35 results about "Training study" patented technology

The invention provides a knowledge map optimal path inquiry method based on depth reinforcement learning, includes two modules, module 1 and module 2, the first module is an offline training module ofthe knowledge map optimal path model, the second module is the on-line application module of the knowledge map optimal path model, the knowledge map optimal path model offline training module is provided with a depth reinforcement learning component, to present the entity carries on the depth intensification training study, so that that next entity is obtain, as the following entity repeats the train learning of the current entity, so that the optimal path model is obtained.The invention adds the generalization ability of the model and improves the calculation accuracy. The logic structure ofthe invention is clear and the calculation mode is flexible. Especially, the reinforcement learning and the depth learning can be distributed and the operation efficiency is improved.

The invention provides a non-supervision feature extraction method based on a self-coding neural network. According to the method, firstly, training data matrix building is performed; then, each component value of a training data matrix is normalized to a position between [0,1]; next, parameter study is performed to obtain a self-coding neural network model; then, the output of a hidden layer is calculated; features are obtained; finally, the number of optimum hidden layer nerve cells is determined according to a halving value taking method; finally, a structure of the self-coding neural network is determined. In the network training study, the expected output of the self-coding neural network is specified to be equal to the input of the network; through such a study target, the providing of the expected network output by training data is not needed in the training process of the self-coding network training process. The method provided by the invention has the advantages that under the condition of being lack of priori knowledge, an internal rule of the equipment mass state data can be excavated and features can be extracted.

The invention discloses a depth study-based network risk early warning method. The method comprises the following steps of: A1. collecting cyberspace asset risk sample data of a full network segment, and storing the cyberspace asset risk sample data in a database; A2. extracting data from the database, and performing convolutional neural network distributed training study, so as to form an initial risk prediction model; and A3. inputting production data into the risk prediction model, evaluating a risk value of the production data, if the risk value reaches an early warning threshold, alarming. Through adoption of the method and device, security risk evaluation and early warning are performed on multiple target networks or targets without obvious vulnerability, and the security state of one network can be evaluated overall; the response speed is improved, and the risk points can be found rapidly; and the maintenance cost is lowered, and the manpower is saved.

The invention provides a character relation classifying method and device. The method comprises the following steps: acquiring a labeled vector set, a unlabeled vector set and a candidate relation test set; expanding the labeled vector set according to the labeled vector set and the unlabeled vector through robust path similarity measurement; carrying out training study on the expanded labeled vector set by a nonlinear semi-supervised classifying function to obtain a character relation classifying model; and carrying out character relation classification on the candidate relation test set by the character relation classifying model. The labeled vector set is expanded by the robust path similarity measurement, influence of noises and isolated points is eliminated by the effect of robustness, manually annotated corpus is reduced, consumption and waste of manpower resources are reduced, the bottleneck problem of manually labeled corpus when a large amount of labeling is required is solved, a regularized semi-supervised classification algorithm is applied to character relation extraction, and the character relation classification efficiency is improved.

The invention provides a system for special dictation training study related to numbers in a foreign language. The system comprises an obtaining unit capable of obtaining dictation materials, a judging unit conducting judgment according to the dictation material, an initializing unit conducting initialization, a setting unit setting difficulty levels and a storage unit for storing a dictation question bank. The obtaining unit, the judging unit, the initializing unit and the setting unit are connected in sequence. The storage unit is connected with the judging unit. According to the system and method for special dictation training study related to the numbers in the foreign language and based on the current capacity levels of response to the dictation numbers in the foreign language of a learner training is automatically and gradually enhanced from easy dictations to difficult dictations and training is intensified in a concentrated mode.

