246 results about "Perception model" patented technology

The invention relates to a prediction-based virtual network function scheduling method for 5G network slices, and belongs to the field of mobile communication. The method specifically comprises: establishing a delay-based service function chain queue model for service function chain features having dynamically changing service traffic; establishing a multi-queue cache model, and determining, at different time, priorities requested by the slices and a lowest service rate that should be provided, according to the size of a slice service queue; discretizing the time into a series of consecutive time windows, and establishing a prediction-based traffic perception model by using the queue information in the time windows as a training data set sample; and searching a scheduling method for the best service function chain VNF under the resource constraint that the caching of the slice service queue does not overflow according to the predicted size of each slice service queue and the corresponding lowest service rate. The method realizes on-line mapping of network slices, reduces the overall average scheduling delay of multiple network slices, and improves the performance of network services.

A network driving environment integrated perception model based on a convolutional and hollow convolutional structure simultaneously realizes object detection and semantic segmentation. A video imageof a road environment is shot through a forward-looking camera system mounted on a vehicle. The residual network model is used to get the bottom feature map of the image. The converged network is designed, which includes two sub-modules: object detection and semantic segmentation. The two modules share the bottom feature map. Among them, the object detection module is responsible for predicting the confidence level of the object frame and the category, and the semantic segmentation module is responsible for predicting the pixel level of each category. The appropriate loss function is selectedfor each of the two modules, and the perceptual model tends to converge in both modules after alternate training. Finally, the joint loss function is used to train the two modules simultaneously to get the final perceptual model. The model can simultaneously complete object detection and semantic segmentation with small computation amount, and the perceptual model uses a large amount of data of object detection to assist the semantic segmentation module to learn the image distribution law.

The embodiment of the invention provides a data processing method and device. The method comprises following steps: acquiring data perception data which denotes one of the following such as image data, video data and audio data; determining the target scene where the target perception data is located; determining a corresponding target perception model; and calculating an identification result of the target perception data based on the target perception model. Therefore, the data processing method and device provided by the embodiment have following beneficial effects: by determining the scene for perception data, the target perception model corresponding to the scene is utilized for calculating the identification result of perception data; and compared with the prior art, calculation complexity is decreased in order to increase efficiency of data processing.

The embodiment of the invention discloses an image recognition method, device and equipment, a computer readable storage medium and a confusion perception convolutional neural network. The confusion perception convolutional neural network comprises a prediction classifier, a confusion perception model, a correction classifier group and a probability average layer, and the prediction classifier, the confusion perception model, the correction classifier group and the probability average layer are trained by using a training sample set and are used as traditional convolutional neural network classifiers. The confusion perception model is constructed based on a confusion matrix obtained by cross validation of a prediction classifier on the training sample set. Each correction classifier is obtained by using a confusion perception model as a decision-making system and training confusion class sample data with fuzzy boundaries in the training sample set. The probability average layer outputsa classification result of the to-be-identified image according to the category probability output by the prediction classifier and the category probability output by the target correction classifier, and the target correction classifier is a correction classifier selected by the confusion perception model according to the prediction category of the prediction classifier. The accuracy of image recognition can be improved.

A method and system for encoding and decoding an input signal in relation to the most perceptually relevant aspects of the input signal. A two-dimensional (2D) transform is applied to the input signal to produce a magnitude matrix and a phase matrix that can be inverse quantized by a decoder. A first column of coefficients of the magnitude matrix represents a mean spectral density (MSD) function of the input signal. Relevant aspects of the MSD function are encoded at a beginning of a data packet. The MSD function is also processed through a core perception model to determine bit allocation. The matrices are then quantized and priority ordered into a data packet, with the least perceptually relevant information at the end of the packet so that it may be ignored or truncated for scalability to the channel data rate capacity.

The invention discloses a multipath utilization through-the-wall radar imaging method based on the Bayes compression perception. A single station single transceiving antenna frequency stepping radar system is used for emitting pulse strings in multiple positions parallel to a wall face and acquiring echo signals; compression sampling is performed on the echo signals by randomly measuring a matrix so as to obtain relatively few measurement signals; then, by combining a dictionary construction compression perception model designed based on the multipath reflection characteristics of the wall face, the Bayes compression perception is used for reconstructing information of scenes after the wall face from the measurement signals; and finally the reconstructed information is sued for finishing through-the-wall radar imaging. According to the invention, cost is reduced; recording time is shortened; by use of the Bayes compression perception technology, calculation speed is quite fast, and required observation data are quite few; and the precision of images obtained through combination of the multipath utilization method is quite high and occurrence of false shadows is effectively inhibited.

