46results about How to "Embody relevance" patented technology

Embodiments of the invention provide a human face feature point locating method and apparatus. The method comprises the steps of obtaining a human face image; and processing the human face image through a deep neural network to obtain position information of human face feature points in the human face image, wherein the deep neural network is a network model obtained by training according to humanface samples, and the human face samples comprise human face sample images under multiple backgrounds and poses and the position information of the human face feature points corresponding to the human face sample images. Through the method and the apparatus, the accuracy of locating the human face feature points under complex backgrounds and multiple poses can be improved.

The invention relates to an ML-kNN (machine learning-k-nearest neighbor) improving method and an ML-kNN improving system applicable to multi-label classification. The ML-kNN improving method includes counting the sum of samples of each class of labels in original data sets, utilizing the sum of the samples of each class of labels in the original data sets as a label sample number, counting the sum of samples in each class of features in the samples of each class of labels, utilizing the sum of the samples of each class of features in the samples of each class of labels as a feature sample number and computing feature label weights according to the label sample numbers and the feature sample numbers; splitting each sample in the initial data sets into a plurality of original single-label samples with single labels, and updating feature values of each original single-label sample according to the feature label weights to generate first data sets; acquiring to-be-measured samples to be predicted, splitting the to-be-measured samples into to-be-measured single-label samples with single labels, sequentially predicting the labels of the to-be-measured single-label samples according to the first data sets and determining label sets of the to-be-measured samples. Each feature corresponds to the single corresponding feature value. The ML-kNN improving method and the ML-kNN improving system have the advantage that accurate prediction results of the samples in the aspect of multi-label classification can be obtained.

The invention relates to an optimal operation method for a multi-energy complementary system, which includes the following steps: S1, carrying out interval description on distributed energy and cooling-heating-power equipment based on the structure of a multi-energy complementary system; S2, establishing a system coupling model according to the interval description result; S3, establishing a multi-objective function considering environmental, economic and energy factors, wherein the multi-objective function includes the operation cost and the pollution emission cost; S4, acquiring constraintsused to ensure interval balance, wherein the constraints include electric power balance, cooling-heating-power equipment balance and grid interaction conversion constraints; and S5, solving the systemcoupling model based on the interval theory and affine operation to obtain a final solution set. Compared with the prior art, coupling modeling is carried out on the energy devices of the system, andthe model is solved based on the interval theory and affine operation, which can reduce the dynamic impact of prediction error and energy fluctuation on the system, realize optimal allocation of resources and improve the energy utilization efficiency of the multi-energy complementary system.

The present invention discloses a target data identification method and apparatus. The method comprises the steps of calling a preset information training model including association relationships between sample data identification and sample information templates; marking target data identification on target data in target information according to the information training model, and obtaining target information templates; and identifying the to-be-identified target data in the target information according to the target information templates. The target information templates can be automatically generated, so efficiency of marking the data identification is improved.

The invention discloses a multi-modal label recommendation model construction method and device of a multi-level attention mechanism. The method comprises steps of extracting features of the image, performing bilinear fusion on the features of the image by using an outer product, obtaining an attention factor of each region in the image through an attention network layer, and performing element-by-element product on the attention factors and the original features to obtain final image feature expression; performing word embedding on the text, extracting text features by using a Bi-LSTM network, and then multiplying the text features by an attention network layer to obtain final text information expression, then, fusing the image features and the text features through a bilinear fusion layer, inputting the fused features into a high-level attention layer to obtain final joint feature expression, and finally, sending the final joint feature expression into a classification layer for label classification and recommendation. Under the condition of multi-modal information processing, the recommendation accuracy is improved through the method of combining the hierarchical attention mechanism.

The invention discloses a multi-target tracking method based on a depth condition random field model. The method comprises the steps of obtaining a multi-target tracking data set, carrying out data association on detection responses of any two continuous frames in all frames of the input video sequence; track sheet, generating a vertex according to the time relationship between any two track pieces in the obtained track piece set; calculating the matching degree between two track pieces corresponding to each vertex in the vertex set; determining a difficult vertex pair set according to a timerelationship and a position relationship between any two vertexes in the vertex set; and obtaining the appearance characteristics and position information of each difficult vertex pair in the obtaineddifficult vertex pair set, combining the appearance characteristics and motion characteristics of the difficult vertex pair into a difficult vertex pair characteristic vector, and inputting each difficult vertex pair characteristic vector into an LSTM network. According to the method, the correlation of real data in the multi-target tracking process can be effectively embodied, and the accuracy of a tracking result is high.

