49results about How to "Rich sample" patented technology

The embodiment of the invention discloses a language model training method and device, electronic equipment and a computer readable medium. According to the embodiment, the language model training method is characterized by comprising the steps of: selecting a first text sample set based on a natural language processing task, and training a first pre-training model through the first text sample set to obtain an initial language model; testing a test text in a preset test text set by using the initial language model, and generating a second text sample set based on a test result and the test text; training the first pre-training model by using the second text sample set to obtain a second pre-training model; and training the second pre-training model by using the first text sample set to obtain a target language model. According to the embodiment, the model applied to the natural language processing task can be obtained, and the model can improve the accuracy of the processing result.

The invention relates to a detection method for specific antibodies IgM of mycoplasma pneumonia and influenza viruses based on a micro-fluidic chip and application of the detection method. The method is characterized by taking the micro-fluidic chip with a plurality of pipelines as a reaction platform and detecting by using a chemical luminescent signal so as to independently or simultaneously detect the serum specific antibodies IgM of mycoplasma pneumonia and influenza viruses. As the micro-fluidic chip is combined with chemiluminiscence, the detection method integrates the advantages of rapidness, high efficiency and small sample amount of the micro-fluidic chip and high specificity of chemiluminiscence, also has the advantages of simplicity in operation, low cost, rapidness and the like, is a multi-target, real-time and high-sensitivity detection method and can be applied to rapid diagnosis of pneumonia and influenza.

The invention discloses a car-following trajectory dynamic acquisition method and system. The trajectory of a front car is obtained by collecting the trajectory of a rear car and the distance between the front and rear cars, wherein the distance between the front and rear cars is calculated based on images shot by a binocular camera mounted on the top of the rear car. The method mainly comprises the following steps: carrying out distortion correction and stereo correction on original images shot by the binocular camera; selecting a target area in one image, and realizing automatic matching of targets in two images based on a template matching algorithm; carrying out calculation according to the triangulation principle to obtain the distance between the actual target area and the binocular camera; keeping the original images shot by the binocular camera, shooting time and the distance corresponding to the target front car; and when carrying out target distance acquisition at the next moment, automatically tracking the target area in the image in the next moment through a target automatic tracking algorithm. The car-following trajectory dynamic acquisition method and system are low in equipment cost, high in data acquisition precision, high in applicability, rich in data content and simple in operation and have a good application prospect in the traffic data acquisition field.

The invention discloses a method for constructing a transient state fault feather sample library of a high-voltage direct-current power transmission system. The method is developed on the bases of the RTDS simulation technology, the circuit simulation technology and a fingerprint feather algorithm for the specific high-voltage direct-current power transmission system, fault information is obtained through the method that practical faults, RTDS simulation faults and circuit simulation faults are collected on site, the fault information comprises SER warning information, protective action recording, fault recorder recording, a lightning arrester motion situation, a fault traveling wave monitoring result, polar line voltage and current data, the fingerprint algorithm is utilized to carry out feature extraction on the recording result and the waveform, the recording result and the waveform are stored in a database, and the transient state fault feather sample library is established. The transient state fault feather sample library is combined with the transient state fault recognition technology of the direct-current power transmission system to be applied, the fault analysis speed, the fault distinguishing speed and the fault exclusion speed can be substantially increased, and the transient state fault feather sample library has good popularization prospects.

The invention discloses a water quality sampling device for lakes, and relates to the field of water quality sampling equipment. According to an adopted technical scheme, the device comprises an unmanned aerial vehicle for providing flight power, connecting rods are hinged to four corners below the unmanned aerial vehicle respectively, and a floating platform for floating on the water surface is hinged to the lower ends of the connecting rods; a sampling apparatus for collecting a water quality sample is arranged below the floating platform; the floating platform comprises two floating platesarranged in parallel, a support is arranged above the floating plates, and the two ends of the support are fixedly connected with the floating plates respectively; and the unmanned aerial vehicle is connected above the support through the connecting rods. The water quality sampling device has the beneficial effects that: the unmanned aerial vehicle can be remotely controlled to automatically navigate to a designated position to collect a water sample and sample water in the middle layer or deep layer of a lake, the whole sampling process is efficient and time-saving, and the sampling efficiency and breadth are greatly improved.

