1926 results about "Learning network" patented technology

The present invention relates to a system and methodology facilitating automation security in a networked-based industrial controller environment. Various components, systems and methodologies are provided to facilitate varying levels of automation security in accordance with security analysis tools, security validation tools and / or security learning systems. The security analysis tool receives abstract factory models or descriptions for input and generates an output that can include security guidelines, components, topologies, procedures, rules, policies, and the like for deployment in an automation security network. The validation tools are operative in the automation security network, wherein the tools perform security checking and / or auditing functions, for example, to determine if security components are in place and / or in suitable working order. The security learning system monitors / learns network traffic patterns during a learning phase, fires alarms or events based upon detected deviations from the learned patterns, and / or causes other automated actions to occur.

Methods and apparatus for deep learning-based system design improvement are provided. An example system design engine apparatus includes a deep learning network (DLN) model associated with each component of a target system to be emulated, each DLN model to be trained using known input and known output, wherein the known input and known output simulate input and output of the associated component of the target system, and wherein each DLN model is connected as each associated component to be emulated is connected in the target system to form a digital model of the target system. The example apparatus also includes a model processor to simulate behavior of the target system and / or each component of the target system to be emulated using the digital model to generate a recommendation regarding a configuration of a component of the target system and / or a structure of the component of the target system.

Methods and apparatus to automatically generate an image quality metric for an image are provided. An example method includes automatically processing a first medical image using a deployed learning network model to generate an image quality metric for the first medical image, the deployed learning network model generated from a digital learning and improvement factory including a training network, wherein the training network is tuned using a set of labeled reference medical images of a plurality of image types, and wherein a label associated with each of the labeled reference medical images indicates a central tendency metric associated with image quality of the image. The example method includes computing the image quality metric associated with the first medical image using the deployed learning network model by leveraging labels and associated central tendency metrics to determine the associated image quality metric for the first medical image.

The present invention provides improved unicast routing, multicast routing and unicast load sharing as compared with conventional methods. Preferred implementations of the invention provide improvements to IEEE 802.1Q. According to preferred aspects of the invention, each bridge is the root of its own multiple spanning tree instance (“MSTI”). Preferred implementations of the invention require no learning of media access control (“MAC”) addresses on the backbone of a network. Some methods of the invention can resolve spanning tree asymmetries. Preferred implementations of the invention require a very low computational load for control protocols.

The invention discloses a high-efficiency anti-DDoS security gateway system, which can effectively detect and defend a DNS denial-of-service attack. The system comprises two core components, namely a detector and a filter. The system supports two deploying ways, namely serial connection and bypass. A detecting method is based on the self-learning network flow state statistic and characteristic disinterment, achieves the detection and positioning of the network abnormal flow by setting network performance parameter thresholds, and can effectively identify the suspected attack flow. A defending method carries out a thought of deep defense, and the system is deployed with two defense steps of attack characteristic defense and baseline defense so as to ensure the attack defending effect of thesystem under normal network condition and the basic defense capacity of the system in the individual and severe attack environment. The methods can effectively improve the security and attack resista nce of a DNS server and can ensure the normal operation of the DNS service.

The invention discloses an automatic driving system based on enhanced learning and multi-sensor fusion. The system comprises a perception system, a control system and an execution system. The perception system high-efficiently processes a laser radar, a camera and a GPS navigator through a deep learning network so as to realize real time identification and understanding of vehicles, pedestrians, lane lines, traffic signs and signal lamps surrounding a running vehicle. Through an enhanced learning technology, the laser radar and a panorama image are matched and fused so as to form a real-time three-dimensional streetscape map and determination of a driving area. The GPS navigator is combined to realize real-time navigation. The control system adopts an enhanced learning network to process information collected by the perception system, and the people, vehicles and objects of the surrounding vehicles are predicted. According to vehicle body state data, the records of driver actions are paired, a current optimal action selection is made, and the execution system is used to complete execution motion. In the invention, laser radar data and a video are fused, and driving area identification and destination path optimal programming are performed.