The invention provides an urban road dynamic traffic network structure information system. According to the system, characteristics of division of a road network are taken into consideration, the whole traffic road network is divided into multiple network subareas on the basis of the unique characteristics of the network, structure information of the different network subareas is calculated through an information entropy formula of Shannon by utilizing the network, the number of edges of the respective subareas and the weight-containing degree of access, the degree of access weight between onesubarea and the adjacent subarea is added by taking consideration of connections among the subareas, and then structure information containing subarea boundary characteristics is calculated. Therefore, dynamic analysis is carried out on an urban road network on the basis of the subarea structure information and the structure information containing subarea boundary characteristics.

The invention discloses a method for generating an unexpected pattern. The method comprises the steps that user data of a plurality of users is obtained; according to the obtained user data, a user relationship diagram is built; in the built user relationship diagram, at least one kind of associated features of each personnel node are extracted according to each personnel node and the attribute values of the other nodes which are connected with the personnel node through sides; for each kind of associated features, the exception reference value of the associated features is computed according to the associated features marked as the associated features of the personnel node without malicious behaviors; according to the associated features marked as the associated features of the personnel node without malicious behaviors, the non-exception reference value of the associated features is computed; according to the computed exception reference value and the computed non-exception reference value of the at least one kind of associated features, the unexpected pattern is generated. The invention further discloses a method for determining whether users have the malicious behaviors, and a corresponding computing device and a computer readable storage medium.

The invention discloses a method for detecting S7 protocol abnormal communication behaviors based on PSO-SVM. The method comprises the steps of obtaining connection from an industrial control network,the connection comprising a plurality of S7 protocol communication data packets; analyzing each S7 protocol communication data packet to obtain a function code or a sub-function code corresponding tothe S7 protocol communication data packet; and forming a function code sequence corresponding to each connection by a plurality of function codes and sub-function codes corresponding to all S7 protocol communication data packets included in each connection, and inputting the function code sequence corresponding to the connection into the trained S7 protocol anomaly detection model to obtain a detection result of the connection. According to the method, the technical problems that in an existing abnormal communication behavior recognition method, the S7 protocol abnormal communication behaviorin the industrial control network cannot be detected, and the recognition rate is low due to the fact that the relevance among the multiple data packets in the same connection is not considered can be solved.

The present invention discloses a deployment scheme of a wireless multi-hop network buffer queue. The deployment scheme is applied to a train car WiFi system, a train central deployment gateway node and train and ground communication equipment are used for the communication with a ground base station BS, each car is provided with one AP for bearing the access needs of multiple users in the car, the cars are wirelessly connected by using a wireless relay, and the AP in the car and a car relay are connected in a wired way. By the setting problem of the buffer queue length of each hop of data transmission in a wireless multiple-hop network, under the premise that a system satisfies a maximum end to end delay, the system packet loss rate is reduced as much as possible, and thus the balance between the system resource configuration and system acceptable end to end QoE is achieved.

The invention discloses a context-based dialogue generation method and system. The method comprises the steps of obtaining a current dialogue statement, a previous statement of a current dialogue anda next statement of the current dialogue; and inputting the current dialogue statement, the previous statement of the current dialogue and the next statement of the current dialogue into a pre-trainedcodec, and outputting a predicted dialogue statement of the current dialogue statement by the codec. According to the method, the method can play a role in linking the previous part and the next part; the contextual information related to the current dialogue can be combined to generate more effective information, on one hand, the dialogue can be smoother, and the consistency and logicality of the dialogue before and after the dialogue are kept; on the other hand, more premise input information can be provided for the neural network, and research shows that the rich input information is beneficial for the neural network to generate more meaningful answers.

