37 results about "Decision area" patented technology

The invention provides a software defined network control optimization method facing large-scale application. The software defined network control optimization method comprises the following steps: 1, an OpenFlow controller analyzes an OpenFlow flow table and divides a decision control region into N decision regions; 2, the OpenFlow controller updates a data packet forwarding rule of the decision regions into OpenFlow decision switching equipment; 3, OpenFlow entrance switching equipment analyzes a data packet and sends the data packet to OpenFlow exit switching equipment or forwards a forward path request flow table of the data packet to the OpenFlow decision switching equipment; 4, the OpenFlow decision switching equipment analyzes the forward path request flow table and forwards the data packet to the OpenFlow exit switching equipment corresponding to a forwarding rule; 5, the OpenFlow controller updates the forwarding rule of the data packet to the OpenFlow entrance switching equipment. Compared with the prior art, the software defined network control optimization method facing the large-scale application provided by the invention can effectively reduce the round-trip-delay of a network and the dependency of the data packet on the controller, improves the network throughput, and is applicable to large-scale network application.

The invention discloses an intersection-based V2P collision avoidance method comprising the steps that firstly, the participants of the game are ensured to be vehicles and pedestrians, and the participants are classified to define their passing strategy sets; secondly, street crossing delay and risk payment of the vehicles and the pedestrians of the intersection are solved and a V2P game model is established; thirdly, traffic reward and punishment payment is added on the basis of delay and risk payment so as to enhance the priority rate of the vehicles; and fourthly, an SEIR virus propagation model is improved, and an IEVRI model is established to analyze the psychology of the traffic participants so as to prove the necessity of introducing traffic rewards and punishments. The pedestrian flow and the vehicle flow in the traffic flow, the distance and the speed of the pedestrians and the vehicles from a decision area to a conflict area and the relative potential of the number of the unit batch of street crossing pedestrians are integrally considered, and the traffic reward and punishment factors are added and the V2P game model is established so that the road traffic safety can be enhanced and the passing efficiency can also be guaranteed.

The invention discloses a resource optimal allocation decision-making method based on deep reinforcement learning and blockchain consensus. The method comprises the steps: constructing a calculation task model and a server state model; calculating the energy consumption and economic overhead of local calculation and unloading calculation of the main controller, and the calculation economic overhead generated in the blockchain consensus process, thereby guiding and adjusting controller selection, unloading decision, block size and server selection through training a deep neural network and a strategy network, and completing optimal resource allocation in a scene. According to the invention, the problems of industrial Internet data security, over-high energy consumption of equipment due to processing of calculation tasks, short working period, over-high overall economic overhead of the system and the like are solved. Simulation experiments show that the industrial Internet resource optimal allocation decision-making method based on deep reinforcement learning and blockchain consensus has certain advantages in the aspects of saving controller energy consumption and system economic overhead and prolonging the total working duration of a controller group.

An information processing device is provided with a display controller configured to display a data display window on a screen of a display device. At least one decision area is defined within the display screen. A storage stores, in association with the at least one decision area printer selection information identifying a printer and/or print setting information. The information processing device further includes a control unit configured to identify a printer to be used based on the printer selection information corresponding to the at least one decision area and/or print setting based on the print setting information if the judging unit judges a dropped position of a data file is within the at least one decision area, print data being generated in accordance with the printer and/or print setting identified by the control unit.

The invention discloses a signal-to-noise-ratio estimation method and device based on hard decisions and also discloses a method and device used for signal-to-noise-ratio estimation. Each hard-decision area on a planisphere is divided into two subareas which are different in hard-decision reliability in advance and a weighting factor corresponding to each subarea is set. When reception signals are available, a subarea which the reception signals belong to is determined and estimated transmission-signal power is corrected by using the weighting factor corresponding to the subarea and then the corrected power is used to estimate a signal-to-noise ratio so that errors in signal-to-noise-ratio estimation are reduced.