The invention belongs to the video and image processing field and provides a stereo image quality evaluation method which is in accord with related characteristics of a human visual system and is more effective. Through the method, quality of a stereo image can be more accurately and effectively evaluated, and development of the stereo imaging technology is further facilitated to a certain degree. The method comprises steps that step 1, a monocular perception model in accord with the human visual characteristics is constructed; and step 2, an image distortion degree QD of a perception image Xv acquired at the step 1 is calculated, 1), structure similarity SSIM of left and right view image subblocks is calculated; 2), the final image distortion degree QD is constructed, in combination with eye stereo visual binocular characteristics and visual center significance characteristics, weighted summation of structure similarity indexes of the image subblocks is carried out to calculate a final image distortion degree evaluation score QD. The method is mainly applied to video and image processing.

The application provides a data flow prediction device and method. The device comprises a signal acquisition module, a confirmation module, a signal processing module and a computation module. The confirmation module is used for confirming that waveforms corresponding to data flow signals have self-similarity according to the data flow signals. The signal processing module is used for decomposition and reconstruction of the waveforms by adopting a wavelet analysis technology. The confirmation module is also used for confirming at least one approximation signal and signals, of which smoothness is less than a first preset threshold value, of multiple detailed signals to be the first type of data flow signals, and confirming the signals of which smoothness is greater than or equal to the first threshold value to be the second type of data flow signals. The computation module is used for predicting the first type of data flow signals by adopting a compression perception model and predicting the second type of data flow signals by adopting a linear model so that the first type of prediction result and the second type of prediction result are synthesized. With employment of the device or the method of the application, data flow prediction precision can be enhanced.

The invention discloses a multi-domain network security situation perception model and a multi-domain network security situation perception method based on SDN. The model comprises a streaming data extraction module, a streaming data abnormity detection module, a security situation factor extracting module, a security situation evaluation module and a network security situation evaluation knowledge base. According to the multi-domain network security situation perception model and method based on SDN, the flow table mechanism in OpenFlow is used, and thereby a controller can obtain the network flow information more timely and efficiently, and extra network loads are not added. Compared with obtaining the flow information based on data packets or data packet integrated flow in the traditional network, the method does not need the software support of the router to NetFlow, does not need to additionally configure a hardware chip in the exchanger to support sFlow either, thus the system is cost-saving and convenient to deploy.

A speed control method of a vehicle including the steps of obtaining a steering angle, a velocity error and a distance error. The velocity and the distance error being determined by mathematical combinations of a GPS position, a required path and speed set points. The steering angle, velocity errors and distance error are applied to fuzzy logic membership functions to produce an output that is applied to a velocity rule base. An output from the velocity rule base is defuzzified to produce a speed signal.

The present invention discloses a spread spectrum through-the-wall radar imaging method based on the compression perception technology. After an index average background elimination method is employed to remove the direct wave and the wall clutter, measurement signals are obtained through a random measurement matrix, a dictionary having been set and the measurement matrix are combined to construct a compressed perception model of a multi-measurement vector (MMV), target image information after the all is reconstructed from the measurement including few measurement values through an orthogonal matching pursuit (OMP) method, and finally, the through-the-wall radar is completed according to the target after the wall. Compared with a traditional synthetic aperture radar imaging method, the spread spectrum through-the-wall radar imaging method based on the compression perception technology is able to greatly reduce the number of the observation points and the data storage amount and shorten the recording time. The problems are solved that the spread spectrum through-the-wall radar SAR imaging method in the prior art requires lots of observation antennas, large data storage space and long data recording time.

The invention discloses an MLN-based network space security situation prediction method and system. The method comprises the steps that asset information data in a specific network space is collected; the collected asset information data is preprocessed, and a network space security situation perception model is constructed and trained; the current network space security situation is evaluated according to the network space security situation perception model and actual data in the current network space; and the future network space security situation is predicted according to a security situation evaluation result of the network space to obtain a security situation prediction result. According to the method, the relation between objects and object properties can be easily represented according to a first-order logic rule by applying a Markov logic network; and in addition, by introducing background knowledge, the object relation in the network space can be mastered more accurately, and then the evaluation accuracy rate is effectively increased. The method and system can be widely applied to the computer network field.

