177results about How to "Reduce operational overhead" patented technology

The invention provides a power weight quantification-based nerve network forward operation hardware structure comprising the following elements: an input buffer memory module used for caching input data and nerve network power weight transferred to a chip from a global memory, wherein the nerve network power weight comprises a convolution core and a full-connecting matrix; a computing module used for carrying out convolution and full-connecting matrix vector multiplying operation based on power weight quantification; an output buffer memory module used for caching the convolution and full-connecting matrix vector multiplying operational result obtained by the computing module; a control module used for carrying out data transfer and operational resource scheduling for the input buffer memory module, the computing module and the output buffer memory module. The power weight quantification-based nerve network forward operation hardware structure can effectively reduce the nerve network memory space, and can effectively reduce nerve network operational expenses, thus obviously improving the application space of the nerve network calculating system on the terminal equipment.

This invention puts forward a method for encrypting double images based on fractional stage Fourier transformation, which disorders pixels of one of two original images, then carries out pure phase coding to the disordered result to multiply the other original image to generate a composite image with a duplicate swing and then makes fractional stage Fourier domain double random phase coding and encryption to get an encrypted image having generalized stable white noise approaching to gaussion distribution, which presents a method for reducing errors by threshold value by analyzing influence to decryption result by disturbance.

The invention discloses a joint estimation method for synchronizing frames, frequencies and fine symbols for orthogonal frequency division multiplexing (OFDM). The method comprises the following steps: carrying out autocorrelation operations of small point length and big point length on base band data sequences at the receiving terminal in parallel; jointly determining the frame synchronization position in real time and in parallel according to the small point autocorrelation peak and the big point autocorrelation peak; computing integer frequency offset coarse estimation value and decimal fraction frequency offset coarse estimation value according to the small point autocorrelation peak and the big point autocorrelation peak; estimating the final frequency offset detection value from the integer frequency offset coarse estimation value and the decimal fraction frequency offset coarse estimation value with a frequency offset dereferencing decision device; carrying out frequency offset correction on the base band data sequences according to the estimated final frequency offset detection value; and carrying out cross-correlation on the corrected base band data sequences and the local data sequences and determining the accurate fine symbol synchronization position through the cross-correlation peak. The contradiction of complex system measure function hardware design in the traditional time-domain joint estimation systems is solved by using the method.

The invention relates to a data center energy saving method based on a central processing unit (CPU) dynamic frequency modulation technology. The method includes two stages of initialization arrangement and dynamic management of a virtual machine. The method is different from methods in the prior art, a virtual machine arrangement technology used by energy saving of a data center is divided into the two stages of initialization arrangement and dynamic management. The initialization arrangement is applicable to the occasions of operation beginning and regular integral adjustment of the data center. The adopted method achieves the overall effect and the long-term effect due to time permission. The dynamic management is used for local management or moving facing different tasks during data center operation. The adopted method relates to local resource and ensures real-time performance.

The invention discloses a wireless camera sensor network position and posture calibration method for a wireless camera sensor network, and the method comprises the steps: enabling an unmanned plane carrying a photoelectric identification to enter a region covered by the wireless camera sensor network; taking a plurality of pictures comprising the photoelectric identification at equal intervals in a recognition range of one wireless camera sensor; carrying out the decoding of the photoelectric identification, extracting the pictures which have no overlapped view and are taken by the wireless camera sensor at three different moments, solving and obtaining the position and posture information of the wireless camera sensor; carrying out the feature matching of the wireless camera sensors which have an overlapped view, calculating an intrinsic matrix, decomposing and obtaining a rotation matrix; solving and obtaining the position and posture information of the wireless camera sensors which have the overlapped view, and determining the rotation matrix and a translation amount. The method achieves the quick, simple and precise calibration of the wireless camera sensor, and the wireless camera sensor network calibrated through the method can be quickly applied to disaster relief and other situations where the cameras need to be calibrated.

The invention discloses a multi-strategy-combined OFDM (Orthogonal Frequency Division Multiplexing) frame synchronization method, which can be used for effectively resisting wrong synchronization caused by interferences of noise, carrier frequency offsets, same-frequency signals and the like and meets the requirement of frame synchronization of an OFDM system. The method comprises the following steps of: (1) performing symbol cross-correlation operation on a received signal sequence and a local training sequence, performing time delay auto-correlation operation on the received signal sequence, and performing energy operation on a time delay received signal sequence; (2) judging and detecting a cross-correlation operation result and a preset cross-correlation judgment threshold T1, judging and detecting an auto-correlation operation result and a preset auto-correlation judgment threshold T2, and judging and detecting the auto-correlation operation result and a self-adaptive judgment threshold T3; (3) when frame arrival is detected, searching for a peak value of the auto-correlation operation result to obtain a timing position; and (4) adjusting the position of the received signal sequence according to the timing position to obtain a frame synchronization output signal.