The invention discloses an early fault diagnosis method for complex equipment. The early fault diagnosis method for the complex equipment comprises the specific steps: extracting collection signals through a sensor; carrying out wavelet transform on the extracted collection signals for de-noising; and establishing a fault diagnosis model based on a BP (Back Propagation) neural network and a fault diagnosis model based on an SVM (Support Vector Machine) according to mechanical features and circuit features, and carrying out fault diagnosis. A kernel function in the fault diagnosis model based on the SVM is a polynomial kernel function, a radial basis function (RBF) or a Sigmoid kernel function. According to the early fault diagnosis method, the fault diagnosis of the complex equipment is divided into a fault diagnosis with the mechanical features and a fault diagnosis with the circuit features, and the fault diagnosis models are established respectively according to the mechanical features and the circuit features, so that mechanical tests are easy, many samples can be obtained, quick convergence can be realized by using the BP neural network, and accuracy is higher; furthermore, the circuit sample data is less. With the adoption of the advantages of small samples of the SVM, the fault diagnosis of the complex system, such as the complex equipment, can be realized.

The invention discloses an intersection recognition method based on a GoogLeNet neural network. The intersection recognition method comprises the following steps: S1, sampling: selecting road networkdata of main cities throughout the country as a sampling source of a complex intersection sample; S2, constructing a Delaunay triangulation network: carrying out the initial positioning of the complexintersection, and preliminarily determining the central position and spatial range of the complex intersection; S3, enhancing the sample size: enhancing the sample size through sample simplification,sample rotation and mirroring; according to the method, a research hotspot GoogLeNet neural network in the field of machine vision is introduced into complex intersection recognition, and rapid and high-quality positioning of the OSM data complex intersection is realized in a mode of combining vector data and raster images.

The invention discloses a method and a system for obtaining smart city evaluation indexes by utilizing big data. The method comprises the following steps: obtaining data related with each index Ik from an NNI (network to network interface), a UNI (user networks interface) and an ANI (access network interface) by a big data acquisition, storage and calculation module respectively; comparing the obtained data with stored historical data, if a deviation exceeds a threshold value, discarding the data Di,j, or otherwise, storing the data Di,j; performing normalization processing on the data Di,j and performing calculation to obtain an assigned value Vk of the index Ik; performing weighting calculation on the assigned value Vk of the index Ik according to an evaluation function to obtain comprehensive result data ISC of a smart city evaluation index system; outputting a comprehensive report of smart city evaluation, an analysis report of a single index and improvement comments and suggestions of smart city construction. A big data sampling method is adopted, the sample amount is relatively high, and the smart city evaluation indexes can be accurately obtained.

The invention discloses a method for controlling stroke of a clothes dryer, and the method comprises the following steps of pre-storing the correspondence relationship between the descent speed of a drying rod and the running current of a constant torque motor under different weight loads as a first relationship model, setting the descent stroke of the drying rod in advance, obtaining the startingtime and running current of the motor after the motor is started, querying the first relationship model to get the corresponding descent speed of the drying rod, obtaining the running time of the motor by dividing the descent speed of the drying rod by the descent stroke of the drying rod set in advance, calculating the terminating time of the motor through the cooperation with the starting timeof the motor, and controlling the motor to stop at the terminating time of the motor. Through the above method, without adding components or modifying the structure or circuit of the clothes dryer, the stop position of the drying rod of the clothes dryer can be automatically and accurately controlled.

An embodiment of the invention relates to the technical field of virtual reality and discloses a positioning method and a handle. In the invention, a method for positioning a lamp group-equipped object to be positioned comprises the following steps: identification codes corresponding to all lamps in a lamp group are obtained; position information of all lamps in the lamp group is determined according to all the identification codes; when position information of one lamp is determined, N prestored models are tested via an identification code, and a model corresponding to the identification code is determined; according to position information which corresponding to the models in a one-to-one manner, the position information of the lamp corresponding to the identification code is determined, N is a natural number greater than one, and a position of the object to be positioned can be determined via three dimensional attitude estimation based on position information of all the lamps. Via the positioning method and the handle provided in the invention, a problem that codes are difficult to identify due to luminance codes which are not obvious can be solved, and effects of one side stereoscopic visual sense can be exerted.

