Patents
Literature
Hiro is an intelligent assistant for R&D personnel, combined with Patent DNA, to facilitate innovative research.
Hiro

42results about How to "Reduce computational redundancy" patented technology

Service index obtaining method and device, server and computer readable storage medium

The invention provides a service index obtaining method and device, a server and a computer readable storage medium. The method comprises the steps of obtaining log data of terminal equipment; extracting dimensions and indexes of a service for log data, performing structured storage on the log data according to the dimensions and the indexes to obtain a first data table, enabling a field of the first data table to comprise a plurality of dimension identifications and a plurality of index identifications, and abstracting each piece of log data into one record in the first data table; and inputting the target dimension combination, the to-be-counted target index identifier and the first data table into a preset operation system for processing to obtain a target index, corresponding to the target index identifier, of the service under the target dimension combination. When the business indexes are counted, only the target index identifier and the target dimension combination need to be input to the preset operation system, so that the calculation redundancy in the business index counting process is greatly reduced, and the code maintenance difficulty is reduced.
Owner:BEIJING QIYI CENTURY SCI & TECH CO LTD

AGV (automated guide vehicle) path planning method based on A* algorithm

The invention discloses an AGV (automated guide vehicle) path planning method based on an A* algorithm. The AGV path planning method comprises steps as follows: firstly, an environmental map is established with a topological method, and the abstract environment is described with a specific data structure; secondly, a heuristic function is established with the A* algorithm, a reasonable evaluationfunction is set, and a path between nodes on the map is solved; finally, according to a specific task requirement of an AGV, a reasonable price mechanism is made to guide the algorithm for obtaining of the optimal solution. With adoption of the method, the working environment of the AGV can be described accurately, storage space of the map is reduced, and efficiency and accuracy of the algorithm are improved.
Owner:NANJING UNIV OF SCI & TECH

Virtual view generating method based on shielding region detection

The invention provides a virtual view generating method based on shielding region detection. The virtual view generating method comprises the following steps of performing edge detection on a left view and a right view of a left reference viewpoint and a right reference viewpoint and extracting edge information; respectively dividing pixel sections of the left view and the right view according to the edge information and performing matching on the pixel sections; checking matching results of the pixel sections so as to perform secondary matching according to the shielding situation and detecting shielding regions of the left view and the right view; generating a virtual view according to the matching result of the pixel sections and detection results of the shielding regions. The virtual view generating method can fully utilize information of two reference viewpoints and avoid large-area void rendering. In addition, calculation redundancy can be further decreased and the calculation efficiency can be improved by adopting the method. Furthermore, virtual view rendering is achieved based on a shielding region and non-shielding region local change method without pixel calculation according to parallax values, and accordingly the drawing speed can be improved.
Owner:TSINGHUA UNIV

Illegal parking detection system and detection method thereof

The invention discloses an illegal parking detection system and a detection method thereof. The illegal parking detection system is composed of a sensor unit, an imaging control unit, and an intelligent illegal parking capture unit. The sensor unit is responsible for converting an optical signal into an electrical signal. The imaging control unit is responsible for adjusting imaging parameters of the sensor unit, and transmitting a taken image to the intelligent illegal parking capture unit for analysis. The intelligent illegal parking capture unit is used for feeding part of control parameters back to the imaging control unit. By adopting the illegal parking detection system and the detection method thereof, not only capture in the day can be realized, but also capture at night can be realized. A driver away-from-car detection technology and a night specific detection technology are fused into the system, which ensures the effectiveness of all-weather capture.
Owner:青岛比特信息技术有限公司

Long-time target tracking method and system based on layered convolution feature

The invention discloses a long-time target tracking method and a system based on layered convolution characteristics. The invention adopts a pre-trained depth convolution neural network to extract convolution characteristics of each layer for each frame of video data; according to the model of correlation filter, the response of convolution layer is calculated and weighted, and the maximum value of the final response is taken as the target center of the current frame; in the process of frame-by-frame updating, the threshold value is set, and the correlation filter model is updated only when the tracking response value is greater than the threshold value, otherwise, the correlation filter model of the previous frame is adopted, and at the same time, when the tracking response value is lowerthan the set threshold value, the stochastic fern algorithm is used to re-detect the target. The invention adopts depth learning for feature extraction, and improves the precision of target tracking;the modified model updating method reduces the computational redundancy in the whole tracking process and improves the tracking speed; a random fern algorithm is used to re-detect the target in bad conditions such as target occlusion, which ensures the reliability of the actual tracking results.
Owner:济南市中未来产业发展有限公司