Apparatus, systems, and methods to improve imaging quality control, image processing, identification of findings, and generation of notification at or near a point of care are disclosed and described. An example imaging apparatus includes a processor to at least: evaluate first image data with respect to an image quality measure; when the first image data satisfies the image quality measure, process the first image data using a trained learning network to generate a first analysis of the first image data; identify a clinical finding in the first image data based on the first analysis; compare the first analysis to a second analysis, the second analysis generated from second image data obtained in a second image acquisition; and, when comparing identifies a change between the first analysis and the second analysis, trigger a notification at the imaging apparatus regarding the clinical finding to prompt a responsive action.

The invention discloses a passenger abnormal behavior identification method based on a human skeleton sequence, which comprises the following steps: 1) shooting an escalator area monitoring video image by a camera; 2) detecting that passenger face through the SVM and trac the passenger face with the KCF to obtain the passenger motion trajectory in the escalator; 3) extracting that human skeleton from the image by using the OpenPose depth learn network; 4) matching that human skeleton to the corresponding passenger trajectory to construct the human skeleton sequence of the passenger; 5) detecting that abnormal behavior skeleton sequence from the passenger human skeleton sequence through template match; 6) using DTW to match it with various abnormal behavior skeleton sequence template to identify abnormal behavior. That invention can accurately and real-time identify a plurality of abnormal behavior of passengers in the escalator base on human skeleton sequence, according to the abnormalbehavior type to control the operation situation of the escalator, and avoid the occurrence of safety accident.

The invention belongs to the field of depth learning, and provides a rail surface defect detection method based on the depth learning network, aiming at solving various problems existing in the priorrail detection methods. The depth learning method first automatically resets the input rail image to 416*416, and then extracts and processes the image. Image extraction mainly by Darknet-53 model complete. The processing output is mainly accomplished by the FPN-like network model. Firstly, the rail image is divided into cells. According to the position of the defects in the cells, the width, height and coordinates of the center point of the defects are calculated by dimension clustering method, and the coordinates are normalized. At the same time, we use logistic regression to predict the fraction of boundary box object, use binary cross-entropy loss to predict the category contained in the boundary box, calculate the confidence level, and then process the convolution in the output, up-sampling, network feature fusion to get the prediction results. The invention can accurately identify defects and effectively improve the detection and identification rate of rail surface defects.

The present invention discloses a surveillance video pedestrian re-recognition method based on ImageNet retrieval. The pedestrian re-recognition problem is transformed into the retrieval problem of an moving target image database so as to utilize the powerful classification ability of an ImageNet hidden layer feature. The method comprises the steps: preprocessing a surveillance video and removing a large amount of irrelevant static background videos from the video; separating out a moving target from a dynamic video frame by adopting a motion compensation frame difference method and forming a pedestrian image database and an organization index table; carrying out alignment of the size and the brightness on an image in the pedestrian image database and a target pedestrian image; training hidden features of the target pedestrian image and the image in the image database by using an ImageNet deep learning network, and performing image retrieving based on cosine distance similarity; and in a time sequence, converging the relevant videos containing recognition results into a video clip reproducing the pedestrian activity trace. The method disclosed by the present invention can better adapt to changes in lighting, perspective, gesture and scale so as to effective improve accuracy and robustness of a pedestrian recognition result in a camera-cross environment.

The invention discloses a three-dimensional human body posture estimation method and device. The method comprises the following steps of S1, using a monocular camera to collect the depth images and the RGB color images of a human body at different angles; S2, constructing a human body skeleton key point detection nerve network based on the RGB color images, and obtaining a key point labeling image; S3, constructing hand joint node 2D-3D mapping network; S4, calibrating the depth image of the human body at the same angle and the key point labeling image, and then carrying out three-dimensionalpoint cloud color conversion on the corresponding depth image so as to obtain a colored depth image; S5, based on the key point labeling image and the colored depth image, using a preset learning network to predict the corresponding position of the labeled human body skeleton key point in the depth image; and S6, combining the outputs of the step S3 and the step S5 to achieve the refined estimation of three-dimensional human body posture estimation.

