155results about How to "Fully trained" patented technology

The invention provides an intelligent device failure diagnosis method based on a support vector machine. The method includes the steps that preprocessing operation is conducted on device data; a failure diagnosis case knowledge base is built; failure diagnosis is conducted on the support vector machine; the failure information is obtained, and troubleshooting guide is conducted. By means of the intelligent device failure diagnosis method based on the support vector machine, the failure feature of a device is highlighted to the maximum degree, the situations that the device data are incomplete and imprecise are reduced, the method provides the possibility for building a precise and reliable failure diagnosis model, the problem of aging of the diagnosis model along with the runtime of the device is solved, the misdiagnosis rate of the failure diagnosis model is reduced, and the correct rate and speed of the device failure diagnosis are increased to the maximum degree.

The invention discloses an SVM classification model-based equipment fault diagnosing method. The method comprises the following steps: performing a preprocessing operation on equipment data; constructing a fault diagnosis case knowledge base; performing fault diagnosis on a support vector machine on the basis of an SVM classification model; acquiring fault information and performing maintenance guide. According to the support vector machine-based intelligent equipment fault diagnosis method, the fault features of equipment are highlighted to the maximum degree; the situations that the equipment data are incomplete and inaccurate are reduced; the possibility is provided for constructing an accurate and reliable fault diagnosis model; the problem that the diagnosis model ages along with the operating time of the equipment is solved; the misdiagnosis rate of the fault diagnosis is reduced; the accuracy and the speed of the equipment fault diagnosis are greatly improved.

The invention discloses an advanced persistent threat (APT) attack detection method based on deep belief network-support vector data description. A deep belief network (DBN) is used for feature dimension-reduction and excellent feature vector extraction; and support vector data description (SVDD) is used for the data classification and detection. At a DBN training state, the feature dimension-reduction is performed by using the DBN model after obtaining a standard data set; a low-level restricted Boltzmann machine (RBM) receives simple representation transmitted from the low-level RBM by usingthe high-level RBM so as to learn more abstract and complex representation after performing the initial dimension-reduction, and back propagation of a back propagation (BP) neural network is used forrepeatedly adjusting a weight value until the data with excellent feature is extracted. The data processed by the DBN is divided into a training set and a testing set, and the data set is provided for the SVDD to perform training and identification detection, thereby obtaining the detection result. The attack detection method disclosed by the invention is suitable for the unsupervised attack datadetection with large data size and high-dimension feature, is fit for the APT attack detection and can obtain an excellent detection result.

The invention discloses an automatic license plate identification method based on a deep convolutional neural network. The automatic license plate identification method comprises the steps of: firstly, designing a network structure and an input format of the neural network; adopting random affine transformation to synthesize a training sample, synthesizing a real scene picture and a grey-scale image license plate, adding noise to simulate and generate a large number of license plate images in a real scene; subjecting the neural network to back-propagation training, and training the neural network by adopting a supervised back-propagation algorithm; conducting sliding window searching, positioning a license plate through sliding a window, segmenting a picture and converting the picture into grey-scale images, and standardizing the grey-scale images to the standard input format. The automatic license plate identification method can effectively handle the influence on identification caused by image translation and rotation, can avoid the dependence on the specific environment and font in the identification process, is simple in algorithm implementation and high in robustness, and is easy to transplant.

The invention provides an aerially-photographed vehicle real-time detection method based on deep learning and mainly aims to solve the problem that in the prior art, it is difficult to perform precisedetection on an aerially-photographed vehicle target under a complicated scene on the basis of guaranteeing instantaneity. The method comprises the implementation steps that 1, an aerially-photographed vehicle dataset is constructed; 2, a multi-scale feature fusion module is designed, and a RefineDet real-time target detection network based on deep learning is optimized in combination with the module, so that an aerially-photographed vehicle real-time detection network is obtained; 3, a cross entropy loss function and a focus loss function are utilized to train the aerially-photographed vehicle real-time detection network in sequence; and 4, a trained detection model is used to detect a vehicle in a to-be-detected aerially-photographed vehicle video. According to the method, the designedmulti-scale feature fusion module can effectively increase the information utilization rate of the aerially-photographed vehicle target, meanwhile, the aerially-photographed vehicle dataset can be trained more sufficiently by use of the two loss functions, and therefore the detection accuracy of the aerially-photographed vehicle target under the complicated scene is improved.

