166results about How to "Good recognition effect" patented technology

The invention provides a station area electricity utilization monitoring method based on bad electric quantity data identification. The station area electricity utilization monitoring method based on bad electric quantity data identification comprises the following steps of: calculating a line loss rate according to the acquired meter-reading electric quantity data of a station area during a time period, and initially judging whether an abnormal electricity utilization phenomenon exists or not; in the case that the abnormal electricity utilization phenomenon is found, turning to a B-spline function identification method, and identifying the line loss rate, and abnormal data in the daily electricity consumption data of each user; and locating suspicious electricity-stealing users by comparing the abnormal time periods corresponding to the line loss rate data of the power grid of the station area and the daily electricity consumption data of each user. The method provided by the invention only needs the meter-reading electric quantity data of the power grid of the station area, as well as is low in requirements on original data, strong in engineering applicability, and easy to be popularized and applied in the existing power grids of station areas in China. The method provided by the invention is adequate in theoretical basis and good in identification effect by identifying and locating the suspicious electricity-stealing users based on a B-spline function.

The invention belongs to the technical field of digital communication signal modulation, and discloses a CNN convolutional neural network-based digital signal automatic modulation identification method. The method comprises the following steps of performing cyclic spectrum analysis on the received digital signal, extracting an axial projection profile map of an amplitude normalization cyclic spectrum, and obtaining a one-dimensional feature vector x belonging to Rn * 1; Carrying out dimension reduction processing on the feature vector by utilizing an improved Fisher algorithm to obtain a low-dimensional feature vector y belongs to xm * 1; And finally, designing a deep CNN network structure, and determining network initialization parameters. According to the method, a Keras deep learning framework is utilized, an existing network layer function is directly called, and a deep network structure is built; An early stop strategy is adopted in the network training process, the network overfitting phenomenon is effectively prevented, after network training is completed, the training effect is verified through a test data set, and automatic signal modulation recognition is completed. For an MQAM signal, when the signal to noise ratio is greater than 0dB, the recognition rate reaches 97% or above; For an MPSK signal, the signal to noise ratio is greater than or equal to-. And when the identification rate is 4dB, the identification rate reaches 95%.

The invention provides a verification code identification method based on window sliding and the convolutional neural network for traditional English letter+digit+Chinese character verification code pictures. According to the method, firstly, a small quantity of verification code pictures are acquired, after noise reduction, to-be-identified character sets of the verification codes are extracted,each character set is turned and distorted, and background noise is added to each character set; secondly, the convolutional neural network is then utilized for the character sets, and each characterset is trained to acquire a single character classifier; and lastly, after the to-be-identified verification code pictures are pre-processed, connected domain segmentation is carried out, for each connected domain, window sliding is carried out, the trained single character classifiers are utilized for classification, and the final identification result is acquired. The method is advantaged in that a problem of segmentation difficulty caused by overlapping of the verification codes and excessive random character jitter can be effectively solved, through the method of employing the small quantity of verification code pictures, extracting the character sets from the pictures and autonomously generating the correlation training sets, data acquisition and data marking cost is greatly reduced.

The invention discloses a residual convolutional neural network and PCA dimensionality reduction fused SAR automatic target recognition method. The method includes: acquiring an SAR target image dataand labeling a target category of the SAR target image data to form a training set; performing enhancement and expansion and pre-processing on training set image data, and constructing a residual convolutional neural network; inputting training samples into the network to perform training; inputting the training samples into the trained network model, and using feature vectors obtained after the samples pass all the hidden layers as a new training set; using the PCA dimensionality reduction method to perform dimensionality reduction on the obtained feature vectors, and then inputting the feature vectors into an SVM classifier to perform training; and finally pre-processing samples to be recognized, inputting the pre-processed samples to be recognized, obtaining feature vectors, performingPCA dimensionality reduction on the feature vectors, and using the trained SVM classifier to perform recognition. The method solves the problem that the existing SAR automatic target recognition technology has low recognition accuracy.