The present invention proposes an unsupervised hash image retrieval system and method based on the convolution neural network. The system and method based on the data enhancement technology, proposes an efficient unsupervised hash model used for the field of fast image retrieval by using the existing hash algorithm structure. Through the data enhancement method, a triad training sample is constructed for the unlabeled data; and the triad loss function, the minimum quantization error loss function and the maximum entropy loss function drive the network to make full use of the information of each picture, so that a series of parameters with more expressive ability is learned to improve the accuracy of fast picture retrieval. The method provided by the present invention is a hash fast image retrieval method capable of using the unlabeled data learning network, the data enhancement is used to construct the triad training sample with more powerful expression ability to train the network, and the accuracy of fast picture retrieval is significantly improved.

A system and related techniques permit a search service operator to access a variety of disparate relevance measures, and integrate those measures into idealized or unified data sets. A search service operator may employ self-learning networks to generate relevance rankings of Web site hits in response to user queries or searches, such as Boolean text or other searches. To improve the accuracy and quality of the rankings of results, the service provider may accept as inputs relevance measures created from query logs, from human-annotated search records, from independent commercial or other search sites, or from other sources and feed those measures to a normalization engine. That engine may normalize those relevance ratings to a common scale, such as quintiles, percentages or other scales or levels. The provider may then use that idealized or normalized combined measure to, for example, train the search algorithms or heuristics to arrive at better or more accurate results.

Systems and processes for operating a digital assistant are provided. In accordance with one or more examples, a method includes, receiving training data for a data-driven learning network. The training data include a plurality of word sequences. The method further includes obtaining representations of an initial set of semantic categories associated with the words included in the training data; and training the data-driven learning network based on the plurality of word sequences included in the training data and based on the representations of the initial set of semantic categories. The training is performed using the word sequences in their entirety. The method further includes obtaining, based on the trained data-driven learning network, representations of a set of semantic embeddings of the words included in the training data; and providing the representations of the set of semantic embeddings to at least one of a plurality of different natural language processing tasks.

The invention discloses an attitude detection method using a deep convolutional neural network, and is suitable for being performed in computing equipment. The method comprises the steps that a data set is divided into training and testing and preprocessed; identification learning model training of human body joint feature areas is performed so as to identify the learning network of human body joint image areas; joint coordinate positioning learning model training is performed; detection image size preprocessing is performed, and the images requiring human body attitude identification are adjusted into the size of the network input requirements; identification of the image joint areas is performed through the network and corresponding rectangular areas are defined to be saved as sub-images; the acquired sub-images act as the input to be transmitted to the joint coordinate positioning learning model to perform joint coordinate acquisition; and the acquired joint points are connected according to the human skeleton model so as to form human body attitude description. The invention also provides storage equipment and a mobile terminal.

The invention provides a multi-task deep learning network-based training method, a multi-task deep learning network-based training system, a multi-task deep learning network-based identification method and a multi-task deep learning network-based identification system. The training method includes the following steps that: the face region of a face image in a training set is obtained; key point detection is performed on the face region, so that key feature point positions are obtained; affine transformation is performed on the face image according to the key feature positions, so that an aligned face image can be obtained; and the aligned face image is inputted into a multi-task deep learning network, so that training can be carried out, and therefore, a multi-task deep learning network model can be obtained. The identification method includes the following steps that: affine transformation is performed on a face image to be identified according to the key feature positions of the face image to be identified, so that an aligned face image can be obtained; the aligned face image is inputted into a trained multi-task deep learning network model, so that feature extraction can be carried out, and feature information can be obtained; and the feature information of the face image to be identified is matched with feature information corresponding to each face image in a registration set, so that identification results can be obtained. With the methods and systems adopted, the training and identification efficiency of the multi-task deep learning network can be improved.