The invention discloses an SAR image target detection method based on a full convolutional neural network. The method mainly solves the problem of low accuracy and slow detection speed in the prior art, and is characterized by obtaining an SAR image; expanding a training dataset; constructing a nine-layer full convolutional neural network; training the full convolutional neural network through the expanded training dataset; inputting a test image into a trained model for significance test to obtain an output significance feature graph; carrying out morphological processing on the significance feature graph; carrying out connected domain labeling on the processed feature graph; with the mass center of each connected domain being the center, extracting a detection slice corresponding to each target mass center; and labeling each detection slice in the input original SAR image and obtaining a target test result of the test data. The full convolutional neural network is applied to SAR image target detection, thereby improving SAR image target detection speed and accuracy; and the method can also be used for object identification.

The invention discloses an age identification method based on an integrated convolution neural network. The method includes following steps: S1, obtaining and expanding training sub-sets in an age identification training database, obtaining the expanded training sub-sets, and selecting M convolution neural network classifiers obtained by training of the expanded training sub-sets as base classifiers; S2, obtaining a to-be-tested face image; and S3, inputting the to-be-tested face image into M base classifiers obtained in step S1 during a test, fusing age categories output by M base classifiers, and obtaining a final age category. According to the method, the accuracy of age identification is high, the dependency on people by age feature extraction of the face image is reduced, ages of a variety of people can be estimated, and the application is wide.

The invention relates to a multi-agent cross-modal depth deterministic strategy gradient training method based on image input. Firstly, a mechanical arm training environment in a simulation platform is constructed; Then two director intelligent bodies and a student intelligent body which are input by utilizing different modalities are constructed; Secondly, based on a depth deterministic strategygradient algorithm, an actor module and a critic module of a director and an actor module of a learner are trained, and finally a cross-modal depth reinforcement learning mechanical arm training algorithm based on image input is achieved; When the overall training is finished; a mob actor network can be used only; high-dimensional image input is received; the action capable of completing the taskis output; Moreover, the method is very suitable for being migrated to a real environment, and since the real environment cannot provide full-state modal information and the image modal information isrelatively easy to obtain, after an actor network of a mob is trained, the demand of the full-state modal information can be abandoned, and a relatively good output strategy can be obtained by directly utilizing image input.

The invention discloses a neural network assisted integrated navigation method for an underwater vehicle. The neural network assisted integrated navigation method is implemented by use of strapdown inertial navigation system (SINS), a Doppler velocity log (DVL), a magnetic compass pilot (MCP) and a terrain aided navigation system (TAN), wherein the integrated navigation is completed by use of a decentralized filter structure of Kalman filtering and a fault-tolerant method, assisted by a radial basis function neutral network (RBFNN). In a fault-free time period, RBFNN is in an online learning model, the observed quantity difference between the SINS and each auxiliary system is taken as the expected output of the RBFNN, and the output fb of an accelerometer after error compensation and the output shown in the specification of a gyroscope are taken as the inputs of the RBFNN; when a sub-system composed of the SINS serving as a reference system and each auxiliary system is out of order, an RBFNN prediction mode is immediately activated, and the predicted output is taken as the measurement input of a corresponding sub-filter. Compared with the SINS mode out of order, the RBFNN mode has the advantages that the navigation accuracy is improved; especially when the fault recovery time is relatively long, the improvement of the navigation accuracy of the RBFNN mode is particularly obvious.

The invention relates to a knee joint rehabilitation device, comprising a base, a motor, a thigh connecting rod, a calf connecting rod, a guide rail, a first shaft and a second shaft. The thigh connecting rod, the calf connecting rod and an electric motor are arranged in pairs, the motor drives the thigh and calf connecting rods, a tie is arranged between the thigh connecting rods, one end of thethigh connecting rod is fixedly installed on the first shaft through a sleeve, two ends of the first shaft are provided with bearings, the first shaft is installed on the base through bearings, two sides of the base and corresponding positions of the first shaft are provided with bearing seats, and the other end of the thigh connecting rod and the lower leg connecting rod are connected together through a connecting rod pin. One end of the calf connecting rod is connected with the head of the base through a second shaft. At both end of that second shaft, rollers are arranged, and the rollers slide on the guide rails. The rehabilitation device of the invention is simple and portable, and easy to operate. The invention also provides a rehabilitation training method, which can train the knee joint to rotate in multiple directions, so as to obtain more sufficient exercise and promote the therapeutic recovery effect.

The invention discloses a real-time network flow anomaly detection method based on big data, which comprises the following steps: S1, obtaining collected and analyzed historical flow data with attacktags stored in a database to obtain attack types; S2, performing data feature preprocessing on the historical traffic data in the S1, and constructing a first class of feature vectors; S3, constructing a clustering model based on the first class of feature vectors in the S2, and obtaining a target model meeting a preset condition by utilizing model evaluation and optimization; S4, storing the target model obtained in S3 and deploying the target model online; S5, capturing and collecting real-time network data flow packet information transmitted in a local area network; S6, performing data feature preprocessing on the real-time network data traffic packet in S5, and constructing a second type of feature vectors; and S7, according to the target model in the S3 and the second type of featurevectors in the S6, performing real-time online analysis and detection, and judging whether the current real-time network data traffic is abnormal traffic or normal traffic.