Computing unit, array, module, hardware system and implementation method

The invention discloses a computing unit, an array, a module, a hardware system and an implementation method, and belongs to the field of artificial intelligence algorithm hardware acceleration. Aiming at the problems of huge data and long calculation time of a sparse convolutional neural network algorithm in the prior art, an invalid data removal mechanism is designed in a calculation unit, invalid weights or input image data can be removed, the calculation time is reduced, and the power consumption caused by multiplication and accumulation calculation is reduced; a multi-channel sub-computing unit is designed, a multiplexing accumulation channel mechanism is adopted to complete convolution operation, and resource consumption is reduced; under the condition that invalid data is removed, asupply number rotation mechanism is further designed, and sufficient supply number of the calculation unit can be kept; the method is low in power consumption, small in area, high in throughput rateand high in recognition speed, is suitable for application of mobile terminals, such as smart home and smart city, and can efficiently complete license plate recognition, face recognition and the like.
Owner:南京宁麒智能计算芯片研究院有限公司

Text processing method and device, electronic equipment and readable storage medium

The invention relates to a text processing method and device, electronic equipment and a computer readable storage medium, and the method comprises the steps: obtaining a target text sequence of a target text, and extracting a subject term sequence from the target text sequence; respectively obtaining semantic distances between all words in the target text sequence and the subject term sequence, and determining a keyword sequence of the target text based on the semantic distances; obtaining a vector of each keyword in the keyword sequence to construct a word vector matrix; obtaining a semanticweight parameter matrix of the target text through a preset random gradient descent algorithm; and obtaining a semantic vector of the target text according to the word vector matrix and the semanticweight parameter matrix. According to the text processing method provided by the invention, the semantics of the target text can be expressed more accurately.
Owner:BEIJING INST OF ENVIRONMENTAL FEATURES

Signal capturing method and device

The invention discloses a signal capturing method and device. The method is characterized in that a satellite signal is obtained; frequency mixing is carried out on the obtained satellite signal and a local carrier wave, and Doppler stripping is realized; accumulative summation is carried out on a frequency mixed signal in a chip; correlation integration is carried out on an accumulated signal and a locally generated code signal; and a coherent result is sent to a capturing decider to carry out capturing decision. According to the method, intermediate results can be reused, and the computing resource is saved. The device which realizes the method can accelerate the signal capturing process, and has higher application value.
Owner:PEKING UNIV

Coding distributed computation method and coding distributed computation apparatus based on edge computation environment

ActiveCN108809520AEdge Computing ImplementationNo need to change the migrationForward error control useData migrationNetwork parameter
The invention discloses a coding distributed computation method, a coding distributed computation apparatus, a coding distributed computation device based on edge computation environment, and a computer readable storage medium. The method comprises the following steps: acquiring network parameters of an edge computation network; obtaining a computation task arrangement scheme based on the networkparameter; performing coding distributed computation to obtain a coding scheme of data migration in the edge computation network; determining any one participated node as a target node, determining acomputation task of a corresponding median of any one output function as a special computation task when the any one output function on the target node and corresponding input data both are located onthe target node, and the other participated nodes, except for the target node, do not have such any output function, determining the special computation task of the other participated nodes, except for the target node, as an abnormal computation task, and deleting the abnormal computation task in the computation task arrangement scheme; and realizing edge computation based on the coding scheme and the computation task arrangement scheme. With the method, the apparatus and the device provided by the invention, computation redundancy is reduced greatly, and network computation resource is saved.
Owner:四川边缘算力科技有限公司

Unsupervised clustering method used for large data volume spectral remote sensing image classification

The invention discloses an unsupervised clustering method used for large data volume spectral remote sensing image classification, comprising the following steps: dividing the original data into a plurality of data blocks, and obtaining a cluster center of each data subblock by virtue of a peak density searching method; dividing each cluster center into a plurality of data blocks again, and clustering again by virtue of the peak density searching method, so that number of the cluster centers is reduced; and repeating a partitioning-clustering process until similarity of any two cluster centerscan be represented by using one two-dimensional matrix, and then obtaining a final classification result. The unsupervised clustering method disclosed by the invention has the advantages that applicability is good, so that the method not only can be used for hyperspectral remote sensing image classification with more spectrum bands but also can be used for hyperspectral remote sensing image classification with fewer spectrum bands after multispectral remote sensing image or spectrum band selection; and operation efficiency is relatively high, blocked processing reduces computation redundancyof a similarity matrix, and clustering processing of all the data blocks is mutually independent, so that parallel processing can be adopted, and classification rate is increased.
Owner:BEIHANG UNIV

