51results about How to "Rich learning" patented technology

The invention discloses an image quality testing method based on a deep convolutional neural network. The image quality testing method based on the deep convolutional neural network comprises the following steps of firstly, establishing a sample set, establishing a deep convolutional neural network model, training the deep convolutional neural network model under different initial conditions, connecting the optimal deep convolutional neural network models obtained through the multiple times of training in parallel to obtain an image quality testing system, and using the obtained image quality testing system for testing images to be tested. According to the image quality testing method based on the deep convolutional neural network, characteristic learning is carried out by simulating the learning process of human brains, the problem that characteristics are difficult to select in an existing image quality testing method is solved, the contingency of predicting results is reduced, integration is high, generalization ability is strong, and testing effects are good.

The invention discloses a method for diagnosing a fault of a transformer on the basis of a clustering algorithm and a neural network. The method comprises the following steps that (a) the type of the fault is determined according to an IEC standard and a DL / T722-2000 guideline, and the characteristic quantities of a fault sample set are selected from an original sample set; (b) clustering is carried out on samples by utilizing a K-means clustering method; (c) an RBF neural network is established; (d) parameter learning is carried out to determine the number, the center position, the width and the output weight of hidden layers; (e) optimization training is carried out by adopting PSO to determine the positions of the centers of the hidden layers, and the number, the width and the weight of the hidden layers are determined by utilizing a test method, a minimum distance method and a pseudo-inverse method respectively; (f) training samples are input, the posterior probability is solved, and the type of the fault is judged. According to the method for diagnosing the fault of the transformer on the basis of the clustering algorithm and the neural network, the training samples and the test samples can be evenly divided from the total samples, more complete test on the neural network can be carried out by good test samples, and therefore the neural network can be evaluated correctly and reasonably.

The invention provides a VR-based distribution network live working simulation training method, device and system, which relate to the technical field of electric power training simulation. The methodcomprises the following steps: constructing a three-dimensional simulation virtual scene of distribution network live working; When a virtual reality display device and a virtual reality interactiondevice associated with a three-dimensional simulation virtual scene are detected to be worn by an operator, the operator is bound to the virtual worker; Laser scanning the operator based on the presetscanning equipment to obtain the current state information corresponding to a plurality of parts of the operator in a three-dimensional simulation virtual scene and load the current state informationto the virtual operator to keep the virtual operator and the operator in synchronous action; The three-dimensional simulation virtual scene is shot by the preset virtual camera from the third personperspective, and the three-dimensional simulation virtual scene is converted into two-dimensional images to be displayed to the viewer. The invention can effectively improve the effect of live operation training of the distribution network.

The invention relates to a DCF tracking confidence evaluation and classifier updating method based on a neural network, and belongs to the technical field of computer vision. The method comprises thefollowing steps that firstly, a small-scale convolutional neural network of a response graph analysis network is designed and trained; in correlation filtering tracking, after convolution is carried out on a classifier and features of a search area, a generated response graph is input into the network, and output serves as a tracking confidence score of the classifier. When a score is lower than preset low confidence threshold value a target is severely disturbed so that update is stopped, thereby preventing a target model from being polluted, adaptively adjusting the updated learning rate andtime interval by a confidence score, and determining that the appearance of the target is in a highly similar state when a classifier obtains a higher confidence score for continuous multiple frames,thereby improving the updating interval to alleviate an overfitting phenomenon. the adaptability of correlation filtering tracking to interference factors such as illumination change, shielding and visual field can be remarkably enhanced, and the space and time efficiency is improved.

The invention discloses a demonstration system for teaching in visual communication major. The demonstration system comprises a pedestal; a storage box is arranged above the pedestal; a hydraulic cylinder is installed inside the storage box; a first turntable is installed at the output end of the hydraulic cylinder; and a bottom plate is mounted above the first turntable. A writing board is mounted above the bottom plate through a third hinging shaft; a first mounting board is mounted above the bottom plate at the rear side of the writing board; a second mounting plate is installed at the rearside of the first mounting plate by a second hinging shaft; and a display screen is installed on the surface of the second mounting plate. According to the invention, with a demonstration box, a demonstration table, the second turntable, the first mounting plate, the writing board, the display screen, the first turntable, an electric push rod and the hydraulic cylinder structure, problems that the demonstration effect is not comprehensive and the height and angle can not be adjusted conveniently are solved.