The invention discloses a dual-channel convolution neural network semantic segmentation method sensitive to small targets. The method comprises the following steps: a Caffe depth learning frame is used to build a non-weighted learning network and a weighted learning network; for the two-channel network, the corresponding semantic segmentation model is obtained by two-stage training. The output scoring charts of two channels are obtained by two semantic segmentation models, and the output scoring charts of two channels are fused by different model fusion algorithms, and the optimal model fusionalgorithm is selected according to the specific evaluation index. The test image is segmented according to the semantic segmentation model and the selected optimal model fusion algorithm. The invention can ensure that on the premise that the overall segmentation accuracy of the data set is better, the invention is more sensitive to the small target area existing in the image.

The present invention discloses a deep learning-based insulator identification method. The insulator identification method comprises the steps of pre-processing an aerial image, and secondly, extending the data via the methods, such as the geometric transformation, the contrast enhancement, an analog noise adding method, etc.; acquiring the insulator samples, aiming at the insulators of different types, classifying to acquire; determining a to-be-trained model structure; inputting the samples in the to-be-trained model, and continuously adjusting the weights and the bias parameters by the forward propagation and backward propagation methods, and finally determining an optimal model parameter, based on the trained model, taking a to-be-detected image as an input signal, and by the network multi-layer convolution, pooling and full-connection operations, obtaining a final detection identification result. According to the present invention, by a deep learning method, the insulator characteristics are learned continuously, a learning network model is determined, the different insulators are identified under different background environments, and support is provided for the electric power maintenance decisions.

The invention discloses an intelligent monitoring system and an on-site violation monitoring method for personnel violating rules and regulations on a working site. The system comprises a data storagelibrary and a plurality of video acquisition units arranged on a working site. The video acquisition units are in communication connection with the data storage library. The data storage library comprises a personnel information data unit, a face detection unit, a personnel information matching unit and a violation behavior detection unit, the personnel information data unit is in communication connection with the face detection unit, the face detection unit is in communication connection with the personnel information matching unit, and the personnel information matching unit is in communication connection with the violation behavior detection unit. Also provided is an onsite violation monitoring method. Through the machine vision learning network, different violation behaviors of the operating personnel can be early warned rapidly and effectively, factors such as weak personnel safety awareness, behavior habits and difficulty in traditional video monitoring and recognition can be effectively avoided, potential safety hazards caused by the violation behaviors are effectively prevented, and the safety of site construction is improved.

The invention discloses a wind power short-term prediction method based on a long-term and short-term memory network, comprising a long-term and short-term memory neural network training algorithm, ashort-term wind power prediction error distribution algorithm and a wind turbine generator power short-term prediction model design. A long-term and short-term memory network algorithm (LSTM)-based wind pow prediction model is established based on the depth learning network, and the Gaussian mixture model (GMM) is used to analyze the error distribution characteristics of the short-term wind powerprediction. The invention can obtain different confidence intervals of two units through the GMM model. It is proved that LSTM method has higher precision and faster convergence rate, and GMM method has practical application value for wind power dispatching.

The present invention relates to a system and methodology facilitating automation security in a networked-based industrial controller environment. Various components, systems and methodologies are provided to facilitate varying levels of automation security in accordance with security analysis tools, security validation tools and / or security learning systems. The security analysis tool receives abstract factory models or descriptions for input and generates an output that can include security guidelines, components, topologies, procedures, rules, policies, and the like for deployment in an automation security network. The validation tools are operative in the automation security network, wherein the tools perform security checking and / or auditing functions, for example, to determine if security components are in place and / or in suitable working order. The security learning system monitors / learns network traffic patterns during a learning phase, fires alarms or events based upon detected deviations from the learned patterns, and / or causes other automated actions to occur.