The invention discloses a gesture recognition system and a gesture recognition method. The gesture recognition system at least comprises an intelligent mobile terminal which is provided with accelerated speed, direction and magnetic field sensors, and used to record sensor data of user gestures and extract features, and recognizes the user gestures according to the features by using a training model obtained in training of a model training platform, and further comprises the model training platform which is used to mark and perform normalization processing on the gesture features of the various user gestures, collected by the intelligent mobile terminal, perform feature training study and obtain the training model. The gesture recognition system and the gesture recognition method can input information through gesture information so as to complete corresponding tasks under the circumstance that a user can not or does not want to input the information through a touch screen or a keyboard, and enables a smart phone user to complete the corresponding tasks by simply waving a hand.

This invention relates to mobile terminal orientation and its programming service method, firstly roughly orientating, which is to orientate the honeycomb net position or the small zone, secondly exactly orientating, which is the manual assistant one key orientation or the small zone exactly orientation or the slightly orientation; wherein the slightly orientation includes the character distilling, comparatively identifying, exactly orientation and training studying; the slightly net is the blue tooth slightly net, the mobile terminal adds the slightly net to obtain the information such as the orientation; service condition, service and the service parameter, wherein the service condition is composed of the special position, time and service; and this invention satisfies the service requirement of the different orientation precision and the mobile communication requirement of the different occasion.

The invention provides an automatic waste empty house site information extraction method based on a remote-sensing image. According to the method, the high-resolution remote sensing image is partitioned through vector cadastre data constraint, a method that the contextual feature of a multi-dimensional space structure serves as the recognition feature of a waste empty house site, and a method that classified recognition conducted through a classifier is adopted, the recognition accuracy of the waste empty house site is effectively improved, and automated extraction of the waste empty house site is achieved. The method mainly includes the steps that (1) the high-resolution remote sensing image and the vector cadastre data of the region to be extracted of the waste empty house site are obtained, the high-resolution remote sensing image is partitioned through the vector cadastre data constraint, and a house site object is extracted; (2) the house site object serves as the father object, partition is conducted continuously, and internal sub objects of the house site and related house and courtyard features of the waste empty house site are extracted to form a multi-dimensional feature space; (3) samples and the classifier are selected to conduct training study of the classifier; (4) by the adoption of the method of classified recognition conducted through the classifier is adopted, a waste empty house site plaque is recognized, and the recognition result is output.

The invention relates to a DMP-based robot workspace adaptive walking control system and method. The control system comprises: a DMP centroid track generator which generates a centroid track of a humanoid robot based on DMP training according to a planning target and feedback information; a DMP foot sole track generator which generates a foot sole track of the humanoid robot based on DMP trainingaccording to the planning target and the feedback information; and a joint mapping module which performs inverse kinematic calculation on the centroid track and the foot sole track, and uses a motionengine to map the adaptability of the workspace to a joint space, so as to achieve the adaptive walking control of the humanoid robot. The DMP centroid track generator and the DMP foot sole track generator are obtained based on the example track training study of the robot working space. Compared with the prior art, the system and the method have the advantages of adjusting the gait and body posture in real time, and enhancing the environmental adaptability of the humanoid robot walking.

The invention provides a VANETs vehicle accident risk prediction model based on AdaBoost-SO (trichotomy Adaboost with SMOTE and One-Hot encoding) and using an SMOTE algorithm and a one-bit effective coding three-point adaptive lifting algorithm. The model can provide the theoretical basis for an ITS (International Traffic System) and driving safety assistance. According to the invention, firstly aresearch data set is filled, samples in the data set are balanced through an SMOTE algorithm, a few types of oversampling are synthesized, and each sample feature is subjected to One-Hot coding, andthen the trachomy Adaboost-SO algorithm is used for train and research the data set to obtain a system model, and finally the traffic data is imported through the VANETs to obtain the probability of vehicle accidents.