The invention relates to a trademark image retrieval model training method, which comprises the following steps: obtaining a plurality of groups of sample data, and selecting a most difficult positiveexample sample and a plurality of difficult negative example samples for each query sample according to similarity; taking one query sample, the corresponding most difficult positive example sample and the plurality of corresponding difficult negative example samples as a group of training data, and performing trademark image retrieval model training by utilizing a neural network according to theplurality of groups of training data; and updating the trademark image retrieval model according to the multi-negative example comparison loss function until the verification effect of the trademarkimage retrieval model on the verification set is not improved any more, and ending the training. According to the method, easy samples are removed and difficult-to-divide samples are mined according to the similarity, a small number of difficult-to-divide samples are fully utilized, neural network parameters are adjusted in a more targeted mode, model convergence / over-fitting can be well delayed,training is more sufficient, and the effect is better. The invention further relates to a trademark image retrieval model training system, a storage medium and a computer device.

The invention discloses an SAR image object classification method based on distribution and structure matching GAN. The method includes selecting real data in a training set and pseudo data generatedby the generator to train a discriminator in DSM-ACGAN, updating its parameters; fixing discriminator parameters, generatinga generator in a pseudo-data training DSM-ACGAN again and updating its parameters; calculating the difference in distribution and structural characteristics between generated data and real data, and guiding the feature learning in DSM-ACGAN training and discriminator as sample weights; the trained discriminator is used to predict the test SAR image and calculate the classification index. At the same time, the invention integrates the statistics of the real SAR image and the image characteristics into the generated antagonism network as a discriminant a priori, effectively realizes the discriminant feature learning, and remarkably improves the classification performance.

A multifunctional leg training machine includes a framework, a weight-bearing mechanism mounted at said framework and movable in direction toward or away from said framework and adapted to impart a downward pressure to the shoulders of the user using the machine, allowing the user to perform a squat exercise, a lunge exercise, a calf-raise exercise, a simple transverse body movement exercise, or a mixed training exercise of moving the body transversely and upwardly to fully train different muscle groups of the legs.

The invention discloses a large-scale remote sensing image content retrieval method based on deep adversarial Hash learning, and the method comprises the steps: firstly, building a remote sensing image library, and selecting a plurality of remote sensing images; using the constructed training sample to train a deep adversarial Hash learning model; performing Hash coding on the whole remote sensingimage library by using the trained adversarial Hash coding model to obtain a Hash database; after normalization processing is conducted on the query image input by the user, Hash coding is conductedthrough the trained confrontation Hash coding model, and Hash codes of the query image are obtained; calculating the similar matching distances between the Hash codes of the query images and all samples in a Hash database, returning the image indexes required by the user according to the matching distances from small to large, finding the corresponding images in the remote sensing image library according to the indexes, and completing the image retrieval. The method has the advantages of being high in retrieval precision, small in quantization loss and more efficient in Hash coding.

The invention belongs to the technical field of image classification and transfer learning, discloses a transfer learning method based on paired sample matching, and realizes mining of internal relations of samples based on different domains. The method specifically comprises the following steps of (1) data preprocessing, (2) double-chain model construction based on transfer learning, (3) instancenormalization and batch normalization, and (4) calculation of comparison loss and maximum mean distance loss. The method has the advantages that instance normalization and batch normalization are combined for learning at the same time, styles and semantic association characteristics of different images are fully mined, and efficient recognition of a small number of target domain samples under theassistance of a source domain is achieved.

The invention discloses a sparse backward propagation compression method of a neural network model, belongs to the field of information technology, and relates to machine learning and deep learning technologies. In the process of backward propagation, each layer of the neural network model uses the output gradient of the previous layer as the input to calculate the gradient, and performs k large-value sparse processing to obtain the sparsely processed vector and the number of sparse return times, and record k The index corresponding to the value; use the sparse gradient to update the parameters of the neural network; according to the k-value subscript index, delete the neuron with a small number of return times, and compress the model. The present invention adopts a sparse method based on a large value of k in the backward propagation process, eliminates inactive neurons, compresses the size of the model, improves the training and reasoning speed of the deep neural network, and maintains good precision.