A method and a system for operating bank business data memory cache are provided. The system comprises a bank business application system, a memory cache cluster and a bank business database. The memory cache cluster comprises a plurality of memory cache servers; the bank business database is used for outputting bank business information data; the bank business application system is connected with the memory cache cluster and is used for receiving a user operation request and transmitting an access request according to the user operation request; the memory cache cluster is respectively connected with the bank business database and the bank business application system and is used for converting the bank business information data as an unstructured memory data structure and then storing in the memory cache servers, and distributing address information corresponding to the memory cache servers to the bank business application system, and inquiring the bank business information data in a local application memory, and outputting a feedback result to the bank business application system. Thereby, the method and the system of the invention provide an integrated, concise and standardized data access method and improve entire interactive response speed of a bank application system under the condition of super-large scale concurrent access.

The invention provides a method for monitoring live broadcast content in real time. The method is characterized by comprising the steps of: receiving and analyzing live broadcast streaming data and viewer streaming data in real time, so as to obtain live broadcast data and viewer data; extracting feature data from the live broadcast data and the viewer data; calculating a score value corresponding to each piece of feature data separately; and acquiring an arbitrary score value by means of a first classifier to be compared with a first preset threshold value, judging a safety level of the live broadcast content, and outputting a judgment result. Therefore, by adopting the method for monitoring the live broadcast content in real time provided by the invention, the live broadcast content can be detected through combining features such as video, audio and user behaviors in real time, and the real-time performance of detection can be improved; moreover, the final detection result is determined in a manner of cascading a plurality of classifiers during detection, thus the computation overhead can be reduced, and the precision of the detection result can be improved.

The invention provides a data segmenting method and system of a distributed graph calculating system. The data segmenting method includes determining a similarity measurement value between each data node and an own first adjacent node in to-be-processed data; acquiring appearing times of a label of each first adjacent node, and determining whether at least two labels identical in appearing times exist or not; if yes, determining second adjacent nodes corresponding to the labels, and determining a label of each data node according to a similarity measurement value between each data node and the corresponding second adjacent node; dividing the data nodes with a same label into a same community, and storing the data nodes belonging to the same community into a same processing host. Community dividing of the data nodes is realized by taking similarity features among the data nodes into full consideration and on the basis of the labels, and operation expenditure is saved; the data nodes close in relationship are distributed into the same processing host, so that expenditure in communication among different processing hosts is reduced.

The invention discloses an EKF-SLAM (Extended Kalman Filter-Simultaneous Localization And Mapping) algorithm based on a straight line feature map, which is characterized in that straight line features in an environment are extracted by using laser radar, and a feature map is established according to the straight line features. The EKF-SLAM algorithm taking a straight line as the feature map is creatively designed. In an experimental environment with obvious straight line features, straight lines features of the environment are extracted very easily. The complexity of the algorithm is low, the operation overhead of a processor is reduced, the real-time performance of the system is improved, and the environment processing means of the SLAM algorithm is enriched.

The invention provides a parallax image generation method and system based on binocular stereo vision matching. The parallax image generation method comprises the steps of obtaining a pair of speckleimages, wherein the pair of speckle images are images of speckles captured by an object through a pair of binocularly calibrated cameras under the irradiation of a set light beam; inputting the pair of speckle images into a preset deep learning neural network to generate corresponding parallax images, wherein the deep learning neural network is obtained by training a plurality of pairs of speckleimages. According to the method and the system, a pair of speckle images is used as input, a parallax image is directly output through binocular vision stereo matching network (deep learning neural network) processing, end-to-end calculation of the parallax image and an end-to-end network design framework are realized, post-processing steps of stereo matching, such as sub-pixel enhancement, filtering and noise reduction, refinement and the like, are eliminated, and the operation overhead is greatly reduced.