The invention discloses an image recognition method for electric meter terminal fault recognition. The method comprises the steps of establishing a classification network based on deep learning, an electric meter terminal detection network and a configuration matching and fault recognition network; the input of the deep learning classification network is an image, and the output is the model of ammeter terminal equipment; wherein the output of the electric meter terminal detection network is a feature corresponding to each electric meter terminal model; acquiring information characteristics ofeach ammeter terminal panel by using a configuration matching detection method; a deep learning network is used for identification, typical'no-fault 'pictures and'faulty' pictures are sent into the deep learning network to be trained, during identification, pictures to be identified are sent into the network, and the network can output a result of whether faults exist. According to the identification method, priori knowledge of the electric meter terminal is utilized to the maximum extent, and the identification success rate of the algorithm is greatly improved. And necessary fault information is accurately provided, so that field operation and maintenance engineers are helped to quickly locate faults, and the efficiency is improved.

The invention relates to a social relation data generation method of a generative adversarial network. The method comprise: A, storing relation data of a social group, including user node informationand user relation information; B, constructing a generator which comprises a first graph neural network for receiving random noise and is used for generating pseudo-relation data; C, constructing a discriminator which comprises a second graph neural network for receiving the relational data and the pseudo-relational data and is used for judging the difference degree of feature structures between the relational data and the pseudo-relational data; and D, alternately training the generator and the discriminator until the discriminator judges that the feature structures of the pseudo relationshipdata and the relationship data are the same, and using the current generator to generate a social relationship data sample. According to the method, the social relation data with the same characteristics as the real relation data can be generated, the social relation data can be simulated more conveniently, the characteristics of the social relation data can be researched more conveniently, and richer samples are provided for related machine learning models of the social network relation data.

The invention discloses a multi-point geostatistics modeling parameter optimization method based on the variation function. The method comprises the steps of combining the correlation recognition of multi-point geostatistics model features and modeling parameters, taking the size of a data sample plate of multi-point geostatistics as an example, wherein along with the increase of the size of the sample plate, the morphological visual features of a model and a training image are more and more similar, and the modeling parameters are more and more similar; based on a variation function and an Hsim similarity function, evaluating the spatial correlation and structural feature similarity between the multipoint geostatistics random model based on the (ordered) modeling parameter set and a training image, and further establishing a relationship curve between spatial correlation evaluation indexes based on the variation function and modeling parameters; selecting a corresponding modeling parameter value when the evaluation index begins to converge and enters the platform area as an optimal parameter as a preferred parameter. Compared with a traditional artificial visual discrimination method, the method can efficiently and objectively optimize the multi-point geostatistics modeling parameters.

The invention provides a method for determining nickel-based single crystal high-temperature alloy residual segregation. By means of a manufactured metallographic specimen, primary dendrite distance of nickel-based single crystal high-temperature alloy is measured, and mass fraction distribution of all elements on all points of the nickel-based single crystal high-temperature alloy are measured on the manufactured metallographic specimen through a dot matrix method. The segregation directions of the elements are determined according to obtained mass fractions of the elements on the measurement points. A computer simulates solution treatment to obtain distribution of the elements after solution treatment, and accordingly the residual segregation ratios of Al, Ta, W and Re elements are obtained respectively. According to the method, experiments and simulation are combined so that residual segregation of the high-temperature alloy can be accurately and quickly predicted after solution treatment, cost of solution treatment research is reduced, and the period of solution treatment research is shortened. The mode that the mass fraction distribution of the elements on the points is obtained through the dot matrix method has the advantage of being large in specimen number, and result deviation caused by large errors in common representing methods is prevented.

The invention discloses a visibility analysis method based on a camera image, and belongs to the technical field of visibility. The core of the video visibility detection algorithm is to extract imagecharacteristic parameters capable of reflecting the influence of visibility change on images, so that the value of visibility is calculated by analyzing the image characteristic change. The road marking is used as a preset standard, the image contrast is used as an image characteristic parameter, the exponential distribution of the image characteristic along with the distance is analyzed, and thevisibility value is calculated through an image contrast threshold value at the critical visibility distance. According to the technical scheme provided by the embodiment of the invention, real-timeanalysis of images shot by cameras along the road is realized, quantitative measurement of visibility is completed, a reasonable judgment reference is determined, and errors which cannot be avoided byhuman eye measurement are reduced.