A multifunctional leg training machine includes a framework, a weight-bearing mechanism mounted at said framework and movable in direction toward or away from said framework and adapted to impart a downward pressure to the shoulders of the user using the machine, allowing the user to perform a squat exercise, a lunge exercise, a calf-raise exercise, a simple transverse body movement exercise, or a mixed training exercise of moving the body transversely and vertically to fully train different muscle groups of the legs.

The invention provides a Chinese semantic matching method based on pinyin and BERT embedding. The method comprises the following steps: constructing a semantic matching model comprising a data preprocessing module, a BERT embedded layer module, a pooling layer module and a classifier module, and training the semantic matching model so as to carry out Chinese semantic matching on a to-be-matched statement by utilizing the trained semantic matching model; enabling the data preprocessing module to perform pinyin conversion and pinyin segmentation on each character in the two to-be-matched Chinesesentences to obtain a corresponding pinyin sequence; enabling the BERT embedded layer module to generate an embedded vector for each pinyin according to the context of the obtained pinyin sequence toobtain an embedded vector sequence; enabling the pooling layer module to aggregate the embedded vector sequence into a one-dimensional semantic representation vector for classification; and enablingthe classifier module to perform classification according to the one-dimensional semantic representation vector to obtain a prediction result corresponding to the semantic relationship between the twoChinese sentences. According to the method, the data volume required by pre-training can be greatly reduced, and a relatively good effect is ensured.

The invention provides a medical image segmentation method based on a lightweight full convolutional neural network. The method comprises the following steps: carrying out preprocessing such as graying, normalization, contrast limited adaptive histogram equalization (CLAHE) and gamma correction on a data set; randomly extracting patches from the training set and sequentially extracting patch graphs from the test set to complete data enhancement; building a full convolutional neural network architecture composed of a contraction path (left side) and an expansion path (right side), and designinga left-one-out training method for a data set with a small number of images; and finally, completing BN channel model cutting through channel sparse regularization training, cutting channels of whichscaling factors are smaller than a set threshold, finely adjusting the cut network to obtain a lightweight full convolutional neural network, and inputting test data into the network for rapid test to complete image segmentation. The lightweight full convolutional neural network not only ensures the advantage of high segmentation precision of the deep network, but also improves the test speed ofthe image segmentation network.

The invention relates to the technical field of energy consumption, in particular to an integrated energy predicting method, namely a consumed energy predicting method. The method includes selecting regional factor historical data and energy consumption requirement actual value as data samples according to particular years; combining with the acquired data, excavating relationships of different years and types, and searching for sample weights of regional factor samples corresponding to an energy consumption requirement actual value so as to determine the affecting level of regional factors of different years on energy consumption requirements; adopting the prediction algorithm based on linear mapping to predict a total annual consumption requirement value of one region of one year. The relationships of factors and prediction results can be represented objectively, the efficiency of the algorithm is improved, the energy consumption predicting process is more effective and rapid, the energy prediction accuracy can be improved on the economic development fresh normalcy and energy environment strong constraint conditions, the regional energy balance is calculated, and the reasonable and feasible energy development and safety guarantee policy can be determined finally.

The invention discloses a multi-level image compression method using Transform, which is characterized in that a multi-level image compression frame is mainly composed of a Transform module and is supplemented by a convolutional layer neural network, the Transform module comprises multiple layers of encoder components and decoder components, the encoder components are adopted at an encoding end, and the decoder components are adopted at a decoding end; the decoder has a cross attention mechanism, and the cross attention mechanism carries out joint calculation on the self attention features input by the decoder and the self attention features of the encoder, and makes full use of the features learned by the encoding end of the compression frame encoder. According to the method, a decoder component and a cross attention mechanism thereof in Transform are reserved, the method is applied to a decoding end to realize full utilization of features learned by a coding end, and a better effect is achieved. And the requirement of the framework on hardware is smaller.

The invention discloses a hyperspectral open set classification method based on self-supervised learning and multi-task learning, which mainly solves the problem of low classification precision caused by the fact that an existing hyperspectral open set classification method cannot fully utilize unlabeled samples of a hyperspectral open set, and the implementation scheme of the hyperspectral open set classification method comprises the following steps: inputting a hyperspectral image and preprocessing the hyperspectral image; performing neighborhood block taking on the preprocessed image to generate a training data set and a test data set; constructing a neural network model based on self-supervised learning and multi-task learning; training the constructed neural network model by utilizing the training data set and adopting a self-supervised learning method and a multi-task learning method; and inputting the test data set into the trained neural network model to obtain a classification result. According to the method, label-free sample information can be fully utilized, the problem that label samples are few is solved, the classification precision is improved, and the method can be applied to environment monitoring, resource exploration, urban planning and agricultural planning.