A Chinese language studying system is installed on a mobile terminal. The Chinese language studying system comprises a listening module, a speaking module, a reading module, a writing module and a game module. The listening module is used for playing a Chinese language sentence which is searched according to a topic or a searching keyword. The speaking module is used for playing demonstration speech and recorded speech. The reading module is used for playing an article for reading. The writing module is used for providing copying display on Chinese characters which are classified and graded according to character components and the number of strokes. The Chinese language studying system settles restrictions of hardware of computer, traditional teaching mode and personnel pairing in studying of the Chinese language by a foreigner, and achieves targets of comprehensive studying and understanding for the Chinese culture by means of a mobile phone and convenient mobile equipment. The Chinese language studying system realizes comprehensive Chinese language studying by means of mobile equipment of mobile phone, etc. The Chinese language studying system realizes high precision of each mode in aspects of listening, speaking, reading, writing, playing etc. Furthermore a modular and combined scientific studying mode is realized.

A robot working platform for teaching comprises a device fixing platform (10) and a control platform, a rotating platform (11) is arranged in the middle of the device fixing platform (10), two hand-actuated mechanical fixing seats (12) and two utensil suction cup devices (13) are evenly distributed on the rotating platform (11), four device fixing seats (16) are evenly distributed all around the device fixing platform (10), and a robot gripping device, a robot polishing device, a robot drilling device and a robot welding device are fixedly arranged on the four device fixing seats (16) in sequence; the control platform conducts connection control on all devices in the device fixing platform (10), the a robot gripping device, the robot polishing device, the robot drilling device and the robot welding device; the device fixing platform (10) can sequentially rotate by 90 degrees, and therefore the hand-actuated mechanical fixing seats (12) or the utensil suction cup devices (13) sequentially correspond to the positions of the robot polishing device, the robot drilling device and the robot welding device.

The invention belongs to the technical field of dynamic network anomaly detection, and discloses an abnormal group detection method based on weighted dynamic network representation learning, which comprises the following steps: step 1, constructing a weighted dynamic network representation learning model based on a deep self-encoding neural network; 2, performing abnormal link identification basedon the constructed weighted dynamic network representation learning model to obtain an abnormal link set; and 3, constructing a full-connection neural network model based on the abnormal link set, and performing abnormal group detection through the full-connection neural network model. According to the method, the exception link is combined with the full-connection neural network exception detection model, the application range of the method is expanded on the basis of the exception link, experimental verification is carried out on a secure mail data set and an AS-level Internet data set, andexperimental results show that the method has a good exception group detection effect.

The invention discloses a calligraphy teaching system comprising: a font library; a retrieval unit used for receiving a retrieval request of a user and searching out Chinese characters satisfying theretrieval request of the user from the font library; an alternative unit for displaying the Chinese characters satisfying the user's retrieval request for selection by the user; a display unit including a first display module, a second display module, and a third display module, wherein the first display module is used for displaying basic knowledge and evolution process of the target Chinese characters, the second display module is used for displaying stroke features and spatial features of the target Chinese characters and the third display module is used for displaying the writing process of the target Chinese characters; an analysis unit used for analyzing and displaying the evolution process of the target Chinese characters from four aspects including the character shape, stroke, structure and writing method of the target Chinese characters.

The embodiment of the invention discloses a Chinese entity extraction method and device. The method comprises the steps that a target source statement is segmented into clauses; the words in the clause are vectorized to obtain a word vector; a probability matrix of each label corresponding to each word obtained by a long short-term memory network LSTM is determined according to the word vector andthe hierarchical bidirectional long short-term memory network BiLSTM; the probability matrix is input into a CRF model to obtain a label with the maximum probability in the labels corresponding to each word; and an entity consisting of the words corresponding to the label with the maximum probability is extracted. According to the embodiment of the invention, the target source statement is segmented into the clauses, so that subsequent semantic representation in the clauses learned at a word level and inter-clause semantic representation learned at a clause level are facilitated; through theCRF model, the tag with the maximum probability in the tags corresponding to each word is determined, and the Chinese entity composed of the words corresponding to the tag with the maximum probabilityis extracted, so that the Chinese entity identification accuracy is improved.

The invention relates to a pre-school education interaction system based on intelligent image identification. The system comprises a mobile intelligent terminal and a service management center. The mobile intelligent terminal is connected to the service management center through a mobile network in a communication connection mode. The system has advantages that the pre-school education interactionsystem based on the intelligent image identification is provided so that a nature image identification speed is rapid and stable; learning becomes interesting and interactivity is high; simultaneously, operation is simple.

