49results about How to "Avoid computational burden" patented technology

The invention provides an intelligent terminal and a video image stabilizing method and device. The method comprises the following steps: in a video compression process, obtaining residuals of motion estimation of macro blocks in an image frame; counting the residuals of the macro blocks in the image frame to obtain the residual of the image frame; adjusting a timestamp offset between a motion sensor and an image sensor according to the residual of the image frame; and carrying out subsequent timestamp correspondence, jitter estimation and jitter compensation on the image frame with the adjusted timestamp offset. According to the video image stabilizing method provided by the invention, the timestamp offset between the motion sensor and the image sensor is adjusted according to the residuals of the motion estimation in the video compression process, so that when the timestamp offset changes due to automatic exposure and other factors, the system can dynamically adjust the timestamp offset used in the timestamp correspondence, the jitter estimation and the jitter compensation to reduce the adverse effect on the image stabilizing effect resulting from the change of the timestamp offset.

The invention discloses a composite fault diagnosis method for a rotary machine based on an improved deep sparse auto-encoder network. The method comprises the following steps: 1) acquiring vibrationsignals of the rotary machine under c conditions in a normal state and different fault states respectively to obtain d groups of time domain vibration signal samples respectively; 2) carrying out Fourier transform on each sample to obtain preprocessed signal samples; 3) constructing a diagnosis sample set to serve as a training sample set; 4) building a rotary machine composite fault diagnosis model to obtain the connection weight and bias parameter of the deep sparse auto-encoder network; 5) acquiring a softmax classifier model; 6) carrying out fast Fourier transform, and selecting a test sample; 7) carrying out deep learning on the test sample by using the test sample as the input of a trained improved deep sparse auto-encoder network, and carrying out feature extraction to obtain a testsample feature signal; and 8) using test feature information as the matching feature of the test sample to obtain the composite fault diagnosis result of the rotary machine to be tested. Through theadoption of the method, the diagnosis accuracy and efficiency are improved.

The invention provides a robot control method capable of being used for multi-vehicle intelligent spraying. The robot control method is used for identifying hubs of different vehicle models and conducting particle spraying. The robot control method includes the steps that two-dimensional images of a target machining area are acquired in real time at the preset position, whether a workpiece entersor not is judged according to changes of feature points in gray values in the two-dimensional images before and after the entry of the workpiece, and meanwhile whether the workpiece is a hub workpieceor not can be further determined when first stereoscopic contour features of a target workpiece are obtained; and by comparing the first stereoscopic contour features with a stereoscopic model set ofan existing preset hub model base, a first stereoscopic model corresponding to the target workpiece is quickly obtained, and the increase of operation burden on equipment caused by regeneration of anew stereoscopic model is avoided.

The invention relates to an observability constraint-based random planet landing track optimizing method, and belongs to the technical field of the deep space probe navigation and guidance. According to the method, a deep space landing guidance control task based on the monocular vision navigation is used as the background, the dual problem between the effective control and the reliable estimation is considered, the system uncertainty is used as partial cost and is introduced into a quadratic performance index through an expansion-state space, so that a random optimization feedback control law can be provided by adopting a linear quadratic type control technique, an expansion-state space description model of a established probe system is substituted, the real-time optimization of a planet landing track is realized. The dynamic planning and the burden in calculation based on a searching method are avoided, the problem of observability shortage in the landing process is effectively overcome, and the navigation estimation performance of the system is improved, so that the overall performance during the planet probe navigation and guidance control is ensured, and the final target that a planet is reliably landed is achieved.

The invention discloses a ship target detection method based on anadaptive layered high-resolution SAR image. The method comprises the steps that firstly a visual attention mechanism combining an image domain and a frequency domain is constructed, not only can better maintain the details of a saliency map, but also has an obvious speed advantage compared with other attention mechanisms; secondly,a candidate target region segmentation threshold value is obtained by adopting a mean dichotomy, so that instability caused by hard threshold value selection is eliminated; and finally, ship target fine detection is carried out by using a frequency domain accelerated block kernel density estimation method. Calculation burden caused by super-pixel segmentation is avoided due to the idea of region partitioning. The mismatch risk of a clutter model is eliminated through kernel density estimation. In addition, the execution efficiency of the detection method is further improved through a frequencydomain acceleration method of the detection method. In addition, compared with an existing high-resolution SAR image ship target detection method, the method has the advantages that the constant false alarm characteristic can be kept while the target is detected.