The invention relates to a filter bank simulating an auditory perception model, comprising an analysis filter bank 2, an each channel gain calculation module 3, a multiplier unit 4 and a synthesis filter bank 5. An audio frequency digital signal X(n)1 is divided into K channels after passing through the analysis filter bank 2, each channel gain calculation module 3 calculates to obtain a specificgain value of each channel, the multiplier unit 4 enables the channel gain value to multiply a corresponding sub-band signal, and an obtained result is synthesized into a path of output signal y(n)6 by the synthesis filter bank 5; the filter bank respectively carries out all-pass transformation and all-pass inverse transformation in the analysis filter bank 2 and the synthesis filter bank 5 of a weighting splicing adding structure by the method of combining the weighting splicing adding structure and the all-pass transformation so as to simulate the ear auditory resolution ratio under the condition of fewer channels. The invention has the high efficiency of the weighting splicing adding structure and meanwhile solves the problems of no real-time realization and phase distortion existing inthe present frequency transformation filter bank method.

A coverage hole eliminating method based on a multi-sensor probability perceptual model comprises the following steps that firstly, coverage hole detection is performed by adopting the grid method after nodes are deployed randomly, wherein a grid is completely covered if the associated perceptual probability of the grid is greater than 90 percent on the basis of the multi-sensor probability perceptual model, or else, the grid is not covered, and each continuously uncovered grid is a hole; secondly, the new position of the nodes for patching the holes is determined according to the geometrical algorithm; thirdly, a mobile target position is saved, the coverage area of the mobile nodes is removed, the nodes are moved to the target position to patch the holes, and coverage loopholes are updated; fourthly, the steps are repeated from the first step until all the loopholes are covered; finally, the steps are repeated until all the holes are covered.

The invention provides a heterogeneous network perception model division and task placement method in assembly line distributed deep learning, which mainly comprises three parts, namely deep learningmodel description, model division and task placement and assembly line distributed training. According to the method, firstly, for resource requirements of deep learning application in the GPU training process, corresponding indexes such as calculation time, intermediate result communication quantity and parameter synchronization quantity in the training execution process of the deep learning application are described and serve as input of model division and task placement; then indexes and heterogeneous network connection topology of the GPU cluster are obtained according to model description, a dynamic programming algorithm based on min-max is designed to execute model division and task placement, and the purpose is to minimize the maximum value of task execution time of each stage afterdivision so as to ensure load balance. And finally, according to a division placement result, performing distributed training by using assembly line time-sharing injection data on the basis of modelparallelism, thereby realizing effective guarantee of training speed and precision.

The invention discloses an image compensation method and device. The method includes the following steps that: a perception model of an image is constructed according to a screen correlation coefficient, the brightness value of the image, the backlight brightness value and the perception brightness value of the image; a brightness compensation function of the image is obtained through calculation according to the perception model of the image; the brightness compensation function is adjusted according to the brightness mean value and the backlight brightness value of the image, so that an adjusted brightness compensation function can be obtained; and a brightness compensation value which is obtained through the adjusted brightness compensation function is utilized to perform compensation on the brightness of the image. The invention also discloses a corresponding image compensation device. With the image compensation method and device provided by the technical scheme of the invention adopted, power consumption can be saved with image compensation quality guaranteed.

The invention, which belongs to the fields of biomedical engineering and machine learning, discloses an EEG epileptic attack detection method based on deep channel attention perception. According to the invention, an attention mechanism is introduced into multi-channel EEG epileptic attack detection to train an end-to-end deep channel attention perception model. With the model, the deep features of a brain wave signal can be extracted; and contribution scores of all channels to epileptic detection can be learned to select a most relevant EEG channel dynamically. Compared with the prior art, onthe basis of combination of deep feature extraction and the attention mechanism, the most relevant EEG channel is selected dynamically and the epileptic features are expressed synergistically, so that the fusion features have the channel perception capability; and the epileptic detection rate is increased and the interpretability is enhanced.