The invention discloses a manipulator dynamic model identification method based in an improved artificial bee colony algorithm. A manipulator linear dynamic model taking joint friction into consideration is established through utilization of an improved Newton-Euler method. A dynamic parameter identification algorithm is designed through importing of the improved artificial bee colony algorithm. An UR industrial robot is taken as an experiment subject; through design of a stimulation track, joint movement of the industrial robot is stimulated; the joint movement data is sampled; after torque is filtered, parameter identification is carried out through utilization of the improved artificial bee colony algorithm; the dynamic parameter estimation of the UR industrial robot is realized; and the dynamic model is verified according to the torque prediction precision. An experiment shows the accuracy and effectiveness of the industrial robot dynamic model the identified by the invention.

The invention discloses a phasor measurement unit (PMU) measurement data-based power line parameter identification and estimation method, and belongs to the technical field of power system line parameter identification and estimation. The method comprises the following steps of: inputting basic data of an identified and estimated line by using a computer through a program, identifying error parameters of the line according to the PMU measurement data of multiple time intervals at two ends of the line, and estimating accurate parameters of the line to acquire estimation values of the accurate parameters of the line. A mean value-based identification index and parameter estimation method is constructed by using the PMU measurement data of multiple time intervals at two ends of the line, anda variance coefficient is used as a convergence precision criterion; and the method is simple, has good identification effect, high estimation precision and high calculation speed, and is convenient for popularization and application. The method can be widely applied to identification and estimation of the centralized parameters of the power line, the two ends of which are provided with PMUs; andthe method is particularly suitable for identification and estimation of the parameters of medium length high-voltage and ultrahigh-voltage power transmission lines.

The embodiment of the invention discloses a target tracking method, which comprises the steps of using a presetting algorithm to extract the corner feature of a reference image to obtain the first feature point set of the reference image, removing abnormal points which exceed the range of images to be matched from the first feature point set to obtain a second feature point set, matching the second feature point set with the images to be matched and finishing the target tracking of the images to be matched according to matching results. The embodiment of the invention additionally discloses a target tracking device. By using the target tracking method and the target tracking device, the extracted corner feature has good identifiability and stability, the matching speed is greatly improved, the feature point set of the reference image is optimized, the corners which exceed the range are removed, the problem that the matching cannot be realized due to a reason that the feature points exceed the range in the prior art is solved and the matching reliability is improved.

The invention discloses an intelligent method for identifying the driving state of a hybrid electric vehicle, belonging to the electric vehicle control technical field. Neural network is mainly adopted by the method for the identification. The working process is mainly divided into two stages of a learning period and a working period. During the learning period, the standard driving condition of the vehicle is collected firstly and segmented; each sample is calculated to obtain a series of sample parameters; the sample parameters are inserted in a neural network calculation formula to obtain parameters required for the intelligent identification control. During the working period, the cut ridge filtering processing is implemented on the obtained speeds; then the recursive average filtering processing is carried out; the speeds within a certain period of time are stored and are calculated; the calculated results are inserted in the neural network calculation formula to obtain the current driving state of the vehicle. With obvious identification effect on the vehicle driving state, the method for identifying the driving state of the hybrid electric vehicle can help a controller of the electric vehicle to reasonably and effectively regulate control parameters, thereby further improving the fuel economy and emission performance of the electric vehicle.

The invention discloses a device for identifying kinetic parameters of terminal loads of a six-degree-of-freedom robot. The device comprises a six-axis industrial robot, a plurality of loads of different masses and a real-time control system, wherein the plurality of loads of different masses are alternatively arranged on the terminal of the six-axis industrial robot, the real-time control system is used for implementing millisecond level real-time motion data collection of the robot, and collected real-time motion data include an encoder value and a torque value. The invention further discloses a method for identifying the kinetic parameters of the terminal loads of the six-degree-of-freedom robot. According to the device and the method, data sampling is performed by adopting an optimized excited track, a kinetic model is established according to a Lagrange equation, and identification parameters are solved by adopting a weighted least square method; a sensing system adopts the efficient real-time control system, the millisecond level real-time data collection of the robot including displacement, motor output and the like can be realized, the data collection quantity can be extended, the characteristic of variability is provided, and the identification of the kinetic parameters of the terminal loads of the industrial robot is satisfied.