The invention discloses a method for generating a signal pulse sequence by a signal data simulator. The method comprises the following steps: S1, acquiring signal parameters of a man-machine interface; s2, establishing a signal pulse parameter description model according to the signal parameters; s3, according to the signal pulse parameter description model, calculating the arrival time of each channel pulse sequence, the repetition period between two adjacent pulse sequences and the duty ratio; s4, searching for sample data meeting a threshold value in each channel pulse sequence according tothe arrival time, the repetition period and the duty ratio; and S5, inserting the sample data between the two adjacent pulse sequences to form a cyclic copy signal pulse sequence. The simulator can simulate complex signals, and is economical, flexible, easy to control and repeat.

The invention discloses a method for establishing a post-coronary artery bypass transplantation acute kidney injury prediction model, which comprises the following steps of: a risk assessment model is firstly established by collecting and sorting hospital data of a large number of domestic patients and combining direct single-factor and indirect single-factor regression screening and multi-factor regression optimization analysis; the accuracy and the reliability of the evaluation model are effectively improved by increasing the sample size and incorporating new risk factors, especially introducing two risk factors of alanine aminotransferase increase and brain natriuretic peptide; therefore, an effective assessment and prediction method is provided for acute kidney injury risk assessment of coronary artery bypass transplantation of heart failure patients in China, and the method is of great significance to healthy development of the medical level in China.

The invention discloses a coronary artery bypass transplantation perioperative period risk prediction model establishment method, which comprises the following steps of: collecting and sorting a large number of hospital data of patients in China, combining a single-factor and multi-factor analysis methods, and analyzing by adopting a logistic regression secondary prediction method to obtain a secondary prediction model; the method not only enriches the types of risk factors investigated in the modeling process, but also effectively improves the accuracy and reliability of model prediction, provides an effective evaluation and prediction method for risk prediction of coronary artery bypass transplantation of heart failure patients in China, can effectively reduce the surgical risk. And the method has positive significance on healthy development of the medical health cause in China.

The invention relates to the technical field of computer information, and provides a cross-modal fusion target tracking method. The objective of the invention is to solve the problem that the difference between cross-modal targets is too large and a good effect is difficult to obtain by purely using feature-based matching. According to the main scheme, the method comprises the steps of constructing and generating an adversarial neural network composed of a pixel alignment module, a feature alignment module and a joint discrimination module, training on a data set to generate the adversarial network, extracting to-be-recognized targets from videos collected by different cameras, inputting the to-be-recognized targets into the trained joint discrimination module, and obtaining the feature similarity between the target and all the to-be-recognized targets; training a logistic regression model for predicting the time similarity between the targets according to the transfer time by using the transfer time data set of the marked targets between the cameras, and calculating the time similarity between the two targets by using the model; and adding the feature similarity and the time similarity to obtain total similarity, and taking the target pair with the highest total similarity as the same target.

The invention relates to a primer group, a kit and a method for simultaneously detecting multiple mutations of nine genes related to congenital adrenal hyperplasia. The kit comprises the following reagents: (1) a reagent for multiplex PCR amplification; and (2) a reagent for constructing a three-generation sequencing library. The method comprises the following steps: (1) preparing a subject sample; (2) performing multiplex PCR amplification on nine genes (CYP21A2, CYP21A1P, CYP11B1, CYP11B2, CYP17A1, HSD3B2, StAR, POR and CYP11A1) in the sample; (3) constructing a three-generation sequencing library; and (4) sequencing and analyzing the gene mutation type.

The invention discloses a method for predicting coal and gas outburst danger of a heading end by using post-blasting gas concentration, which comprises the following steps of: acquiring gas concentration monitoring data, identifying wave crest data not lower than a preset threshold value, and taking a sequence consisting of gas concentrations in a preset duration before and after a wave crest including the wave crest as a gas concentration abnormal sequence; adopting a weighted K-nearest neighbor algorithm to classify the gas concentration anomaly sequence; aiming at the gas concentration abnormal sequence classified as blasting operation, automatically extracting blasting time and corresponding gas concentration data by adopting first-order differential convolution operation, and acquiring post-blasting gas concentration data; and on the basis of the post-blasting gas concentration data, predicting the outburst danger degree of the heading head by adopting a convolutional neural network. According to the method, effective identification of abnormal gas concentration of the coal mine driving working face, accurate extraction of gas emission characteristics after blasting and continuous dynamic prediction of outburst danger can be realized.