A method of managing a local area communication network comprising at least one access equipment for accessing the network is disclosed. At least one communicating object is connected to the network. In one aspect, the method comprises upon detecting connection of a new communicating object to the network and / or upon detecting installation of new firmware on the at least one communicating object connected to the network, a learning phase involving observing interactions of the communicating object with at least one other equipment of the local area network and / or at least one equipment of a wide area communication network accessible via the access equipment. In addition, at least one security rule associated with the communicating object on the basis of the observed interactions is disclosed.

The invention discloses a user trust relationship prediction method and system based on a graph self-encoding network, and the method comprises the steps: obtaining comment interaction data between users, and constructing a user trust relationship network; extracting an adjacency matrix based on the user trust relationship network, and converting the adjacency matrix into a directed activation propagation adjacency matrix; calculating a symbol network reachable matrix in combination with the symbol network activation propagation adjacency matrix, and performing recursion on a high-order symbolnetwork reachable matrix; taking the reachable matrix of the high-order symbol network as the input of a graph convolution network, and encoding the symbol network by using a spectral domain graph convolution method to obtain a network embedding result; and taking the network embedding result as a code of the symbol network, and performing similarity measurement between nodes in the network by using an inner product decoding mode to obtain a reconstructed symbol network adjacency matrix, namely a user trust relationship network link prediction result. According to the invention, the application of the graph convolution network in the symbol network is realized, and the accuracy of user trust relationship prediction is improved.

The invention discloses a multifunctional education robot, and belongs to the field of education robots, the multifunctional education robot comprises a base, the top of the base is provided with a robot main body, the front side of the robot main body is provided with a functional assembly, the outer side of the robot main body is provided with an anti-falling assembly, and the top of the base is provided with an anti-toppling assembly; and electric universal wheels which are uniformly distributed are fixedly mounted at the bottom of the base. According to the robot, children can be helped to learn in a diversified mode through external data supplement, parent remote supervision and video playing learning, meanwhile, the phenomenon that the robot topples over due to the influence of external factors in the moving process can be reduced, the moving stability is guaranteed, and when the robot is used by the children, the stability of the robot is improved; and the phenomenon that the machine falls and is broken is reduced, and the service life of the whole device is prolonged.

The invention relates to a novel English teaching learning language system comprising a data exchange module connected with a teaching library module electrically. The teaching library module is electrically connected with an operation module; the operation module is electrically connected with a display module; and the display module is electrically connected with a voice module. The teaching library module consists of a knowledge storage module and an examination question storage module. The operation module includes a note module and an evaluation module. The operation module plays knowledge points stored into the teaching library module; the note system can consult key and difficult points and wrong questions to carry out consolidation and leak filling; the data exchange module carries out data interaction with a computer; the teaching library module separates knowledge points from examination questions; the note module is used for new knowledge reviewing and consolidation; and the voice module transforms text information into voice information, so that students can understand and receive knowledge conveniently. Therefore, the system can be operated simply; and on the basis of comprehensive knowledge, learning of students with different English learning levels can be realized.

The invention provides a piano renting system. The piano renting system comprises a piano end, an artistic course popularization and management platform, an after-sales maintenance platform and a charging platform, wherein the piano end is used for playing piano by teachers and students, the piano end is provided with import information of an artistic course promotion management platform, an after-sales maintenance platform and a charging platform, so that the teachers and the students can use mobile phone terminals to enter the artistic course promotion management platform, the after-sales maintenance platform and the charging platform; the after-sale maintenance platform is used for providing piano end maintenance service; the charging platform is used for providing lease service of thepiano end and periodically collecting rent; and the art course promotion management platform is used for providing art courses and promotion art information for students. The problem that students andparents are difficult to burden due to the fact that equipment such as pianos is too expensive can be effectively solved.

The invention relates to a video description method based on space-time super-resolution and electronic equipment, and the method is realized based on a video description model, and comprises the following steps: obtaining an input video, and sampling the input video to obtain a video frame sequence containing a plurality of compression size frames; performing multi-modal feature extraction and feature coding on the video frame sequence through the video description model, dynamically fusing the coded multi-modal features, and gradually decoding to generate a video description statement; wherein when the video description model is trained, a frame with an original resolution and a middle missing frame between adjacent sampling frames are reconstructed from two dimensions of space and time, a loss function is constructed by a reconstruction error and a decoding prediction error, and model training is realized. Compared with the prior art, the method has the advantages of rich and accurate description, strong generalization ability, low calculation overhead and the like.