The invention relates to a vehicle positioning method and device, and a terminal device. The method includes the following steps: obtaining positioning information sent by a peripheral device after apositioning instruction is received; selecting the peripheral device that meets a preset condition as a positioning reference according to the positioning information; and obtaining the positioning information of a location of a vehicle according to a wireless signal relationship between the vehicle and the peripheral device as the positioning reference. With the help of a plurality of fixed equipment having a V2X capacity (referring to Vehicle-to-Everything, information exchange between a vehicle and the outside) deployed in a parking stall of an indoor parking lot or near a lane in a vehiclenetwork, information interaction is directly performed on the terminal device having the V2X capacity through V2V, and indoor accurate location information of the vehicle is obtained without deploying a positioning network; and compared with traditional indoor positioning methods, network bandwidth overhead and computation overhead are reduced, and the complexity and cost of the positioning network are reduced.

The invention discloses a power frequency sinusoidal AC electrical signal zero-crossing time detection system and method thereof, relating to the time detection technology in the electrical signal measurement field. The system is characterized in that a power frequency sinusoidal AC signal source (100), an AD sampling circuit (210), a logic-to-logic circuit (221), and a pulse drive circuit (240) are successively connected; and a pulse generating circuit (230) and acontroller (222) are connected to a logic and calculation circuit (221). The method includes (1) initialization, (2) frequency parameter estimation; (3) phase parameter estimation; and (4) zero-crossing time reckoning and output. The power frequency sinusoidal AC electrical signal zero-crossing time detection system and method can effectively detect the zero-crossing time of the power frequency AC analog electrical signal; and controls the error within 1micron second, can be used for the minimal phase error measurement between the analog and digital signals, and facilitates the value traceability of phase accuracy.

The invention discloses a dynamic correction algorithm of a GNSS receiver. The dynamic correction algorithm uses differential coherence technology and does not need to estimate navigation message flip so that the computation cost is lowered greatly. The dynamic correction algorithm includes that carrying out dynamic correction before performing coherent addition, to be more specific, designing a DBZP differential coherence acquisition algorithm output matrix, extracting items related to range walk and range curvature through DFT transformation, designing a Keystone transformation algorithm to remove items related to range walk, and constructing phase compensation functions according to related information of related functions in a frequency domain to compensate the range curvature. The dynamic correction algorithm further weakens the influences of Doppler frequency and the change rate thereof, and accordingly the energy of the signal is more concentrated after performing the coherence addition; and moreover, the dynamic correction does not use an intermediate estimate, so that the dynamic correction algorithm is applicable to the environment with low signal to noise ratio, and the receiver can work normally under the application environment with low signal to noise ratio and high dynamic nature.

The invention provides a method and a device for detecting a spot defect of a face skin based on deep learning. The method comprises the steps of acquiring a training data set; constructing a convolutional neural network, and training the convolutional neural network by use of the training data set to obtain a detection model, wherein the convolutional neural network has the following characteristics: multi-layer convolutional characteristic extraction is performed on a sample to obtain different hierarchies of characteristic graphs, then, sampling is performed by use of sampling frames with multiple sizes and shapes to obtain sampling result corresponding to different hierarchies, then all of the sampling results are classified by use of a softmax loss function and a position is predictedby use of an L1 loss function, a loss value obtained by calculation is returned to a front end of the convolutional neural network, and a convolutional neural network parameter is adjusted b y use ofa gradient descent method; and segmenting a to-be-detected image into test image blocks, inputting all test image blocks into the detection model in a traversal manner to obtain corresponding image block detection results, and then splicing all of the image block test results into a complete image detection result.

The invention relates to the technical field of the detection of the traffic state of a road and particularly discloses a method for distinguishing the traffic jam of a toll plaza of a highway based on a video detection technology, which comprises the following steps: (1) shooting the video of the road of a toll plaza; (2) extracting the picture of the road of the toll plaza from the video; (3) establishing and updating a background model of the picture; (4) extracting a foreground image from the picture; and (5) obtaining the energy value and the absolute value of the variable quantity of the energy value of the foreground image and judging whether the road of the toll plaza is jammed or not. According to the method for distinguishing the traffic jam of the toll plaza of the highway based on the video detection technology, only the parameter of the energy value of the road needs to be obtained, and a jam distinguishing model is constructed, so that the judgment to the jam state of the toll plaza of the highway is finished. The algorithm is simple, the operation expense is little, the real-time property is strong, the problem of distinguishing the traffic jam of the toll plaza of the highway can be accurately and effectively solved, and a jam caution is also output at the time of jam, so that powerful information support is provided for a manager to master a site condition in time and make a management decision so as to reduce the potential safety hazard of pass in traffic.