The invention discloses an image recognition method based on a three-way decision and a CNN. The method comprises the following steps of inputting a sample data set image to train a CNN classifier; introducing three decision-making ideas in an image segmentation stage, and obtaining three decision-making classifiers by utilizing a positive data set image and a negative data set image which are obtained by classification of a CNN classifier; dividing the image into a positive region, a negative region and a delay decision region by using three decision classifiers; performing iterative classification processing on the delay decision area divided by the classifier; judging whether the remaining delay decision area reaches a critical value or not; when the delay decision area reaches a critical value, indicating that the to-be-identified image cannot be segmented again, and then performing CNN image identification on all positive areas. According to the present invention, the CNN image recognition technology and the three decision theories are combined, the useful information in the image can be fully utilized, and the higher recognition rate is achieved.

The invention discloses an unbalanced data classification system based on geometric structure integration, and the system comprises the following modules: an input module which converts collected samples according to the specific description of an unbalanced problem to obtain a sample set in a vector form, and the sample set in the vector form comprises a few types of samples and a plurality of types of samples; the training module is used for training the sample set in the vector form to obtain a few types of decision areas of the system; and the test module is used for inputting a to-be-discriminated sample and judging whether the to-be-discriminated sample is in the minority class decision area of the system to obtain the class to which the to-be-discriminated sample belongs. Accordingto the method, weak classifiers are designed by utilizing a supporting hyperplane principle, so that each weak classifier can identify a plurality of different types of samples, and labor division exists among the weak classifiers; through the combination of corresponding decision area spaces, the designed integration strategy can effectively identify a few types of samples and a plurality of types of samples, so that the problem of imbalance is effectively solved.

The invention discloses a symbol judging method. The symbol judging method comprises the following steps: storing a comparison table in a symbol judging circuit; receiving a first signal, and generating a coordinate signal group corresponding to the first signal according to the first signal, wherein the coordinate signal group is positioned in a first decision area; and reading the comparison table according to the coordinate signal group to output a first symbol corresponding to the first signal, wherein the first symbol is a first constellation point corresponding to the first decision area.

An on-board control device is provided with: an image acquisition unit that acquires an image obtained by an imaging device that images an area around a vehicle; a decision area division unit that divides a predetermined decision range of an imaging range of the imaging device into a plurality of areas; a feature point extraction unit that extracts, from the image acquired by the image acquisition unit, a feature point with a predetermined feature around the vehicle; a score accumulation unit that accumulates, with respect to a plurality of images chronologically acquired by the image acquisition unit, a score for an area of the plurality of areas in which the feature point extracted by the feature point extraction unit is present; and a decision unit that makes a decision as to whether foreign matter is attached in the imaging device on the basis of the result of accumulated scores for the plurality of areas.

The invention discloses a deep learning model compression method based on decision boundaries, and belongs to the technical field of model compression of deep learning. The deep learning model compression method based on the decision boundary comprises the following steps: step 1, carrying out feature mapping; step 2, performing piecewise linearization of an activation function; step 3, sub-decision region calculation: calculating a sub-decision region of a full connection layer; and step 4, decision network construction: calculating a corresponding decision boundary according to the sub-decision region, and using the decision boundary to construct a new decision network. According to the invention, efficient model compression of a full connection layer is achieved, and for a model with an activation function being piecewise linear, compared with an existing method, precision lossless compression can be achieved. And for other non-linear functions with infinite asymptotes as activation functions, model compression under controllable precision can be realized.

The invention discloses a data extraction method and device related to a deep learning model. According to a deep learning model of one embodiment, a data extraction method thereof comprises: a step of receiving an input query; a step of determining a first decision boundary set, the first decision boundary set being a subset of a decision boundary set corresponding to a target layer of the deep learning model; a step of extracting a decision region including the input query on the basis of the first set of decision boundaries; and a step of extracting data included in the decision region.