The invention relates to a sound signal transduction and prediction method of a GTAW (gas tungsten arc welding) welding fusion penetration state. The prediction method comprises the following steps of: utilizing a sound sensor to collect electric arc sound of a GTAW welding dynamic process, carrying out feature extraction on the electric arc sound of the collected GTAW welding dynamic process, and obtaining an electric arc sound characteristic; building a HMM(hidden markov model) model according to the obtained electric arc sound characteristic, and carrying out model training and sound fusion penetration recognition; and forecasting the GTAW welding fusion penetration state. The prediction method provided by the invention can effectively carry out real-time detection and control on welding fusion penetration.

The invention discloses an electrocardiogram (ECG) intelligent analysis method and system based on a deep neural network. The method comprises a training stage and a detection stage; the training stage comprises the following steps: annotating a horizontal layer label and a longitudinal layer label on image information of each collected N-channel static ECG; training to obtain N convolutional neural network models for identifying the horizontal layer and convolutional neural network models for identifying the longitudinal layer; the detection stage comprises the following steps: regarding theN feature sequences of the collected to-be-detected N-channel static ECG and the feature sequence segmented by taking the heartbeat period as the input of N conventional neural network models for identifying the horizontal layer and the convolutional neural network models for identifying the longitudinal layer, thereby obtaining the horizontal identification abnormal analysis and the longitudinalidentification abnormal analysis. All channels are respectively learned and judged by using the convolutional neural networks; experimental results show that the method has better identification effect. The method disclosed by the invention is stronger in operability, better in network generalization capacity, and high in ECG correct identification rate.

The present invention discloses a video human body behavior recognition method based on motion saliency. The method comprises: performing motion saliency detection on a behavior video frame by using a motion saliency detection algorithm, so as to obtain a motion saliency image; by adopting non-maximal suppression (NMS) sampling algorithm, calculating a motion saliency area candidate frame based on the motion saliency image; clipping the video frame around the motion saliency area candidate frame to acquire an image block containing a human body behavior; scaling the image block obtained through clipping into a size required for input data of a depth convolutional neural network; by using the depth convolutional neural network, extracting a depth feature of the human body behavior based on the scaled image block; and performing feature classification on the depth feature of the human body behavior to obtain a human body recognition category result. According to the method, the image block required by the convolutional network is constructed around the behavior saliency motion area, so that the areas of human body behavior changes are effectively captured, the depth convolutional features of human body behaviors with good recognition are effectively extracted, and the accuracy of human body behavior recognition is effectively improved.

The invention provides a face recognition model training method and device and a face recognition method and device. The face recognition model training method comprises the following steps of: inputting a comparison image set into a first neural network and extracting a first feature vector for each comparison image in the comparison image set by using the first neural network; inputting a sourceimage set and a target image set into a second neural network, carrying out feature learning on source images in the source image set and target images in the target image set, and extracting a second feature vector for each source image in the source image set; inputting the second feature vectors into a face classifier to obtain classification results; training the second neural network and theface classifier on the basis of comparison results between the first feature vectors and the corresponding second feature vectors and the classification results; and repeatedly training the second neural network and the face classifier to obtain a face recognition model. The method is capable of effectively recognizing face images with relatively bad quality.

The invention provides a polarization board with polyvinyl alcohol resin and a stretched polyethylene glycol terephthalate mask laminated on one single face of the polarization board through a first adhesive layer wherein the refractive index of b arranged in the stretched polyethylene glycol terephthalate mask is defined as nx; the refractive index of the orthogonal direction with slow axis is defined as ny; the refractive index of the thickness direction is defined as nz. The coefficient of Nz represented by formula (nx-nz) / (nx-ny) is above 4.