The invention belongs to the field of artificial intelligence and visual dialogues, particularly relates to an autonomous evolution intelligent dialogue method, system and device based on a physical environment game, and aims to solve the problems of high calculation consumption, low convergence speed and low information processing accuracy of an intelligent system. The method comprises the stepsof obtaining a to-be-processed image and a corresponding problem text; Encoding the picture into a picture vector by adopting an optimized dialogue model, and encoding the question text into a question vector; Generating a state vector based on the picture vector and the question vector; And decoding the state vector to obtain a response text and outputting the response text. Wherein a discriminator needs to be introduced in the optimization process of the dialogue model, the dialogue model and the discriminator are alternately optimized until the mixed loss function of the dialogue model andthe loss function value of the discriminator are not reduced or lower than a preset value, and model optimization is completed. Compared with a traditional method, the method is low in calculation consumption and high in convergence speed, and the information processing accuracy is further improved.

The disclosure relates to a short message field classification method as well as a short message field identification method and apparatus. The short message field classification method comprises: determining a short message template set corresponding to each short message field in all collected short message fields; determining a training data set of a classifier according to the short message template set corresponding to each short message field in all the short message fields; determining a classifier training frequency corresponding to each short message template according to a frequency corresponding to each short message template in the short message template set corresponding to each short message field, subjected to statistics within a set time period; and training the classifier through the training data set and the classifier training frequency corresponding to each short message template in the short message template set. According to the technical scheme, the problem of computing burden caused by matching received short messages by taking all semantic templates as candidate templates in related technologies can be avoided, and a relatively good bedding is made for deep understanding of short message content.

A hybrid electromagnetic transient simulation method for microgrid real-time simulation, wherein a traditional node analysis method (NAM) and a highly parallel latency insertion method (LIM) are combined, so that the microgrid is firstly divided from a filter of a distributed power generation system to form one latency insertion method (LIM) network containing a power distribution line and a plurality of node analysis method (NAM) networks containing the distributed power generation system respectively, the NAM network being simulated by traditional node analysis method, the LIM network being simulated by the latency insertion method, in an initialization stage, one correlation matrix and four diagonal matrixes containing line parameters used for LIM network simulation being formed according to line topology and parameters of the microgrid, in a main cycle of the simulation, the LIM network solved simultaneously with multiple NAM networks, a parallelism of a microgrid simulation being improved.

The invention provides a semi-supervised image deblurring method based on a fusion attention mechanism, and the method comprises the steps: step 1, obtaining an image training data set, and dividing the image training data set into two training subsets; step 2, constructing an image blurring reduction model based on full supervision and non-supervision, wherein the image blurring reduction model comprises a full supervision network and a non-supervision network; step 3, inputting one subset into a fully supervised network to be trained for learning, and inputting the other subset into an unsupervised network to be trained; step 4, screening image features through the attention mechanism in the network, distributing corresponding weights to feature information extracted by the network, obtaining important image information, carrying out feature fusion on the important image information, and obtaining a final clear image.According to the invention, the attention mechanism and the neural network are combined; a network model for motion blurred image restoration is established, and effective and accurate restoration of the blurred image is realized.

A method for the generation of a set of conflicts for model-based system diagnostics is described, with which system a plurality of sensors is associated for the observation of variables indicative of operation conditions.The method starts from generating a complete set of Analytical Redundancy Relations (ARRs) in implicit form and, for each diagnosis instance: it performs a system simulation, computing the expected values for a first subset of Analytical Redundancy Relations (D-ARRs) including the relations involving only one system observation variable; it compares the expected and observed values of the system observation variables to identify the inconsistent variables, i.e. the unsatisfied redundancy relations, indicating a fault in at least one system component; and, when discrepancies are detected, it checks the consistency of each relation belonging to a second subset of Analytical Redundancy Relations (I-ARR), comprising relations involving a plurality of system observation variables, at least one of which exhibits discrepancies between expected and observed values; therefore identifying the support set of the unsatisfied Redundancy Relations as set of system conflicts.

The invention discloses a method for adaptively performing data encryption and authentication without an authentication center, which comprises the following steps that: (1) system parameters are determined; (2) information is sent and received and both parties are registered; (3) a sender carries out data encryption and authentication protection operation on a plaintext message to be sent to a receiver; and (4) the receiver carries out reading and verification operation on the received data message. The step (3) is realized by computing a key parameter, computing a signature parameter, outputting an electronic signature which is equal to zero when only carrying out data encryption on the plaintext message, outputting an encrypted ciphertext which is equal to the plaintext message when only carrying out electronic signing, computing the electronic signature when carrying out both data encryption and electronic signing, carrying out encryption operation on the combination of the plaintext message and the electronic signature, combining the ciphertext, the signature parameter and the electronic signature to form a data message and sending the data message. The plaintext message can be adaptively subjected to data encryption and content integrity authentication; and the problems caused by adoption of the authentication center and a plurality of functional parts in the conventional method are solved.