SAR echo rapid time domain generation method

The invention discloses an SAR echo rapid time domain generation method. All point targets included by a surface target are merged stage by stage, each merged group of point targets at each stage are replaced with one virtual point target, first-stage echoes are generated by use of a time domain superposition method, as the point targets are merged stage by stage, sub echoes generated at each stage gradually become dense, finally, when the number of sampling points in an echo data orientation becomes the same as the number of orientation sampling points, merging ends, and complete echoes are obtained by superposing sub echoes of the last stage. Compared to a conventional time domain superposition method, the method provided by the invention has the following advantages: the orientation sampling points and the point targets do not have to be traversed one by one, the calculation redundancy generated by the echoes is reduced, the computation amount is greatly reduced, the method is applied to an SAR system of any mode and is especially applied to echo generation of a large-scene surface target, and at the same time, platform motion errors are easily introduced.
Owner:UNIV OF ELECTRONICS SCI & TECH OF CHINA

Deepfake video detection method based on image group and two-stream network

The invention relates to a Deepfake video detection method based on an image group and a two-stream network. The method comprises the following steps: (1) extracting key frames of a video to be detected to form an image group; (2) inputting the first frame of the image group into a spatial stream in the two-stream network to extract spatial features; (3) differentiating the remaining frames of the image group with the first frame to obtain difference images, forming a difference image sequence, and inputting the difference image sequence into the time flow in the two-flow network to extract time features; and (4) fusing the extracted spatial features and time features, and evaluating the authenticity of the video by using a dynamic routing algorithm. Compared with the prior art, the method has the advantages that the calculation redundancy is reduced by using the image group, the network is enabled to focus on the key frame, the spatial and temporal information of the key frame is fully utilized by fusing the spatial features and the temporal features, and classification is performed through the dynamic routing algorithm to obtain a more accurate evaluation result.
Owner:NANJING UNIV OF INFORMATION SCI & TECH

Graphic calculation iteration processing method based on SCC-DAG

The invention discloses a graphic calculation iteration processing method based on SCC-DGA. The method comprises the processing step and the calculation step; specifically, in the stage of preprocessing, all SCCs in a given graph are obtained, and the SCC-DGA is constructed through connection among the SCCs; in the stage of calculation, according to the topological sorting of the SCCs in the SCC-DGA, iteration is performed inside the SCCs of the SCC-DGA in sequence till convergence of all the SCCs is achieved, and data in the convergence status is output as a computational result of a corresponding algorithm. According to the graphic calculation iteration processing method based on SCC-DGA, iteration processing is defined inside the SCCs, and iteration operation is performed independently inside each SCC; when all the SCCs inside the SCC-DGA are subjected to convergence, the whole graph is in the convergence status; according to the graphic calculation iteration processing method, the low preprocessing cost is needed for increasing the graph convergence speed, calculation redundancy is reduced, and the IO resource cost is lowered.
Owner:HUAZHONG UNIV OF SCI & TECH

Method and device for removing scene object models in game

The invention discloses a method and device for removing a scene object model in a game. The method comprises the steps of removing each scene object model in a game scene in a predetermined rejectionprocessing mode, wherein the predetermined rejection processing mode comprises cone rejection processing and / or potential visible object rejection processing; obtaining occlusion information of the scene object model after removing processing; determining the occlusion type of the scene object model after elimination processing according to the occlusion information; obtaining a target scene object model from the removed scene object model based on the occlusion type detection; and rendering the target scene object model. According to the method and the device, the technical problem of relatively low calculation efficiency of a CPU due to calculation redundancy in the process of removing the scene object model from the CPU in the prior art is solved.
Owner:NETEASE (HANGZHOU) NETWORK CO LTD

Inference method and system for dynamic branch selection of deep learning model

The invention provides an inference method and system for dynamic branch selection of a deep learning model, and aims to solve the problem of redundancy of deep features of a convolutional neural network in a decoding part in the inference process of a traditional deep learning model, and can automatically select corresponding decoding branches according to the category of primary features, each decoding branch is trained through the primary feature input and the corresponding output of the corresponding branch category to obtain different feature weight sets, and the method can focus on the processing of the primary features of the corresponding branch category and the recognition of deep features to complete more complex classification or representation generation, therefore, the computational redundancy is reduced, the reasoning speed is increased, and the actual application requirements of edge computing equipment are met.
Owner:魔视智能科技(上海)有限公司