The present invention discloses a face recognition algorithm configuration based on unbalance tag information fusion. The face recognition algorithm configuration comprises two layers of configurations (L1 and L2); the L1 is configured to train the face data and corresponding tag information to obtain an initial face recognition model 1 through adoption of a supervised learning algorithm, train no-tag data through adoption of an unsupervised method, alternately optimize the face data tag information and the parameters of the model 1, and obtaining a final face recognition model 1 through calculation after multiple iterations; the L2 has a thought opposite to the L1, the L2 is configured to initiate parameters of a face recognition model 2 at random, then perform unsupervised training to update the parameters of the model 2, input tag data, continuing training through adoption of the supervised learning algorithm, and finally obtaining the face recognition model 2; and the model 1 and the model 2 are fused to obtain a final face recognition model. Through combination of the advantages of a supervised learning algorithm and an unsupervised learning algorithm, the face recognition algorithm configuration based on unbalance tag information fusion gives full play to mass of no-tag data to allow the algorithm to have excellent recognition capability in a special scene and adapt different scenes.

The invention provides a method for positioning a sound source in the airflow environment, belonging to the technical field of sound source localization. The method comprises the following steps of: setting up a microphone array, and collecting acoustic signals in the airflow environment; according to coordinates and other geometric parameters of array microphones, solving an acoustic propagation path from each microphone to a scanning point by means of an Amiet model, so as to obtain an array manifold matrix A in the airflow environment; estimating the number of sound sources, constructing a cross-spectrum matrix after noise reduction, and establishing a value function among a scanning point sound pressure matrix, the array manifold matrix and the cross-spectrum matrix after noise reduction; solving an acoustic pressure matrix by convex optimization iteration, finally drawing out a sound pressure reconstruction picture, and obtaining the sound source position information. The invention helps accurately locate the position of the sound source in the airflow environment and correct the positioning deviation which may be caused by airflow, achieves high positioning resolution, realizes multi-coherent sound source localization, reduces the calculation quantity in the convex optimization iteration process, reduces the number of side lobes in the reconstruction cloud picture, and achieves good sound source localization effect.

In soft measurement technology frame, the easy-to-measure reliable pressure difference fluctuation signal is adopted as auxiliary variable in detecting flow pattern of two phase flow as the main variable. The acquired pressure difference signal is wavelet analyzed, corresponding characteristic values are extracted, and the characteristic values are used in fuzzy identification of the flow pattern of two phase gas / liquid flow in horizontal pipe. The key soft measurement model is established by means of combining the wavelet analysis technology and fuzzy pattern identifying technology. Then, one flow pattern identifying system for two phase gas / liquid flow is established, and the system consists of test pipeline, pressure sampling hole, pressure sampling pipe, pressure difference sensor, amplifier, data acquiring card, microcomputer and printer. The detected pressure difference signal is processed in the system for pattern identification of two phase gas / liquid flow.

The invention discloses a vehicle identification method and a vehicle identification system for searching images by images. The method comprises the following steps of obtaining a to-be-retrieved image; identifying a vehicle in the to-be-retrieved picture to obtain macroscopic feature information of the vehicle, the macroscopic feature information comprising a vehicle position, a shape and a size;according to the macroscopic feature information of the vehicle, narrowing the recognition range in the to-be-retrieved picture, analyzing the to-be-retrieved picture to obtain multi-dimensional feature information, the multi-dimensional feature information comprising visual word bag features and vehicle depth features; performing deep feature fusion on the multi-dimensional feature information to obtain global features; and obtaining similar images in the image library through the feature index according to the global features, and sorting the similar images according to the image similarityto obtain a retrieval result. According to the method and the system, a plurality of feature fusion and deep neural network methods are used, so that the vehicle searching precision is effectively improved. Through distributed vehicle retrieval, the search performance of mass data is effectively improved.

The invention discloses a rub-impact sound emission fault position identification method based on near field sound source focusing positioning. According to the method, sub-band decomposition of a sound emission signal acquired by a rub-impact sound emission test device is carried out to extract a characteristic signal; a subspace frequency focusing method is utilized to carry out frequency focusing of a sound emission characteristic signal, and decoherence processing on multiple coherent signals is carried out; lastly, a space spectrum estimation method is employed to carry out further precise sound source positioning. Through the method, fast fading, coherent and multi-modal problems of the broadband sound emission signal can be solved, so the rub-impact fault source position can be effectively identified. The method is advantaged in that the generation position of the rub-impact sound emission signal can be accurately identified, precision and complexity indexes are better compared with a positioning algorithm in the prior art, and the method is an effective rub-impact fault position detection method.