The invention relates to a multi-channel wide dynamic range compressing system based on an audition perception model. The multi-channel wide dynamic range compressing system comprises an analysis filter group for simulating an audition perception model, a sound pressure level detecting module, a compression amplification gain calculating module, a multiplier and an integrated filter group for simulating the audition perception model, wherein audio digital signals x (n) are divided into K channels after passing through the analysis filter group, the sound pressure level detecting module is used for detecting the sound pressure level of each channel, the compression amplification gain calculating module is used for calculating the specific gain value of each channel, the multiplier can multiply the gain values of the channels with corresponding sub-band signals, and the obtained results of the multiplier are integrated into a path of output signal y (n) through the integrated filter group, the integrated filter group respectively carries out the all-pass transformation and all-pass inverse transformation in an analysis filter group and an integrated filter group of a weighted splice adding structure through the mode of combining the weighted splice adding structure and the all-pass transformation, and can simulate ear audition resolution of a human under the condition of fewer channels.

A human perception model for a vehicle steering control method including the steps of obtaining a heading error, obtaining a velocity value, obtaining a distance error, applying the heading error, inputting a measure of operator aggressiveness and defuzzifying an output from a steering rule base. The velocity value and the distance error are applied along with the heading error to fuzzy logic membership functions to produce an output that is applied to a steering rule base. A measure of the operator aggressiveness is input to the steering rule base. An output from the steering rule base is defuzzified to produce a steering signal.

The invention provides an intelligent device and a control method and a device thereof, belonging to the field of machine learning. After receiving the execution instruction for the target task, the method can acquire the detection data and input the detection data and the target task to the perceptual model to obtain the representative detection data associated with the target task. Then the target task and the representative detection data can be input to the programming model to obtain the target state data. Then, the target state data and some or all representative detection data can be input to the control model to obtain the control parameters for controlling the intelligent device, and the intelligent device can be controlled based on the control parameters. The invention solves theproblems of large dependence on training samples and unsatisfactory training effect in the control process of the intelligent device in the prior art, and better control of the intelligent device canbe realized.

The embodiment of the invention provides a sleep environment air regulating method and device and electronic equipment. The sleep environment air regulating method and device and the electronic equipment are used for conducting accurate and effective regulation of air status in a sleep environment. The method comprises the steps that: current air status information of a target user in the sleep environment is obtained; the current air status information is input into a target comfortableness perception model to obtain current comfortableness value; whether the current comfortableness value iswithin an ideal comfortableness value interval of the target user under the target comfortableness perception model is determined; if the current comfortableness value is outside the ideal comfortableness value interval, target air status information adapted to the target user is determined based on the comfortableness value in the ideal comfortableness value interval as well as the target comfortableness perception model; based on the target air status information, corresponding air parameter regulating information to be regulated is determined; and target air regulating equipment is controlled to operate based on the air parameter regulating information and adjust the air status in the sleep environment.

The invention discloses a generalized perception model under limited spectral resources and a distributed Q learning access method. According to the model, a channel perception mechanism based on a zero-added latin square matrix is proposed by considering limitation of spectrum resources and dynamism of network environment and aiming at the problem of multiple time-slot channel perception sequenceoptimization. The method includes the following steps that firstly, a game model is constructed, and participants are all cognitive users in a network; based on the generalized perception model, eachuser randomly selects a channel perception sequence strategy from the corresponding zero-added latin square matrix and conducts perception; each active user calculates a current state return value and conducts Q-value update and probability update for the next time slot on the basis of the current state return value; the cognitive users conduct time-slot perception circularly until channel perception sequence strategies of all the cognitive users are converged. By adopting the model and the method, multiple time-slot channel perception sequence conflict is effectively reduced under the limited spectral resources, and handling capacity of the system cognitive users is improved.

The invention relates to a training method and a training device for an image segmentation model under label fault tolerance and related equipment. The method comprises the steps of obtaining a training sample set; processing the sample image through a segmentation model to obtain a prediction segmentation result; determining a segmentation loss function according to the prediction segmentation result and the pixel-level annotation of the sample image; processing the sample image and the pixel-level label thereof through a quality perception model and an anti-overfitting model to obtain a relative quality index; and adjusting parameters of the segmentation model and the quality perception model according to the segmentation loss function and the relative quality index to obtain a trained segmentation model. The invention relates to a training method and device for an image segmentation model under label fault tolerance and related equipment. According to the invention, the relative quality index is generated according to the sample image and the pixel-level label thereof. The segmentation loss function is adjusted according to the relative quality index to complete model training.Therefore, the trained segmentation model can still be ensured to have high accuracy when the training sample set has noise.