The invention relates to a distortion correction method for a large-field-of-view display device, and belongs to the technical field of intelligent information image processing. The method comprises: establishing an artificial neural network with a dual-layer implicit structure and solving a weight and a threshold of each layer of the artificial neural network with the dual-layer implicit structure by utilizing a particle swarm algorithm; taking a value corresponding to a global extreme value as the weight and the threshold of the neural network and substituting the value into the established artificial neural network to perform training study to form an image distortion correction model; and finally inputting distortion image data into the distortion correction model to perform correction to obtain a corrected image. According to the method, the weight and the threshold of the artificial neural network are trained by adopting the particle swarm algorithm to overcome the shortcomings of local minimum, low convergence speed and the like of a conventional artificial neural network; and the method is easy to implement, good in data processing capability, high in correction precision and suitable for distortion correction of the large-field-of-view display device.

The invention discloses an MALDI-TOF-MS-based dead slaughtered meat identification method. The method comprises the following steps: preparing a sample set, wherein the sample set comprises a training sample set, a testing sample set and a to-be-measured sample set; measuring the sample sets by using the MALDI-TOF-MS to obtain a corresponding mass spectrum; establishing a neural work classifier based on the mass spectrum of the training sample set and the testing sample set; classifying the samples in the to-be-measured sample set by applying the neural network classifier to obtain the identification result. The invention also discloses an MALDI-TOF-MS-based dead slaughtered meat identification system. The metabolomic and proteomic difference of dead slaughtered meat and living slaughtered meat is found in the molecular level by virtue of the mass spectrometer, interference of environmental factors and sample factors is reduced maximally through training study and testing of the neutral network classifier, the detection process has high stability, sensitivity and accuracy, the detection result has high reproducibility, and the MALDI-TOF-MS based dead slaughtered meat identification method and the MALDI-TOF-MS based dead slaughtered meat identification system are a scientific, accurate and efficient dead slaughtered meat identification method as well as a scientific, accurate and efficient dead slaughtered meat identification system.

The invention discloses a training method of PSO-BP neural network model for heavy metal content prediction, a storage medium and terminal equipment. The training method includes: measuring content ofheavy metal of sample soil to acquire heavy metal content data of the sample soil; performing spectrum reflectivity treatment on the sample soil to acquire a spectrum reflectivity curve of the samplesoil after being treated; according to the heavy metal content data of the sample soil and the spectrum reflectivity curve of the sample soil after being treated, performing feature waveband selection to acquire feature waveband of the sample soil; building the PSO-BP neural network model; inputting the feature waveband of the sample soil into the PSO-BP neural network model for training study until the PSO-BP neural network model converges. Accuracy of soil heavy metal content estimation and predication through the model trained by the method is improved greatly.

The invention relates to a system for predicting the efficacy of a drug directed at a target to treat a disease, the system comprising a processor configured for: - receiving (602) biomedical documents (214) comprising an identifier of the target and / or of the disease; - specifying (604) an offset time (d) the offset time indicating a time interval ahead of the performing of the prediction; - specifying (606) a time window (706) ending at the begin of the offset time; - extracting (608) a plurality of features (222) selectively from the ones of the received documents published during said timewindow; - providing (610) a classifier (226.3) having been trained on training features (220) extracted from biomedical training documents published within a training time window ending at the beginof the offset time ahead of a moment (OC) the outcome of one or more training studies on training target-disease-pairs was disclosed; - executing (612) the classifier, thereby providing the extractedfeatures as input; - outputting (614) a classification result indicating whether the drug directed at the target can be used to treat the disease.

The invention relates to an electronic mobile training and test system of hygienic knowledge for public service staff. The electronic mobile training and test system of the hygienic knowledge for the public service staff comprises a main test room system and a mobile test system. The main test room system comprises a main server, a first network exchanger and a touch type computer or a desktop computer. The mobile test room system comprises a subordinate server, a second network exchanger and a tablet personal computer. After the mobile test system is applied, a test can be organized by only one person. Testing on main test room and mobile test rooms can be started simultaneously and do not interfere. All test data are automatically collected, organized and undergo statistical analysis. Working efficiency is improved and office resources are saved. Knowledge training study materials for the public service staff can be installed on the electronic mobile training and test system in advance. The mobile test rooms can be converted into mobile classrooms as required at any time and at any places so that the employed staff can be conveniently trained.