The invention discloses a cutter wear state recognition and prediction method based on a hidden Markov model. The method comprises: extracting signal features; obtaining 41 signal characteristic quantities through screening; dividing the cutter abrasion stage into an initial abrasion stage, a stable abrasion stage, a rapid abrasion stage, a serious abrasion stage and a wear-out stage; constructinga wear stage identification model; constructing a cutter remaining service life prediction model; performing online wear stage identification and residual service life prediction: collecting a tool machining process signal in real time in an online link, extracting characteristic quantities according to a model training process, and respectively inputting the characteristic quantities as observation sequences into a wear stage identification model and a residual service life prediction model for stage identification and residual service life prediction; and model training updating: repeatingthe steps S1 to S6 on the new signal and the tool wear state data along with monitoring data accumulation, and updating the wear stage identification model and the remaining service life prediction model. Reference is provided for online identification and prediction of the tool wear state.

The invention provides an anti-dazzle which has a resin base material film and a hard coating layer laminated on the resin base material film surface and having micro concavo-convex shapes on the surface; the hard coating layer is made of a light transparent resin with at least one kind of resin particles dispersed therein; the hard coating layer has a surface haze of 1%-10% and an inside haze of 10%-20%; the volume change rate of the resin particles is above 10% when dipped in vinyl acetate for under 23 degrees centigrade for 48 hours. And the invention further provides an anti-dazzle polarizer and an image display apparatus using the anti-dazzle film.

This invention discloses an image identification method based on the Gabor phase mode including: picture selection: exchanging Gabor to the being compared images to get Gabor character images for them, picking up global and local Gabor phase modes from each of the character image and evaluating the character modes to connect the evaluation results in series to high dimension character vectors to be compared to get the similarity degree among the vectors to identify images.

The invention discloses a full-bore multistage-key switch type fracturing slide sleeve. By adoption of the full-bore multistage-key switch type fracturing slide sleeve, the problems that existing slide sleeves are complex in structure and low in working reliability are solved. The full-bore multistage-key switch type fracturing slide sleeve specifically comprises a slide sleeve actuator and a slide sleeve switch mechanism. The slide sleeve actuator comprises a slide sleeve main body, an inner drum, a pressure adjustment piston, a base and a pulling rod. The slide sleeve switch mechanism comprises a throwing block body, a switch body, a clamping block, an outer slide block, an inner slide block, a slide block spring, a switch spring and an upper cap. According to the full-bore multistage-key switch type fracturing slide sleeve, the circumferential rotating angle of the inner drum is small in the opening process of the slide sleeve, the total rotating angle of the inner drum in the circumferential direction is zero degree after the slide sleeve is opened and closed each time, error superposition in the rotating process is avoided, the working reliability is improved, the recognition effect is good, opening and closing of slide sleeves in different stratums can be controlled by different switch mechanisms, multiple times of opening and closing of sleeves in one single stratum can be achieved, and the comprehensive performance of the slide sleeve is good.

The invention discloses a transformer substation patrol robot digital type instrument identification algorithm. The transformer substation patrol robot digital type instrument identification algorithm comprises steps that a transformer substation patrol robot is used to acquire a device image, and is used for the pretreatment of the image; a digital area is located automatically; the inclination of the digital area is corrected; the segmentation operation of the digital area is executed by adopting a contour detection algorithm, and the digital positioning is executed according to a contour acquired after the segmentation operation, and then a single digital image is acquired; the digital images of the various forms on the scene of the transformer substation are acquired, and a digital identification training sample set is established, and an integrated classifier is formed; the trained integrated classifier is used for identification of a single digit, and a final identification result is acquired by sequencing the digits according to the coordinate of the above mentioned digit in the image. The transformer substation patrol robot digital type instrument identification algorithm is advantageous in that when surfaces of on-site instruments are dirty, and are inclined at a certain angle, a good identification result is still acquired, and the digital type instruments having certain defects are identified correctly, and in addition, the transformer substation patrol robot digital type instrument identification algorithm is suitable for the digits of the various forms in the transformer substation.