The invention discloses application of food grade printing ink. The food grade printing ink comprises the following components in percentage by weight: 15-25% of water soluble acrylic resin, 25-40% of purified water, 15-30% of calcium carbonate, 15-25% of food grade pigments, 3-8% of edible alcohol, 1-3% of a surfactant, 0.30-0.42% of an antifoaming agent, 0.4-0.6% of a self-crosslinking additive, 1-3% of a dispersing agent and 1-2% of additives. The food grade printing ink is characterized by being used for child full-filling, particularly paper and plastic paper serve as a full-filling carrier, and then the food grade printing ink is painted. According to the food grade printing ink disclosed by the invention, due to the usage of the good grade pigments and edible alcohol, the prepared printing ink maintains the characteristic of 'being dried on a printed body' so as to be conveniently used by a user, is enriched in safe food grade standards and is particularly suitable for being used on child full-filling books with high safety. The learning and playing fun of the children can be greatly enriched.

The invention discloses a neural network model training method and device and a storage medium, and the method comprises the steps: respectively storing the teacher networks of different time nodes selected in advance as process models in the training process of the teacher networks; integrating the plurality of stored process models to form a new teacher network; and training the student network by using the new teacher network.

The invention discloses a power load prediction method based on a deep neural network. The method comprises the steps of collecting historical load data; processing the historical load data by using aplurality of parallel convolutional neural network CNN components to obtain sequence data; introducing a parallel structure into a deep neural network model DNN; processing the sequence data througha recurrent neural network to obtain a hidden corresponding state list; and bringing other types of features into a prediction model to serve as input of the DNN part of the model, and carrying out load prediction. Different types of factors possibly influencing load consumption are modeled by using different types of neural network components; the plurality of convolutional neural network (CNN) components are used for learning rich feature representation from a historical load sequence, and then a recurrent neural component based on an LSTM is used for modeling variability and dynamics in historical loading such that the purpose of predicting the power load per hour one day ahead is achieved.

The invention provides a breast tumor benign and malignant classification method and device based on an ultrasonic image and a medium, and the method comprises the steps: carrying out the preprocessing of an original breast tumor ultrasonic image with a classification label, and obtaining a preprocessed image; obtaining a region of interest in the preprocessed image; performing three kinds of processing on the region of interest to respectively obtain a deep residual network feature vector, a texture feature vector and a morphological feature vector; carrying out dimension reduction processing on the deep residual network feature vector and the texture feature vector, and then carrying out feature fusion on the deep residual network feature vector and the texture feature vector to obtain fusion vector data; learning a certain amount of fusion vector data by using a support vector machine classifier, learning morphological feature vectors by using a naive Bayes classifier, and obtaining a tumor classification model by weighting the two classifiers; and inputting the breast tumor ultrasonic image to be classified into the breast tumor classification model to obtain a classification result. By adopting the method provided by the invention, automatic classification of the breast tumor ultrasonic images can be realized more accurately.

The invention discloses a user trust relationship network link prediction method and system based on a gating mechanism, and the method comprises: obtaining comment interaction data between users, andconstructing a user trust relationship network; extracting an adjacency matrix based on the user trust relationship network, and converting the adjacency matrix into a directed activation propagationadjacency matrix; calculating a symbol network reachable matrix in combination with the symbol network activation propagation adjacency matrix; processing the symbol network reachable matrix based ona gating mechanism; taking the processed reachable matrix as input of a graph convolutional network to obtain a symbol network for encoding and a network embedding result; and taking the network embedding result as a code of the symbol network, and performing similarity measurement between nodes in the network by using an inner product decoding mode to obtain a reconstructed symbol network adjacency matrix, namely a user trust relationship network link prediction result. An accurate network embedding result is obtained through the graph convolution network, the speed of user trust relationship prediction is increased, and the prediction accuracy is ensured.