The invention provides a method, a device, and an application for sequentially creating a multi-dimensional feature map of a personal scene. The method comprises the following steps: presetting a multi-level scene tag library, and presetting a corresponding scene logic relationship, an association probability calculation rule between scenes, and a sequential relationship and chain serial expression relationship between the scenes; automatically acquiring and / or calculating a multi-level scene label value of the user by using a base station or / and a satellite positioning system used by a mobilephone of the user, correspondingly storing the value in the preset multi-level scene label library, and generating a user personal scene according to the preset scene logic relationship; andThe invention provides a method, a device, and an application for sequentially creating a multi-dimensional feature map of a personal scene. The method comprises the following steps: presetting a multi-level scene tag library, and presetting a corresponding scene logic relationship, an association probability calculation rule between scenes, and a sequential relationship and chain serial expression relationship between the scenes; automatically acquiring and / or calculating a multi-level scene label value of the user by using a base station or / and a satellite positioning system used by a mobile phone ofthe user, correspondingly storing the value in the preset multi-level scene label library, and generating a user personal scene according to the preset scene logic relationship; based on multiple independent user personal scenes, calculating a correlation probability among multiple independent user personal scenes in a preset time period according to the association probability calculation rule between scenes, and generating the user personal scene map according to the sequential relationship and chain serial expression relationship between the preset scenes. According to the method, the device, and the application for sequentially creating the multi-dimensional feature map of the personal scene, the automatic creation of a personal scene map is realized, the personal scene map is appliedto realize user behavior prediction, and personalized recommendation is realized according to a prediction result.based on multiple independent user personal scenes, calculating a correlation probability among multiple independent user personal scenes in a preset time period according to the association probability calculation rule between scenes, and generating the user personal scene map according to the sequential relationship and chain serial expression relationship between the preset scenes. According to the method, the device, and the application for sequentially creating the multi-dimensional feature map of the personal scene, the automatic creation of a personal scene map is realized, the personal scene map is applied to realize user behavior prediction, and personalized recommendation is realized according to a prediction result.

The invention discloses a sampling apparatus using remote automatic control for transported lime powder of a tank car, and specifically relates to the technical field of lime power sampling. The sampling apparatus includes a vertical plate; a frame is disposed on one side of the vertical plate; a transmission mechanism is disposed inside the frame; a sampling mechanism is disposed on a bottom endof the frame; the transmission mechanism includes a motor, a slide plate, a screw rod and a guide rod; the screw rod is fixedly connected to an output end of the motor; the slide plate is slidably connected to the inside of the frame; both the screw rod and the guide rod penetrate through the slide plate; the screw rod is in threaded connection with the slide plate; and the sampling mechanism includes a material box. According to the sampling apparatus provided by the invention, the motor rotates reversely to drive the material box to move leftward; a bump is squeezed by the guide rod in a movement process, so that the material box moves upward while moving leftward, and the material box samples while moving; and therefore, the sampling range is larger, the sampled specimen is more complete, and the data obtained in later analysis is more accurate.

The invention relates to an indoor personnel detection method under monitoring conditions, comprising the following steps: establishing a sample set; Input sample pretreatment: input samples are labeled one by one, the upper part of the human body is labeled, and the labeled samples obtained above are pretreated by using the idea of region suggestion as the data input of the whole network model; Target feature extraction; Deformable treatment: According to the characteristics that most of the indoor personnel are visible in the upper body under the condition of monitoring, The corresponding component model is designed based on the algorithm idea of deformable component, and the feature map is obtained after the feature extraction of the human body object in the image. After the deformableprocessing layer, the corresponding component feature map is obtained by convolution operation with the corresponding component filter, and the mapping score of the component is calculated. Complete classification and discrimination; Network model training.

