemble; one model is trained to detect hemorrhage and potential regions of hemorrhage in the CT scans, and the second model analyzes hemorrhagic regions found by the first model more closely. The two-model ensemble reduces the error rate by 17% relative to the first model alone, increasing the overall detection accuracy to 97.0%. It also applies Gradient Class Activation Maps (GradCAM), which provide a coarse mapping of the regions of the image that were most influential in the model's predictions. The activation maps provide a strong visual aid for explaining and justifying the model's outputs and can be used by radiologists to assist them in identifying the areas of focus in an image.

An object detection network includes: a hybrid voxel feature extractor configured to acquire a raw point cloud, extract a hybrid scale voxel feature from the raw point cloud, and project the hybrid scale voxel feature to generate a pseudo-image feature map; a backbone network configured to perform a hybrid voxel scale feature fusion by using the pseudo-image feature map to generate multi-class pyramid features; and a detection head configured to predict a three-dimensional object box of a corresponding class according to the multi-class pyramid features. The object detection network can effectively solve a problem that under a single voxel scale, inference time is longer if the voxel scale is smaller, and an intricate feature cannot be captured and a smaller object cannot be accurately located if the voxel scale is larger. Different classes of 3D objects can be detected quickly and accurately in a 3D scene.

A computer-implemented method of detecting one or more objects in a driving environment located externally to a vehicle, and a vehicle imaging system configured to detect one or more objects. The computer-implemented method includes training a first neural network to detect objects in a color video stream, the first neural network having a plurality of mid-level color features at a plurality of scales, and training a second neural network, operatively coupled to color neural network and an infrared video stream, to match, at the plurality of scales, mid-level infrared features of the second neural network to mid-level color features of the first neural network. A pixel-level invisibility map is then generated from the color video stream and the infrared video stream by determining differences, at each of the plurality of scales, between mid-level color features at the first neural network and mid-level infrared features at the second infrared neural network, and coupling the result to a fusing function.

Object parts (20, 21, 22, 23, 24) are detected in a picture using object detector(s) (3) and part location representations (40, 42, 43, 44) are generated for the detected object parts (20, 22, 23, 24). The size of an object (10) comprising object parts (20, 21, 22, 23, 24) is estimated based on a geometric model and the part location representations (40, 42, 43, 44). Search locations (51) in the picture for a search window (52) having a size based on the estimated size are determined based on the part location representations (40, 42, 43, 44). The search locations (51) are then processed by identifying any detected object part (20, 22, 23) that is within the search window (52) positioned at the search location (51). A homography is estimated by minimizing an error between mapped object part(s) from the geometric model and the identified detected object part(s) (20, 22, 23). If the error is smaller than a threshold value, an object location representation is determined for the object (10).

Systems and methods for detecting and aligning land property boundaries on aerial imagery are provided. The system receives an aerial imagery having land properties. The system applies a feature encoder having a plurality of levels to the aerial imagery. A first level of the plurality of levels includes a convolution block and a discrete wavelet transform layer. The discrete wavelet transform layer decomposes an input feature tensor to the first level into a low-frequency band and a high-frequency band. The high-frequency band is cached and processed with side-convolutional blocks before the high-frequency band are passed to a feature decoder. The system applies the feature decoder to an output of the feature encoder based at least in part on one of inverse discrete wavelet transform layers. The system determines boundaries of the one or more land properties based at least in part on a boundary cross-entropy loss function.

Techniques for generating magnetic resonance (MR) images from MR data obtained by a magnetic resonance imaging (MRI) system comprising a plurality of RF coils configured to detect RF signals. The techniques include: obtaining a plurality of input MR datasets obtained by the MRI system to image a subject, each of the plurality of input MR datasets comprising spatial frequency data and obtained using a respective RF coil in the plurality of RF coils; generating a respective plurality of MR images from the plurality of input MR datasets by using an MR image reconstruction technique; estimating, using a neural network model, a plurality of RF coil profiles corresponding to the plurality of RF coils; generating an MR image of the subject using the plurality of MR images and the plurality of RF coil profiles; and outputting the generated MR image.

An illustrative multiscale data system determines a first distance between a first object in a scene and a virtual vantage point at the scene. The multiscale data system also determines a second distance between a second object in the scene and the virtual vantage point. In an example in which the second distance is greater than the first distance, the multiscale data system generates, based on the first and second distances, a tiled representation associated with the virtual vantage point. The tiled representation in this example includes a first representation of the first object rendered at a first quality level and a second representation of the second object rendered at a second quality level lower than the first quality level. Corresponding methods and systems are also disclosed.

A computer vision system including: one or more processors; and memory including instructions that, when executed by the one or more processors, cause the one or more processors to: determine a semantic multi-scale context feature and an instance multi-scale context feature of an input scene; generate a joint attention map based on the semantic multi-scale context feature and the instance multi-scale context feature; refine the semantic multi-scale context feature and instance multi-scale context feature based on the joint attention map; and generate a panoptic segmentation image based on the refined semantic multi-scale context feature and the refined instance multi-scale context feature.

A computer vision system including: one or more processors; and memory including instructions that, when executed by the one or more processors, cause the one or more processors to: determine a semantic multi-scale context feature and an instance multi-scale context feature of an input scene; generate a joint attention map based on the semantic multi-scale context feature and the instance multi-scale context feature; refine the semantic multi-scale context feature and instance multi-scale context feature based on the joint attention map; and generate a panoptic segmentation image based on the refined semantic multi-scale context feature and the refined instance multi-scale context feature.