Image processing method and device, equipment, computer storage medium, and system

The embodiment of the invention discloses an image processing method and device, equipment, a computer storage medium, and a system. The method comprises the following steps: acquiring a plurality of reconstructed blocks, wherein the plurality of reconstructed blocks are obtained after a plurality of image blocks, divided by a to-be-processed image, pass through a preset coding network model and a preset decoding network model; determining a central region of at least one reconstructed block in the plurality of reconstructed blocks; performing quality enhancement on the central region of the at least one reconstructed block to obtain at least one target reconstructed block; generating a reconstructed image according to the at least one target reconstructed block and the plurality of reconstructed blocks; filtering block boundaries in the reconstructed image to obtain a target image.
Owner:GUANGDONG OPPO MOBILE TELECOMM CORP LTD

A coding distributed computing method and device based on an edge computing environment

ActiveCN108809520BEdge Computing ImplementationNo need to change the migrationForward error control useEdge computingGoal node
The invention discloses a coding distributed computation method, a coding distributed computation apparatus, a coding distributed computation device based on edge computation environment, and a computer readable storage medium. The method comprises the following steps: acquiring network parameters of an edge computation network; obtaining a computation task arrangement scheme based on the networkparameter; performing coding distributed computation to obtain a coding scheme of data migration in the edge computation network; determining any one participated node as a target node, determining acomputation task of a corresponding median of any one output function as a special computation task when the any one output function on the target node and corresponding input data both are located onthe target node, and the other participated nodes, except for the target node, do not have such any output function, determining the special computation task of the other participated nodes, except for the target node, as an abnormal computation task, and deleting the abnormal computation task in the computation task arrangement scheme; and realizing edge computation based on the coding scheme and the computation task arrangement scheme. With the method, the apparatus and the device provided by the invention, computation redundancy is reduced greatly, and network computation resource is saved.
Owner:四川边缘算力科技有限公司

Light field matting method and device based on air angle consistency

PendingCN114742847AGuaranteed Empty Corner ConsistencyReduce computational redundancyImage enhancementImage analysisParallaxLight field
The invention discloses a light field matting method and device based on air angle consistency. The method comprises the following steps: step 1, extracting a central sub-aperture image of light field data and calculating the parallax of the central sub-aperture image; step 2, calculating a central sub-aperture image alpha graph; and step 3, propagating the central sub-aperture image alpha image through a light field alpha image propagation model to obtain a light field alpha image alpha u, v (x, y). By adopting the method provided by the invention, accurate and consistent light field matting can be realized, and the air angle consistency of the light field matting can be evaluated.
Owner:BEIJING INFORMATION SCI & TECH UNIV

System and method for supporting computation migration of multiple embedded devices

InactiveCN101937362BMitigate upgrade requirementsMeet information service needsDigital computer detailsSoftware simulation/interpretation/emulationParallel compressionTransfer procedure
The invention relates to a system and a method for supporting the computation migration of multiple embedded devices. The system comprises a computer, a coding machine and the embedded devices, wherein virtual machines corresponding to the embedded devices are arranged in the computer; the coding machine is connected with the computer and is used for receiving video images and audio data from the computer, performing concurrent compression coding and sending the video images and audio data subjected to the concurrent compression coding to the embedded devices; and the embedded devices are used for receiving the data sent by the coding machine and transmitting an interactive service instruction to the computer. The method comprises the following steps that: the embedded devices request for a service; the computer provides an interactive service; the coding machine performs concurrent compression coding on the video images and audio data sent from the computer and distributes the video images and audio data subjected to the concurrent compression coding to the embedded devices; and the embedded devices receive the data subjected to the compression coding by the coding machine and acquire the video images and sound of the requested service. The system and the method have the advantages of simple and compact structure, low cost, simple and convenient operation, simple principle, high processing capacity, small delay, high concurrent access capacity, low requirement on network bandwidth in the transmission process, and the like.
Owner:长沙天乘视讯网络科技有限公司

Reasoning method and system for dynamic branch selection of deep learning model

The invention provides a reasoning method and system for dynamic branch selection of a deep learning model. The invention aims at the problem that the deep features of the convolutional neural network in the decoding part are redundant in the reasoning process of the traditional deep learning model. , automatically select the corresponding decoding branch, and each decoding branch is trained through the primary feature input and corresponding output of the corresponding branch category to obtain different feature weight groups, which can focus on the processing of the primary features of the corresponding branch category and the identification of deep features In order to complete more complex classification or representation generation, thereby reducing computing redundancy, improving reasoning speed, and meeting the actual application requirements of edge computing devices.
Owner:魔视智能科技(上海)有限公司