The invention relates to a frequency detection circuit and method for rapidly calculating electrical parameters of a motor, and the frequency detection circuit comprises a resistor R1, a resistor R2, a resistor R3, a capacitor C1, a capacitor C2, and a comparator U1. A pin 7 of the comparator U1 is connected with a 5V power supply, and a pin 4 of the comparator U1 is grounded; a pin 3 of the comparator U1 is connected with a resistor R3, and the other end of the resistor R3 is connected with an output signal of the motor; a second pin of the comparator U1 is connected with one end of a resistor R1, one end of a resistor R2 and one end of a capacitor C2, the other end of the capacitor C2 and the other end of the resistor R1 are grounded, and the other end of the resistor R2 is connected with a 5V power supply. A pin 8 of the comparator U1 is grounded, the pin 8 is also connected with one end of a capacitor C1, and the other end of the capacitor C1 is connected with a 5V power supply; a pin 6 of the comparator U1 is used as output; by setting the data groups, the effective data corresponding to the output signal can be output after the sampling of at least one data group without waiting for the completion of periodic sampling and then calculating the effective data corresponding to the output signal, so that the waiting time is shortened.

The invention discloses a community physical activity data acquisition device based on resident health. The device is applied to the technical field of data acquisition and comprises a sensor module, a data processing module, a data sending module, a data receiving module, a central processing module and a server. Multiple physiological parameter signals are acquired from samples, so that multi-sample, multi-parameter and real-time monitoring is realized, and the physiological parameter signals are transmitted after being processed and deleted. Through multi-terminal monitoring and health guidance, monitoring and guidance of resident health are facilitated.

The invention discloses a face recognition image processing method, which comprises the following steps: obtaining an identity card face photo of a person to be recognized with time parameters and a real-time face image with time parameters and environmental parameters at the same time; extracting human face features in the identity card human face photo with time parameters and human face features in the real-time human face image with time parameters and environmental parameters; calculating an identity card face feature autocorrelation coefficient, and at least adding a time parameter function to the identity card face feature autocorrelation coefficient to serve as a specific use condition; calculating an on-site face feature autocorrelation coefficient according to face features in the real-time face image, and at least adding a time parameter function and an environment parameter function to the on-site face feature autocorrelation coefficient to serve as specific use conditionsof the on-site face feature autocorrelation coefficient; and identifying the identity of the person to be identified according to the identity card face feature autocorrelation coefficient and the on-site face feature autocorrelation coefficient. The invention further discloses a system corresponding to the method.

The invention discloses a breeding method for a male sterile line of an edible oil sunflower. The method comprises the steps of finding out a male sterile line from a sunflower group to obtain a first generation of female parent plant bags; dividing the sunflower group with objective traits into multiple first generation of male parent groups, bagging and numbering, and pollinating pollen of the first generation of male parent groups into one male sterile plant in a hybrid manner; planting grains of the first generation of female parents to obtain a second generation of female parent groups, planting the grains of the first generation of male parents to obtain a second generation of male parent groups, bagging and numbering; selecting a group with the highest male sterile rate from the second generation of female parents, and dividing the second generation of plants of the previous generation of male parents in the group into multiple male parent groups, collecting the pollen from each group and then pollinating a single female parent plant, and so forth until the sterile rate is 80%, and carrying out one-to-one pollination on the male parents to achieve the breeding target. The breeding method has the beneficial effects that a maintainer line is large in search range, multiple in samples, high in success rate, short in period and suitable for interspecies maintaining, an excellent crossing parent of the sunflower can be provided through 5-6 generations of breeding, the yield and the quality of the sunflower are significantly improved and the market prospect is very considerable.

The invention discloses a reflective TOF device for cluster beam comprehensive deposition, which comprises a cluster ion acceleration device, wherein the cluster ion acceleration device comprises a first acceleration electrode and a second acceleration electrode, the first acceleration electrode applies pulse high voltage to cluster ions and generates an initial signal A to enable the cluster ions to deflect towards a first reflection device and to be initially accelerated, and the second acceleration electrode enables the cluster ions to be accelerated by 2-5keV; and a first flight channel communicated with the output end of the cluster ion acceleration device, wherein the accelerated cluster ions freely fly in the first flight channel. The cluster ion acceleration device is arranged, the first acceleration electrode applies pulse high voltage to cluster ions and serves as an initial signal to enable the cluster ions to deflect towards the first reflection device and be initially accelerated, the second acceleration electrode enables the cluster ions to be accelerated by 2-5 keV, and it is guaranteed that the speed difference of the cluster ions is small.