The purpose of the present invention is to provide a substrate with a transparent electrode in which the pattern cannot be readily visually identified even when the transparent electrode layer has been patterned, and a method for manufacturing the same. This substrate with a transparent electrode has, on at least one of the surfaces of a transparent film, a first dielectric layer, a second dielectric layer, a third dielectric layer, and a patterned transparent electrode layer, in the sequence listed. The first dielectric layer is a silicon oxide layer having SiOx as a principal component. The second dielectric layer is a metal oxide layer having an oxide of a metal as a principal component. The third dielectric layer is a silicon oxide layer having SiOy as a principal component. The transparent electrode layer is an electroconductive metal oxide layer having an indium-tin composite oxide as a principal component. The refractive index (n1) of the first dielectric layer, the refractive index (n2) of the second dielectric layer, and the refractive index (n3) of the third dielectric layer satisfy the relationship n3 < n1 < n2. Each of the dielectric layers and the transparent electrode layer preferably has a film thickness and a refractive index within a predetermined range.

The invention discloses an unstructured road identification method, which comprises the steps of: S1, acquiring a color image in front of a mobile robot in the running process, and preprocessing the acquired color image; S2, performing region segmentation on the image obtained by preprocessing; S3, performing edge detection on the preprocessed image; S4, and performing fusion extraction on the image obtained after edge detection, so as to obtain road regions. The unstructured road identification method is implemented by the steps of: firstly, acquiring the color image in front of the mobile robot, and preprocessing the acquired color image; secondly, subjecting the image to region segmentation, and extracting edge features; and finally, fusing region segmentation and edge feature information to obtain the road regions. According to the unstructured road identification method, an OTSU algorithm is adopted in region segmentation, a multi-directional sobel operator edge detection algorithm is adopted in edge detection, the time complexity is small, the algorithms have good identification effect for a new environment, the robustness thereof is greatly enhanced, and the precision and validity of calculation are improved.

The invention discloses an SCADA data calibration method based on WAMS information. The method includes the steps of firstly, performing self-identification on relevant data of a WAMS system to verify the correctness of the data itself; secondly, linearly estimates a state value of each node in the system through a PMU, substituting the state value of each node into an SCADA measurement equation to obtain a calculation result corresponding to the SCADA, then obtaining a difference between the SCADA measured value and the SCADA measured value by the PMU estimation, and also calculating a standardized residual value to achieve the purpose of detection; and thirdly, subjecting the WAMS data to measurement transformation to form SCADA data at the equivalent time by using a nonlinear model, updating a Jacobian matrix with the iteration, and replacing bad data of the SCADA by the data communication between the two systems to achieve the purpose of calibration. The method applies PMU measurement information to the bad data detection and identification, overcomes the phenomenon of residual contamination and so on, and also has good detection and identification effects on the problem of the occurrence of bad data to the key item measurement in an SCADA system. The operation safety and reliability of a power system are improved.

The invention relates to a risk assessment method and device, computer equipment and a storage medium. The method comprises the steps of obtaining current position information of a to-be-evaluated user, and determining a regional grid and a density clustering grid which take the current position information as a center; acquiring historical user behavior tracks associated with the regional grids and the density clustering grids; and determining a risk assessment result of the to-be-assessed user based on the historical user behavior track associated with the regional grid and the density clustering grid and the risk level data. According to the historical user behavior track and risk level data of the regional grid and the density clustering grid near the position of the user, the active area of the gang fraudulent user is accurately identified, the hidden fraudulent gang is found, and the associated risk case is quickly identified, so that the risk assessment result of the user is obtained, and the identification effect is better.

The invention discloses a web crawler identification method and a web crawler identification device. The web crawler identification method comprises steps that a crawler identification instruction isgenerated; according to the crawler identification instruction, a user identification set and an access time set corresponding to each user identification of the user identification set in a preset time period are acquired; an interval between two adjacent access times in the access time set is calculated to acquire an interval length set; according to the interval length set, the web crawler is identified from the user identification set, and therefore the web crawler can be identified accurately, a missing report rate and a false report rate are reduced, and identification effect is good.