The system includes a processor configured for receiving biomedical documents including an identifier of the target and / or of the disease; specifying an offset time the offset time indicating a time interval ahead of the performing of the prediction; specifying a time window ending at the begin of the offset time; extracting a plurality of features selectively from the ones of the received documents published during the time window; providing a classifier having been trained on training features extracted from biomedical training documents published within a training time window ending at the begin of the offset time ahead of a moment the outcome of one or more training studies on training target-disease-pairs was disclosed; executing the classifier, thereby providing the extracted features as input; and outputting a classification result indicating whether the drug directed at the target can be used to treat the disease.

The invention provides a human body motion state recognition method and system based on an acceleration sensor, which is divided into an offline stage and an online stage; wherein, the offline stage adopts the K-Means clustering method to construct a human body motion state recognition model, based on the existing tagged The data is used for training and research, and a classification strategy is proposed; then in the online stage, a real-time system for human motion state recognition is designed based on Android mobile phones, and the design is carried out from five functions: data collection, data processing, motion recognition, model update, and data display; finally, it is proved by experiments The effectiveness of the clustering algorithm, the experimental results show that it is feasible to build a human motion recognition model based on the clustering method, and the model has the advantages of good real-time performance, light weight and easy adjustment.

The invention discloses a training method of PSO-BP neural network model for heavy metal content prediction, a storage medium and terminal equipment. The training method includes: measuring content ofheavy metal of sample soil to acquire heavy metal content data of the sample soil; performing spectrum reflectivity treatment on the sample soil to acquire a spectrum reflectivity curve of the samplesoil after being treated; according to the heavy metal content data of the sample soil and the spectrum reflectivity curve of the sample soil after being treated, performing feature waveband selection to acquire feature waveband of the sample soil; building the PSO-BP neural network model; inputting the feature waveband of the sample soil into the PSO-BP neural network model for training study until the PSO-BP neural network model converges. Accuracy of soil heavy metal content estimation and predication through the model trained by the method is improved greatly.

The invention relates to an electronic mobile training and test system of hygienic knowledge for public service staff. The electronic mobile training and test system of the hygienic knowledge for the public service staff comprises a main test room system and a mobile test system. The main test room system comprises a main server, a first network exchanger and a touch type computer or a desktop computer. The mobile test room system comprises a subordinate server, a second network exchanger and a tablet personal computer. After the mobile test system is applied, a test can be organized by only one person. Testing on main test room and mobile test rooms can be started simultaneously and do not interfere. All test data are automatically collected, organized and undergo statistical analysis. Working efficiency is improved and office resources are saved. Knowledge training study materials for the public service staff can be installed on the electronic mobile training and test system in advance. The mobile test rooms can be converted into mobile classrooms as required at any time and at any places so that the employed staff can be conveniently trained.

The invention discloses a gesture recognition system and a gesture recognition method. The gesture recognition system at least comprises an intelligent mobile terminal which is provided with accelerated speed, direction and magnetic field sensors, and used to record sensor data of user gestures and extract features, and recognizes the user gestures according to the features by using a training model obtained in training of a model training platform, and further comprises the model training platform which is used to mark and perform normalization processing on the gesture features of the various user gestures, collected by the intelligent mobile terminal, perform feature training study and obtain the training model. The gesture recognition system and the gesture recognition method can input information through gesture information so as to complete corresponding tasks under the circumstance that a user can not or does not want to input the information through a touch screen or a keyboard, and enables a smart phone user to complete the corresponding tasks by simply waving a hand.