The invention relates to a multi-scale video expressing and browsing method. The steps of the method comprise: carrying out preprocessing to videos to be processed at an input end, and users can carry out gesture input and /or noting to the video to be processed. The preprocessed videos load following expressing forms according to orders or breaking timing restriction, and carry out multi-scale expression. An output end chooses the mentioned one or more ways to browse the videos according to the gesture inputting of the users, and outputs browsing results. The multi-scale video expressing and browsing method provides a plurality of different ways to carry out expression and browsing to the users, the limitation of a single expression mode is effectively overcome, the contents from coarse dimensions to finer dimensions can be expressed better according to the requirements of the users, and the users can be helped to understand and locating the contents of the videos. Meanwhile clustering and layering methods relative to the contents are provided, and therefore expressions of the content of different levels are supplied to the users, and different browsing requirements of the users can be met better.

The invention discloses a software requirement priority setting method based on correlation relationship. The method comprises following steps: S1. establishing correlated sets and functional structure diagrams according to correlation among demands; establishing functional structure directed graphs by means of the influence relationship and determining whether there is a loop in the functional structure directed graphs; if there is a loop, removing the loop by performing model elaboration on the functional structure directed graphs to obtain functional structure directed graphs without loops;S2. using the ISM method and calculating according to the functional structure directed graphs without loops to establish a demand layer model and obtain multistage hierarchical directed graph; S3. generating a correlated demand set of each element in the highest layer of the multistage hierarchical directed graph; performing.

The invention provides an alarm method and device, and a storage medium. The alarm method comprises the following steps: obtaining an alarm network element that meets an alarm threshold condition according to network element data and the alarm threshold condition of each network element, wherein the alarm threshold condition is used for determining whether the network element data of the network element is abnormal or not; obtaining a location of the alarm network element in an electronic map according to an actual location of the alarm network element and a corresponding relationship betweenthe actual location and the location in the electronic map; obtaining an icon attribute of the alarm network element according to the attribute of the network element data of the alarm network elementand a corresponding relationship between the attribute of the network element data and the attribute of the icon; and displaying an icon with the icon attribute of the alarm network element at the location of the alarm network element in the electronic map. According to the invention, the alarm is displayed on the electronic map in the form of an icon according to the corresponding relationship between the attribute of the network element data and the icon attribute; therefore, the processing efficiency is improved, and the alarm reflects the relevance of each region.

The invention discloses a task multi-level processing method, system and device and a storage medium, and the method comprises the steps: obtaining a total task, carrying out decomposition of the total task according to a preset rule to acquire subtasks and execution objects; after the subtasks are distributed to the corresponding execution objects, enabling the execution object to execute the tasks according to the content of the received subtasks; and after the task is completed, returning the task data to the superior execution object for auditing. The system comprises a decomposition module, an execution module and a feedback module. The device comprises a memory and a processor used for executing the task multi-level processing method. The task processing difficulty of a user can bereduced. The task multi-level processing method, system and device and the storage medium can be widely applied to the field of task processing.

The invention discloses a correlation-based feature point elimination method. The method is used for visually guiding an assembling/grabbing process and pre-acquiring standard coordinates and theoretical coordinates; acquiring measurement coordinates, and utilizing a first projection error; judging whether wrong feature points need to be removed or not, wherein the method for searching the wrong feature points comprises the following steps: sorting the plurality of feature points; removing the ith feature point in sequence according to the sequence, calculating a standard reprojection error array corresponding to the ith feature point, and measuring the reprojection error array; calculating a correlation coefficient between the standard reprojection error array corresponding to the ith feature point and the measurement reprojection error array; finding an abnormal point from a plurality of correlation coefficients, and then finding an error feature point from the abnormal point; The method is advantaged in that two arrays obtained through the method are high in relevancy, and amplitudes of all elements in the set are close; therefore, accuracy of the remaining feature points is guaranteed, and assembly/grabbing accuracy is improved.

The invention discloses an elevator fault detection method based on gating circulation networks and canonical correlation analysis. The method is a vertical elevator fault detection method based on gating circulation unit neural networks and canonical correlation analysis, and four kinds of current data of motor current, brake current, safety circuit current and car door motor current and vibration data in three directions of elevator operation are detected; off-line data are pre-processed, input into the two gating circulation unit neural networks respectively, and trained at the same time, so that the correlation coefficient obtained after canonical correlation analysis is carried out on the output of the two gating circulation unit neural networks is maximum; and online data are pre-processed and input into the two trained networks, and the correlation coefficient is compared with a threshold value to realize fault detection.