The invention provides a construction method of a deep convolutional neural network with resolution adaptability. The method is used for constructing a deep convolutional neural network model capableof adapting to scale features of target images with various resolutions, and is characterized by comprising the following steps of S1, setting a plurality of target scales according to the target images to form scale levels, and setting a target retrieval step length according to the scale levels; s2, obtaining a training image, and carrying out the standardization of the training image accordingto the order of the size of the training image, thereby obtaining a standardized training image; step S3, designing a deep convolutional neural network model for adapting to multi-resolution input andmulti-scale target detection, training the deep convolutional neural network model through the standardized training image so as to obtain an executable deep convolutional neural network model, wherein the deep convolutional neural network model in the step S3 comprises a feature map extraction part, a step-by-step down-sampling part, a branch convolution operation part and a prediction output part.

A turbogenerator main throttle valve opening predication control method based on a delaying observer is provided. The method comprises the following four steps: step 1: analyzing and modeling a turbogenerator main throttle valve opening control system; step 2: designing turbogenerator main throttle valve opening predication control; step 3: designing the delaying observer; and step 4: finishing the design. Aiming at a main throttle valve opening control system model, a control principle with a closed analytical solution is designed, and the output delaying observer is designed to correct a measured signal, so that the overall stability of a closed loop control system is guaranteed under the time varying delaying condition; meanwhile, the rapid and accurate tracking of a pre-set track by a power angle of a turbogenerator is realized.

The invention provides a maneuvering target tracking algorithm based on a road network. According to the algorithm, a tracking algorithm model adaptive strategy realized through road information is given by aid of a prior road information base, and a variable-structure multi-model method is adopted to realize ground multi-maneuvering-target tracking. In this way, state estimation precision of maneuvering target tracking can be improved, and the target missing rate is lowered; meanwhile, computing burden brought by multiple models adopted in a fixed multi-model algorithm is avoided, and operating time is greatly shortened; and the scheme has practical value in terms of ground target tracking.

The invention provides a steam turbine generator main steam valve opening degree prediction control method based on a non-linear interference observer. The method comprises the four major steps that 1, a steam turbine generator main steam valve opening degree control system carries out analysis and modeling; 2, the steam turbine generator main steam valve opening degree prediction control design is carried out; 3, the non-linear interference observer is designed; 4, the design is completed. The method aims at a main steam valve opening degree control system model, a control rule with the closed type analytical solution is designed, and then, the non-linear interference observer is designed for controlling the interference for compensation, so that the global stability of the closed loop control system is ensured under the condition with higher input interference, and meanwhile, the fast and precise tracking of a steam turbine generator power angle on the preset tracks is realized.

The invention discloses a time-lag-dependent matrix function-based time-varying delay wide-area electric power system stability analysis method; according to the method, the stability problem of the wide-area electric power system under the influence of time lag is further researched from the perspective of the functional structure, so as to obtain a conclusion with lower conservative property. The method comprises the following steps that firstly, a novel Lyapunov-Krasovskii function with time-lag-dependent matrix is constructed, and a novel time-lag matrix different from a constant moment isintroduced, so that the time lag information in the electric power system can be more utilized., the influence of the time lag change rate on the stable operation region of the electric power systemis fully considered, a critical effect of reducing the system conservation performance is achieved, and the conservative property of the stability criterion of the electric power system is effectivelyreduced. In addition, the integral term is segmented into two integral sections for segmented treatment, and a functional derivative is analyzed, so that more time lag information is utilized, the conservation of the result is reduced, and meanwhile, the calculation burden caused by excessive segmentation areas can be avoided.