The invention relates to an electronic product applied to the eating health aspect, which is used as a family health electronic nutritionist and consists of a software part and a hardware part, wherein the software part comprises relevant contents such as the functions of various kinds of menus, the functions of an electronic nutrition dictionary, healthy eating guide and the like, and the hardware part consists of a case, an input disc, a circuit assembly, a display screen and an interface. A healthy eating inquiry device is in two types: the fixed type and the portable type, the fixed type healthy eating inquiry device is fixed in a kitchen or other places requiring the placement of the healthy eating inquiry device, and the portable type healthy eating inquiry device can be carried about and can realize the hand-held operation. Through the product, people can obtain some knowledge about the healthy eating and the ingredient matching and cooking methods of various dishes and food, so the scientific eating is realized by people, and people can realize the health through eating. The detects that in the prior art, people only can obtain incomplete information singly from books, media and televisions are overcome, various kinds of eating information can be fully, simply and conveniently searched and learned according to requirements, and the healthy eating inquiry device belongs to the treasure chest type indispensable product capable of being used for guiding the health life of people.

The invention relates to an infrared image target detection method and device based on difficult sample transfer learning, belongs to the technical field of image processing, and solves the problem of how to utilize rich natural image data to assist in improving the performance of infrared image target detection. The method comprises the following steps: acquiring a target detection data set and dividing the data set into a training set and a test set, wherein the data set comprises a visible light data set and a corresponding infrared data set; the construction of the difficult sample transfer learning infrared image target detection network comprises the following steps: introducing a domain attention module and a path aggregation network module into a ResNet50 network to form an improved feature extraction network; performing different-source image transfer learning on the infrared image target detection network by using the training set to generate a target detection model; and inputting the to-be-detected visible light image and the to-be-detected infrared image in the test set into the target detection model to obtain a target detection result. Different domains activate different data domains through domain attention and predict targets of different sizes on different scales.

The invention discloses an intellectual property teaching managing system which comprises a student client side, an application server, a teacher client side and a receiving and sending module. The student client side comprises a first control module, a ranking module, a consulting module, an examining module, a teaching material library module, a case library module and an examination library module. The teaching material library module and the case library module each comprise a video unit, a document unit and a one-to-one teaching unit. The application server comprises a second control module, a data storage module and a recommend module. The teacher client side comprises a third control module, a teaching management module and a teaching arrangement module. The intellectual property teaching managing system has the advantages that different teaching arrangements can be formulated according to the conditions of students, learning of teaching materials, cases and examines is arranged, the students more thoroughly and more comprehensively learn related knowledge, three modes of videos, documents and one-to-one learning can meet requirements of different students, and the learningenthusiasm of the students can be stimulated through ranking.

The invention relates to a training method of a neural network model for image recognition, and the method comprises the steps: in a first training stage, calculating a first loss function corresponding to a first group of input image samples, and training the neural network model through the first loss function, wherein the calculation of the first loss function comprises: (1) for each input image sample in the first set of input image samples, calculating a classification item corresponding to a difference between a predicted classification label and a real classification label of the input image sample; and (2) for each input image sample in the first set of input image samples, calculating a reconstruction error term corresponding to a difference between the input image sample and its reconstructed image sample.

The invention discloses ink printing equipment for printing layers with many colours in one process. The ink printing equipment comprises a printed body conveying part and an ink storage part, wherein the ink storage part is located above the printed body conveying part. The ink printing equipment is characterized in that the ink storage part comprises an independent and detachable ink division disc; the ink division disc is divided into a plurality of small bins; inks with different colours are stored in each small bin; and ink layers with different colours are applied on a printed body passing through. According to the ink printing equipment, the independent and detachable ink division disc is additionally arranged on the conveying part of the ink printing equipment, the inks with different colours are stored by virtue of each small bin on the ink division disc, and the plurality of ink layers with different colours are applied on the printed body passing through in one process. The ink printing equipment is simple and reasonable in structure, simple and fast to operate, and beneficial to increasing the coating-printing efficiency; and moreover, the ink layers with different colours can be applied in one process, thus the coating-printing quality can be ensured.

The invention discloses a facial expression recognition method based on a deep privileged network. The method comprises the following steps: inputting a facial expression picture through a main network, and carrying out the preprocessing of the facial expression picture, thereby obtaining a preprocessed facial expression picture; the main network learns facial expression features to obtain facial expression feature information, and then performs emotion classification on facial expression emotions to obtain emotion classification information; obtaining privilege information through the privilege network, performing privilege learning on the loss function, optimizing parameters of the main network, and obtaining an optimized deep privilege network; inputting a tested facial expression picture into the main network, and preprocessing the tested facial expression picture; the deep privileged network after privileged learning is adopted to extract expression features, emotion classification is achieved, and a facial expression recognition result is obtained; according to the method, the face movement unit is used as privilege information, and a traditional deep network is trained to extract expression features beneficial to recognition, so that the accuracy of face emotion recognition is improved.