The invention discloses a semantic understanding technology-based intelligent customer service training research platform, which comprises a data filling unit, a database, a learning unit, a training unit, a training result analysis unit and a correction unit, and is characterized in that the data filling unit is connected with the database, and the training unit is connected with the database; the database is a storage library used for storing all data, the data filling unit is used for filling new data in the database, the learning unit is used for intelligent customer service personnel to learn, and the training unit is used for training after the intelligent customer service personnel learn. The intelligent customer service training research platform based on the high-precision semantic understanding technology disclosed by the invention has the advantages that the intelligent customer service can perform real person reply simulation training, enough buffer time is provided for transferring the real person customer service, and the training result analysis unit is used for performing overall analysis on the trained result. And the working pressure of the real customer service is reduced.

The invention discloses a MALDI-TOF-MS-based identification method for slaughtered meat, comprising: preparing a sample set, the sample set including a training sample set, a test sample set and a sample set to be tested; applying MALDI-TOF-MS to the sample set Measure to obtain the corresponding mass spectrum; establish a neural network classifier based on the mass spectrum of the training sample set and the test sample set; use the neural network classifier to classify the samples in the test sample set to obtain the identification result. The invention also discloses a MALDI-TOF-MS-based identification system for slaughtered meat. The present invention finds the metabolomics and proteomics differences between dead slaughtered meat and live slaughtered meat at the molecular level by means of a mass spectrometer, and minimizes the interference of environmental factors and sample factors through the training, learning and inspection of neural network classifiers. The detection process has high stability, sensitivity and accuracy, and the detection result has good reproducibility. It is a scientific, accurate and efficient dead meat identification method and system.

The invention discloses a method for semi-automatic marking of a three-dimensional (3D) model based on fuzzy K-nearest neighbor. The method comprises the following steps of: firstly establishing a 3D model training base; then extracting characteristic vectors of the input model to be marked, and matching similarity of the model with the model in the training base to find k neighbor models; then classifying by using a fuzzy K-nearest neighbor classifier to obtain a fuzzy classification result; and finally calculating classification uncertainty, and marking the models which are uncertain in classification by a feedback and retraining studying method. The marked models are added into the 3D model training base to further expand the model base. In the method for semi-automatic marking of the 3D model, the fuzzy classification result comprising more information is used; classification labels of the models with uncertain classification are determined by aid of feedback so as to solve overlapping problem of feature clustering edges; and a mechanism for training study is used to further automatically mark.

The invention discloses an information hiding detection method based on local learning, which comprises: constructing a training sample database containing positive and negative samples; for any sample to be detected, searching for K positive and negative samples most similar to it in the sample database Yes, it constitutes a local training set; on the local training set, the training and learning of the classifier is carried out. During the learning process, the constraint of pairing positive and negative samples is added, and the optimal classifier is obtained by using the optimization algorithm; the obtained classifier is used to treat The detection samples are discriminated and classified to obtain a detection result of whether the samples to be detected have undergone information hiding. The present invention makes full use of local learning to better overcome the advantages of large intra-class changes, noise reduction and less need for prior knowledge, improves the effect of information hiding detection, and can be applied to an information hiding detection algorithm analysis system based on pattern recognition middle.

The present invention discloses a malicious behavior identification method, system and storage medium oriented to weighted heterogeneous graphs. The method includes the following steps: constructing an inductive graph neural network model, and the inductive graph neural network model includes a subgraph extraction module, Multiple feature vector generation fusion module and classification learning module; train and learn the inductive graph neural network model, extract subgraphs, learn potential vector representations of nodes in subgraphs, obtain multiple subgraph feature vectors corresponding to subgraphs, and multiple subgraphs Graph feature vector fusion, the node feature vectors obtained by fusion are classified and learned in the classification learning module; the trained inductive graph neural network model is used to identify malicious behaviors. The present invention fully combines and utilizes rich topological feature information and attribute information contained in heterogeneous graphs, and on this basis, designs an inductive learning graph neural network model to complete feature extraction and representation learning in heterogeneous graphs, and finally realize malicious behavior recognition.