The invention discloses a monitoring time sequence data prediction method based on a Prophet combination model, and relates to the technical field of engineering construction and operation and maintenance. Comprising the following steps: S1, arranging a sensor on an engineering site to obtain original time sequence data X0 = {x01, x02,..., x0i,..., x0n}, and marking festivals and holidays for interference time sequence data; s2, constructing a Prophet combination model, substituting original time sequence data X0 into a data preprocessing layer of the Prophet combination model, and performing iterative calculation to obtain time sequence data X '= {x' 1, x '2,..., x'i,..., x'n} after the gross error is deleted; the method comprises the steps of (S1) obtaining a Prophet combination model prediction layer, (S2) obtaining a Prophet combination model prediction layer, (S3) substituting X'into the Prophet combination model regression layer, and carrying out regression vacancy calculation to obtain complete time sequence data Y '= {y' 1, y '2,..., y'i,..., y'n}, and (S4) substituting the complete time sequence data Y' into the Prophet combination model prediction layer to obtain predicted time sequence data Z '= {z'n-k + 1, z'n-k + 2,..., z'n}. According to the method, the anti-interference performance of a prediction model is enhanced, and the prediction accuracy is improved.

The invention is suitable for the technical field of information security, and provides a QR code authentication method, a communication device and a storage medium, and the QR code authentication method applied to a user terminal comprises the steps: sending a public key query instruction to a server after determining that the data of a QR code comprises a digital certificate, and obtaining a server public key; verifying the validity of the certificate signature information through the server public key: when the server public key confirms that the certificate signature information is legal, determining that the digital certificate is legal, and at the moment, further confirming the validity of the QR code according to the certificate public key in the digital certificate and the QR code signature information of the QR code. According to the embodiment of the invention, the use safety of the QR code can be improved. In a digital medical scene, by using the QR code authentication method provided by the embodiment of the invention, the security and reliability of the obtained medical data can be ensured.

The embodiment of the invention relates to the technical field of communication, and discloses an image recognition method, electronic equipment and a computer readable storage medium. The method comprises the following steps: obtaining the similarity between an image block to be detected in a video frame and each candidate image block in a candidate object set; if the candidate image blocks of which the similarity with the to-be-tested image blocks is greater than a first preset threshold exist in the candidate object set, taking the candidate image blocks of which the similarity is greater than the first preset threshold as target candidate image blocks; if to-be-detected image blocks always exist in a first preset frame number continuous to the video frames, wherein the similarity between the to-be-detected image block and the target candidate image block is greater than a first preset threshold value all the time, and if the distance between the position information of the to-be-detected image block in the video frame and the position information of the to-be-detected image block in the video frame is smaller than the preset threshold, judging the to-be-detected image block tobe a real image block, so that the image can be identified more accurately through simple operation, and the success rate of image identification is increased.

The invention provides a vehicle lithium ion battery capacity estimation method based on surface temperature, which makes full use of battery temperature data, acquires and constructs a temperature rise curve, determines a capacity sensitive interval from the temperature rise curve, and samples a health factor sequence for capacity estimation. According to the method, the calculation burden causedby extracting specific features from the temperature signal is avoided; the temperature acquisition signals can be fully utilized to realize capacity estimation, supplement can be provided for an existing capacity estimation method based on voltage and current, the reliability and accuracy of capacity estimation in a battery management system are improved, and new hardware does not need to be added, so that the method has many beneficial effects which do not exist in the prior art.

The invention provides a distributed sensor network consistency filtering method with unknown colored noise, which comprises the following steps of: learning state noise and measurement noise parameters of each node in a distributed sensor network according to a group of given training data; meanwhile, performing state extension, extending learned colored noise parameters to a state vector, and finally, state estimation is completed by using a consistency Kalman filtering algorithm, so that estimated values of all sensor nodes are converged to be globally consistent. To improve computational efficiency, only local measurement noise and state noise are integrated into an extended state for each node. Simulation results show that the unknown colored noise in the distributed sensor network can be effectively estimated, and the accuracy of distributed estimation is improved. The method can be applied to the fields of distributed sensor network radars, infrared target tracking, mobile robot positioning and the like.

The invention relates to the technical field of image recognition, and provides a picture filtering method and device, electronic equipment and a storage medium. The method comprises the steps: carrying out the clustering and filing of obtained interference picture data, carrying out the duplicate removal of a first feature value corresponding to the interference picture data, obtaining an initialblacklist file, and enabling the initial blacklist file to comprise the duplicate-removed first feature value; obtaining to-be-archived picture data, and calculating feature value similarity betweena second feature value of the to-be-archived picture data and the first feature value; and filtering the to-be-archived picture data of which the feature value similarity reaches a preset similarity threshold to obtain a blacklist archived picture library. According to the embodiment of the invention, the data volume of data calculation can be reduced, and the filing efficiency of picture data isimproved.