SAR ship detection method based on global and local context information

The invention discloses an SAR ship detection method based on global and local context information. The method comprises the steps: 1, obtaining an SAR image, and carrying out the processing of the SAR image to obtain a binarized image; 2, performing morphological processing on the binarized image; 3, carrying out region selection based on global context information, and screening out sub-images of all hidden ships; 4, carrying out classification and regression prediction on the sub-images of the hidden ship; 5, performing morphological processing on the binarized image again; and 6, carrying out false alarm suppression based on local context information to obtain a ship target. The method comprises screening sub-images to be detected by using global context information of an SAR image, and screening out sub-images containing ships; and then the detection result is screened again by using the detection frame and the local context information of the peripheral neighborhood, so that the calculation redundancy can be reduced, the detection efficiency can be improved, the false alarm can be reduced, and the detection precision can be improved.
Owner:HEBEI UNIV OF TECH

A method to reduce the complexity of turbo-mpa receiver

The invention discloses a method for reducing the complexity of a Turbo-MPA receiver, and belongs to the technical field of wireless communication. The method disclosed by the invention comprises thesteps of: obtaining the posterior log-likelihood ratio of received signals through an MPA receiver in a condition of prior equal-probability, namely in a condition that the prior log-likelihood ratiois zero, calculating extrinsic information of coding bits by utilizing the posterior log-likelihood ratio and a bayes formula, selecting a coding bit set having relatively low credibility from the extrinsic information, and outputting the corresponding log-likelihood ratio; determining a check bit set related to the coding bit set having low credibility through the check relationship; furthermore, updating the log-likelihood ratio of the coding bit set having low credibility by utilizing the check bit set, the check relationship and a sum product algorithm, so that the prior log-likelihood ratio is updated; and, replacing the zero prior log-likelihood ratio by the updated prior log-likelihood ratio, iterating to the maximum iteration time, and finally, outputting decoding bits. By means of the method for reducing the complexity of the Turbo-MPA receiver disclosed by the invention, the calculation complexity and the processing delay of the Turbo-MPA receiver in the iteration process can be reduced.
Owner:BEIJING INSTITUTE OF TECHNOLOGYGY

ViBe algorithm performance improvement method

PendingCN111798495AFast updateReduce the number of calls to the random number functionImage analysisComplex mathematical operationsPerformance enhancementAlgorithm
The invention provides a ViBe algorithm performance improvement method. The method comprises the following steps of: 1, creating two Boolean matrixes M1 and M2 with the same size as a ViBe input image; 2, modifying an operation performed when the current pixel (x, y) is judged as a background point in a ViBe algorithm; and 3, after the ViBe scans the current image, updating the sample points of the ViBe, and executing the step 2 again. According to the invention, the updating logic of the background model is updated by modifying the ViBe, therefore, the ViBe can be in a high-resolution video sequence, the random number function calling frequency of the ViBe algorithm is greatly reduced, the background model updating speed of the ViBe algorithm is increased, the calculation redundancy is reduced, the calculation efficiency is greatly improved, and meanwhile, the influence on the segmentation quality of the foreground and the background of the ViBe algorithm is extremely low.
Owner:ZHEJIANG UNIV OF TECH

Light field matting evaluation method and device based on air angle consistency

PendingCN114742846AGuaranteed Empty Corner ConsistencyEmpty Angle Consistency ValidImage enhancementImage analysisData centerLight field
The invention discloses a light field matting evaluation method and device based on air angle consistency. The method comprises the following steps: step 1, extracting a central sub-aperture image of light field data; 2, calculating a disparity map of the light field data and a light field data center sub-aperture image alpha map; 3, propagating the alpha image of the central sub-aperture image to obtain a light field alpha image; and 4, evaluating the air angle consistency of the light field data in the light field alpha graph through an air angle consistency evaluation index S-cons obtained in a spatial domain, wherein the value range of the index S-cons is (0-1), the closer the index S-cons is to 1, the smaller the variance of the boundary region is, the better the air angle consistency of the light field data is, and the closer the index S-cons is to 0, the poorer the air angle consistency of the light field data is. By adopting the method provided by the invention, accurate and consistent light field matting can be realized, and the air angle consistency of the light field matting can be evaluated.
Owner:BEIJING INFORMATION SCI & TECH UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products