The invention discloses a multi-functional rehabilitation training device for rats with brain injury, and relates to the technical field of medical simulation experiments. The main points of the technical scheme are that the multi-functional rehabilitation training device comprises a main box body and a main controller; the main box body is internally provided with an annular plane runway, an annular three-dimensional runway and a static training room which are coaxially arranged from outside to inside; a plurality of dynamic training rooms which are uniformly distributed along the circumferential direction are communicated between the annular plane runway and the annular three-dimensional runway, and a dark light training pipeline is arranged between the static training room and the annular plane runway; and the training path is from the annular plane runway, the dynamic training rooms, the annular three-dimensional runway, the static training room, the dark light training pipeline to the annular plane runway in sequence. Diversified training conditions meeting actual conditions are provided for rat brain injury rehabilitation training, and basic data are provided for brain injury rehabilitation training research after data in the training process are collected.

The invention belongs to the field of daily living articles, and provides a study chair which comprises a chair face, a supporting part, two pairs of chair legs for supporting the chair face, a study desktop and a containing part. The supporting part is perpendicular to the chair face and arranged above one lateral side of the chair face. The containing part is formed among the two pairs of chair legs, and comprises supporting rods connected with the chair legs, and a plurality of connecting rods arranged between the supporting rods. The study desktop is arranged on one side of the chair face and higher than the chair face. By adding the study desktop and the containing part to an ordinary chair, the training study chair is formed; the study desktop and a drawer box can be used for writing and stationery containing, the containing part can be used for containing backpacks and other personal belongings, and the study chair is simple in structure, high in practicability, firm, attractive and durable.

The invention discloses a multifunctional rehabilitation training device for brain-injured rats, which relates to the technical field of medical simulation experiments. The circular plane runway, the circular three-dimensional runway and the static training room are set up. The circular plane runway and the circular three-dimensional runway are connected with multiple dynamic training rooms evenly distributed along the circumferential direction. There is a dark space between the static training room and the circular plane runway. Light training pipeline; the training path is the circular plane track, the dynamic training room, the circular three-dimensional track, the static training room, the dark light training pipeline, and the circular plane track. The invention provides diversified training conditions that meet actual conditions for rat brain injury rehabilitation training, and provides basic data for brain injury rehabilitation training research after collecting data in the training process.

The present invention relates to an abnormal behavior recognition method based on feature extraction of video motion information and error backpropagation network (BP Adaboost) based on adaptive enhancement algorithm, comprising: firstly calculating optical flow according to adjacent image frames of the video, by horizontal direction and The optical flow in the vertical direction is used to calculate the optical flow direction, and the optical flow direction histogram is calculated with the intensity of the optical flow as the weight, and the histogram features are converted into feature attributes with probability attributes, and then training based on the normal and abnormal training samples. The classifier is obtained by adapting the error backpropagation network (BP Adaboost) of the boosting algorithm. In the test phase, before using the trained classification model, the optical flow direction histogram of the test sample is obtained according to the same calculation method as the optical flow histogram of the adjacent frame, and finally the abnormal behavior in the test sample is checked according to the classification model obtained through training and learning. identify. The invention has the characteristics of high recognition rate and low computational complexity, and can be widely used in the fields of abnormal behavior recognition and motion analysis.

The invention discloses a user event relevance calculation method based on content environment enhancement. The method comprises the following steps: lowering the dimensions of archives of a user and social events by using a topic model, and converting the archives into topic distribution; calculating user preference features of the user archives and the social event archives; extracting online and offline social influence features based on a collaborative filtering method by taking the user preference features as probabilities of the fact that the user participates in corresponding events; acquiring local interested topic distribution according to events in a city where the user is located, and comparing the local interested topic distribution with the events to obtain local popularity features; acquiring user event relevance through a training study sequencing model. According to the method, content environment relevant information in an event social network is fully mined to extract the features of user preferences, social influences and local popularity for calculation, and a plurality of features are combined, so that the accuracy of a final result is increased, and the technical problem concerned with recommendation specific to the type of novel objects such as social events in a personalized recommending system is solved.