Embodiments of the present invention relate to the technical field of communications, and disclose an image recognition method, an electronic device, and a computer-readable storage medium. In the present invention, the similarity between the image block under test in the video frame and each candidate image block in the candidate object set is obtained; if there is a candidate object whose similarity with the image block under test is greater than the first preset threshold in the candidate object set Image block, the candidate image block whose similarity is greater than the first preset threshold is used as the target candidate image block; if in the first preset number of frames continuous with the video frame, there is always a similarity with the target candidate image block greater than the first The image block to be tested with a preset threshold value, and the distance between the position information of the image block to be tested in any frame in the first preset number of frames and the position information of the image block to be tested in the video frame is smaller than the predetermined distance. If the threshold is set, the image block to be tested is determined as a real image block, so that the image can be recognized more accurately through simple operations, and the success rate of recognizing the image can be increased.

The present invention provides a calculation unit and a determination method for the distribution index of carbon emission flow in a power system. The calculation unit collects local active power flow and direct carbon emission information, and two adjacent calculation units communicate iteratively to perform system-wide Calculation of carbon emission flow distribution; the determination method is to set up calculation units on each power system carbon emission flow distribution equipment in the target power system; the calculation unit collects local active power flow and direct carbon emission information; iteration between adjacent calculation units Communication, calculation of system-wide carbon emission flow distribution values. The calculation unit and determination method proposed by the present invention realize the accurate distribution of the carbon emission flow of the whole system obtained through fast calculation, avoid increasing the calculation burden of the central server, provide users with accurate and reliable carbon emission flow of the power system, and reduce carbon emissions. The allocation of emission responsibilities is more equitable, and at the same time helps to eliminate carbon leakage, and stimulates the enthusiasm of users to participate in the power system to reduce carbon emissions.

The invention relates to a power distribution area three-phase load data transmission method based on dynamic cloud coding. According to the method,encoding is performed according to data such as voltage, current, phase, fault signal, network address and / or physical address in a power supply network; a dynamic coding table is set in a power distribution area three-phase load data transmission process; and the time interval of the dynamic coding table is optimized and updated according to the characteristics of a power distribution area in the power supply network, and the transmission efficiency of power distribution area three-phase load data improved.

The present invention relates to a fuel cell hybrid power energy management method and system for a vehicle. The method updates the working condition table at intervals, and calculates the demand power transfer matrix generated based on the current updated working condition table data and is generated based on the current new working condition table. The Pearson correlation coefficient ρ between the demand power transfer probability matrix and the matrix generated based on the previous working condition table is used to determine whether to update the energy management strategy; when the absolute value of the Pearson correlation coefficient ρ is less than the reference threshold λ, the Q‑ The learning algorithm uses optimization iterations to calculate a new optimal energy management strategy to update the energy management strategy and apply it to the system; when the absolute value of ρ is greater than or equal to λ, the original energy management strategy is retained. The invention uses the Pearson correlation coefficient of the required power transfer probability matrix as a basis to update the energy management strategy when the working condition changes, thereby enhancing the real-time performance of the energy management strategy and improving the energy utilization rate.

The invention discloses a multi-cell filtering method for pose positioning of a mobile robot in a production workshop, and belongs to the technical field of manufacturing process control. The method locates the mobile robot by adopting the filter method of the fully symmetrical polytope, so that the geometric area of ​​the fully symmetrical polytope is minimized, thereby effectively reducing the conservatism of identification and increasing the accuracy of robot pose positioning. Furthermore, the dimensionality reduction of the fully symmetrical polytope is performed through singular value decomposition, which simplifies the final calculation amount, avoids the calculation burden brought by the increase in dimension, and greatly improves the calculation efficiency. The pose positioning method provided by this application is applied to the pose positioning of mobile robots in the production workshop, which improves the positioning accuracy and the efficiency of state estimation, thus improving production capacity; the method of this application also provides technical inspiration for the next intelligent manufacturing .

The invention discloses a soft landing attitude and orbit integrated control method based on a parallel variable thrust engine, and belongs to the technical field of spacecraft guidance control, the method comprises the following steps of: S1, limiting the amplitude of guidance command thrust according to the total variable thrust capacity range of an engine unit, and calculating the minimum normsolution of the command thrust of each engine by utilizing the limited guidance command thrust to serve as the nominal command thrust of each engine; S2, according to the instruction torque, determining the final adjustment amount of each engine on the basis of the nominal instruction thrust; and S3, determining a thrust instruction of each engine according to the nominal instruction thrust of each engine in the S1 and the final adjustment amount of each engine on the basis of the nominal instruction thrust in the S2.