The invention provides a social network cross-media search method based on adversarial learning and semantic similarity. A text and image feature extraction network, a public semantic space mapping network, a semantic similarity network and a modal discrimination network are included. The method has outstanding innovativeness, and is mainly used in social network cross-media search. The method isapplied to the field of image and text processing, cross-media data in different modes can be processed, and retrieval between the cross-media data is efficient and accurate.

The invention discloses a website error-reporting screenshot classification method based on feature fusion. The method comprises the following steps: firstly, carrying out data enhancement on an imagedata set of error-reporting screenshots; zooming the image data to a uniform size, and randomly dividing the image data into a training set, a verification set and a test set; performing feature extraction on the image by using a part of network layer of the VGG16 convolutional neural network; extracting features of the image by using a scale-invariant feature transformation operator; fusing thetwo features through feature splicing to serve as final features of the image; and enabling the final features of the image to pass through a full connection layer, a Dropout layer and a Softmax layerto realize correct classification of error-reporting screenshots. According to the invention, machine learning is used to train the neural network for image classification, the workload of customer service staff is reduced, and the enterprise operation efficiency is improved; the data set is expanded by performing data enhancement on the data set image, so that the training is more sufficient; and the two image features are fused to obtain better classification accuracy.

The invention provides a data balancing strategy and multi-feature fusion-based image labeling method. The method comprises the following steps of: 1, carrying out semantic grouping on training imagesto obtain semantic groups; 2, extending the semantic groups by adoption of a data balancing strategy; 3, inputting the training images into a trained deep convolutional neural network to obtain a deep feature of each image in each semantic group; 4, calculating multi-scale fusion feature of each image in each semantic group; 5, carrying out multi-feature fusion on the multi-scale fusion feature and the deep feature so as to obtain a fused feature of each image in each semantic group; 6, extracting a shallow feature and a deep feature of a to-be-tested image and carrying out feature fusion toobtain a fused feature of the to-be-tested image; and 7, calculating a visual similarity between the fused feature of the to-be-tested image and the fused feature of each image in each semantic group,and sorting the visual similarities to obtain an image labeling result and then obtain a category label. According to the method, the problems that training set images are unbalanced and the single feature expression ability is not strong are solved.

The invention discloses a cloud mobile terminal cooperative fault early warning method, and a related device and system. User data information is monitored through a mobile terminal, the cloud terminal receives the data sent by the mobile terminal, the data sent by the mobile terminal are processed, a fault early warning model is obtained and sent to the mobile terminal, and the mobile terminal obtains the fault early warning model for fault early warning. According to the method, the high computing power and the large storage space of the cloud are utilized to effectively solve the requirements of limited resources of the user mobile terminal and the computing power and the storage space in the training process of the machine learning model. Meanwhile, the cloud end can provide a large amount of user data, the problem that the data size of a single user mobile terminal is insufficient is solved, the early warning model is fully trained, and the accuracy of fault early warning can be effectively improved by training one early warning model for each type of mobile terminal.

The invention provides a SAR image classification method and device based on a hierarchical automatic encoder, the method comprising the steps of: amplifying a sample of SAR images through a generalized regularized automatic encoder model; establishing a hierarchical automatic encoder network model according to the generalized regularized automatic encoder model; inputting the amplified sample ofthe SAR images into the hierarchical automatic encoder network model so that the hierarchical automatic encoder network model encodes the input samples; inputting the feature codes output from the hierarchical automatic encoder network model to a classifier, and classifying the amplified sample of the SAR images by the classifier. The technical scheme of the invention realizes the full training ofthe depth network through the amplification of the SAR image samples and improves the accuracy of the SAR image classification. By establishing a hierarchical automatic encoder network model to classify the input samples, the adaptability of SAR image samples is improved.

The invention relates to an image recognition model generation method and device, computer equipment and a storage medium. The method comprises the steps of obtaining a sample image set; wherein the sample image set comprises a plurality of sample image subsets of which the number of images is sequentially decreased, and the plurality of sample image subsets comprise the same number of image categories; training a to-be-trained image recognition model according to the sample image set to obtain a loss value of the to-be-trained image recognition model; wherein the to-be-trained image recognition model comprises a plurality of branch neural networks; wherein the loss value comprises a target classification loss value and a classification loss value, the target classification loss value is aloss value of the model for the sample image set, and the classification loss value is a loss value of the branch neural network for the corresponding sample image subset; and adjusting model parameters according to the loss value until the loss value is lower than a preset threshold. According to the invention, the image types with a small number of images in the training process can be fully trained, and the effect of image recognition model generation is improved.