The invention provides a deep learning network training method and device based on artificial intelligence. The method comprises the steps of: when a user initiates a task submission request for a target task via a Web front end, acquiring an operating environment packet indicated by the task submission request from a cluster server operating a deep learning framework, then building a deep learning network conforming to the network configuration indicated by the task submission request by using a network configuration generating tool in the operating environment packet, and generating a task instruction for indicating the cluster server to carry out distributed training on the deep learning network by using the operated deep learning framework according to the training information of the target task, so that the cluster server carries out distributed training on the deep learning network according to the task instruction and sends an execution progress of the distributed training to the Web front end. The method and the device solve the technical problem that the deep learning framework is inconvenient to use in the prior art.

The invention discloses a semi-supervised image instance segmentation method based on stepwise adversarial learning. The semi-supervised image instance segmentation method comprises the steps of 1, constructing Mask R-CNN instance segmentation model; 2, Training FPN in a Mask R-CNN Based on DCGAN; 3, adopting labeled data to perform preliminary training on other modules in the Mask R-CNN; 4, constructing a discriminant convolutional network, and forming an adversarial learning network with the Mask R-CNN, and optimizing parameters of the adversarial learning network through adversarial training; 5, feeding back the output of the discriminative convolutional network to Mask R-CNN for retraining the instance segmentation model; And 6, segmenting the to-be-segmented image by using the instance segmentation model. According to the method, the sample set only partially labeled with the image is used for model training, the workload of sample processing is reduced, and a segmentation model with high precision can be obtained.

The invention discloses a content awareness deep learning network-based remote sensing image super-resolution reconstruction method. A comprehensive measurement index and a calculation method for content complexity of images are proposed; based on this, the sample images are classified by the content complexity; deep GAN models with low, medium and high complexity are built and trained; and according to the content complexity of the to-be-classified super-resolution input images, corresponding networks are selected for performing reconstruction. In order to improve learning performance of a GAN, an optimized loss function definition is given. The method overcomes the over-fitting and under-fitting contradiction ubiquitous in machine learning-based super-resolution reconstruction, and effectively improves the super-resolution reconstruction precision of the remote sensing images.

The invention discloses a remote sensing image ground object classification method based on a depth learning semantic segmentation network. Firstly, pixel-by-pixel labeling is carried out on various ground objects in the remote sensing image, and a remote sensing ground object labeling image database is constructed as a training label. A method for constructing a multi-scale feature map group based on structural feature is designed, the feature group and the original image are combined as the input of the depth learning network, in addition, the invention designs an improved network structureof the full convolution network according to the deeplab algorithm, trains the parameters through convolution and deconvolution, finally overlaps and segments the wide remote sensing images, and combines the classification results to obtain the final wide remote sensing image land object classification results. The invention also discloses an improved network structure of the full convolution network according to the deeplab algorithm. High-resolution remote sensing images can be efficiently and quickly achieved pixel-level classification of various objects, simplifying the complex process oftraditional classification methods, and achieving good segmentation and classification results.

The invention provides an automatic vertical parking system and a parking method based on multi-stage planning and machine learning. The image is collected by a ring-view camera, and an ultrasonic sensor detects obstacle information to judge the position, heading information and the validity information of the storage location relative to the storage location. The automatic parking process is initiated when a suitable size and no vehicle-occupied space is identified. The automatic parking system plans the parking route according to the current self-parking posture and the information of the garage. If necessary, the automatic parking system uses multiple R-S curve is adjusted from the parking position to the proper position, and then the secondary helix parking trajectory is generated according to the learning network. The steering wheel, throttle and brake pedal are controlled electronically for parking and garage entry. The invention utilizes the secondary helix training set and thelearning network to improve the efficiency of the parking process and the adaptability to the path deviation, and combines R-S curve for multi-stage planning, to achieve a very small range of parkingplanning high success rate, a wider range of application, more reliable parking process.