106results about How to "Guaranteed recognition rate" patented technology

The invention provides a real-time face recognition method based on a deep neural network and adjacent element analysis. According to the invention, at first, a multi-layer neural network is practiced through a large-scale face database having the advantage of good diversity, wherein layers besides the last layer are non-linear layers, and the tail layer is a linear layer; then, the obtained network is practiced continuously on the basis of a hybrid face database through the supervised adjacent element analysis method to make the network has more deep understanding of face images so as to achieve the purposes of shorting the face image distance between same individuals and increasing the face image distance between different individuals; and finally, in the actual face recognition stage, the invention provides a concept of 'search radius', so the recognition time can be shortened on the premise of ensuring the recognition rate to realize real-time face recognition. According to the invention, the advantage of high recognition rate can be realized, and at the same time, according to the invention, the advantage of fast recognition speed can be realized, so that the methof of the invention is suitable for being applied in real-time face recognition tasks.

The invention discloses a Chinese speech control system and method with mutually interrelated spectrograms for a driver. The control system is composed of a special Chinese speech identification control device (1), a silicon microphone (2), an infrared CCD (Charge-Coupled Device) camera (3) and a vehicle-mounted electronic device (4). The special Chinese speech identification control device (1) is electrically connected with the silicon microphone (2), the infrared CCD (Charge-Coupled Device) camera (3) and the vehicle-mounted electronic device (4). The special Chinese speech identification control device (1) is composed of a DSP (Digital Signal Processor) chip (11), a video AD (Analog-to-Digital) module chip (12), a synchronous FIFO (First Input First Output) chip (13), a speech encoding / decoding chip (14), a CPLD (Complex Programmable Logic Device) chip (15), a 1# relay (16) and a 2# relay (17). The vehicle-mounted electronic device (4) comprises an air conditioner (4a) and a trunk lock (4b). The invention also provides a method for controlling the vehicle-mounted electronic device (4) by using the system.

The invention discloses a method for sorting postal letters according to addresses driven by an address information base, wherein each delivery address in the address information base at least comprises a character expression containing address information for realizing transit sorting, local port sorting and road sections; the address character expression obtained by image acquisition and character identification is in transversal matching with the delivery addresses in the address information base; and sorting information of the postal letters can be obtained according to the matching degree, so that the transit sorting, the local port sorting and the road section delivery after the local port sorting can be realized. The invention is actually applied to an identification module for an information sorter; and under the conditions that the address base is relatively accurate and complete and the identification rate is basically ensured, the method can be effectively used for analyzing and correcting a letter address identification result to obtain accurate letter sorting information, i.e. the method can be used for sorting the letters in complete accordance with the letter address identification result.

The present invention discloses a single depth map-based human body joint point recognition method for virtual reality. The method comprises: training a gesture public data set NYU in an offline manner by using a convolutional network to obtain a model with good robustness, high accuracy and a fast recognition speed; and capturing a depth image in real time by using a depth camera, after performing a series of preprocessing on the image, transmitting the preprocessed image to a human bone recognition module and a gesture recognition module respectively, returning recognized joint point three-dimensional information, and mapping a three-dimensional human body model. The method disclosed by the present invention has good interactive functions in the virtual reality environment, and has goodperformance in the recognition rate, the operation speed, universality, accuracy, and the like.

The embodiment of the invention provides a pavement crack detection method, device and equipment and a storage medium. The method comprises the steps that acquiring multiple preset pavement sample images; preprocessing the plurality of preset road surface sample images based on gamma correction; training the convolutional neural network model; predicting the trained convolutional neural network model; if it is determined that training is stopped according to the prediction result, obtaining a test convolutional neural network model; testing the tested convolutional neural network model; if thetest result is within a preset target result range, obtaining a target convolutional neural network model; and inputting a three-channel image corresponding to a to-be-detected pavement image into the target convolutional neural network model to obtain pavement crack information in the to-be-detected pavement image. According to the method provided by the embodiment of the invention, the problemsthat the crack recognition rate cannot be ensured and time and resources are wasted in the prior art can be solved.

The invention discloses a method for identifying synonyms. The method is executed in a computing device, multiple query records of multiple users are stored in the computing device, and each query record comprises user identification, query time and a query phrase. The method comprises the steps that multiple conversations are determined according to multiple stored query records within preset time, each conversation comprises one or more query phrases of the same user identification; for each conversation, each query phase in the conversation is divided into one or more words; candidate synonym pairs are extracted from each conversation according to preset rules; for each candidate synonym pair, word vectors corresponding to two words in the candidate synonym pair are determined according to a word vector model, the similarity of the two word vectors is calculated, and if the similarity is larger than a first threshold value, the two words in the candidate synonym pair serve as the synonyms. The invention further discloses a device capable of implementing the method and a computing device comprising the device.

The invention discloses a three-decision unbalanced data oversampling method based on a Spark big data platform, and relates to a Spark big data technology in the field of data excavation. The method comprises the following steps: firstly, carrying out data transformation with an RDD (Resilient Distributed Dataset) of Spark to obtain a normalized sample set with the LabeledPoint format <label: [features]>, and dividing the sample set into a training set and a test set; secondly, carrying out data variation by adopting the RDD of Spark, calculating a distance between samples, determining the radius of a domain, and classifying the samples in the whole training set into positive domain samples, boundary domain samples and negative domain samples according to a neighborhood three-decision model; then respectively oversampling the boundary domain samples and the negative domain samples; and finally, calling a Spark Mllib machine learning algorithm to verify a sampling result. According to the three-decision unbalanced data oversampling method based on the Spark big data platform, the problem of classification of a large-scale unbalanced data set in the field of machine learning and mode recognition is effectively solved.

The invention relates to a method for improving the parameter number of a deep-layer convolution neural network, which comprises the following steps: step 1, constructing a deep-layer convolution neural network; 2, improving deep-seated convolution neural network; 3, training and testing the deep-seated convolution neural network; 4, optimizing network parameters. Compared with the prior art, theinvention effectively reduces the complexity of the model through the improvement of the structure of the convolution neural network and the reduction and optimization of the network parameters in themodel, while maintaining the original recognition rate of the model, greatly reducing the training time and reducing the hardware requirement.

The invention discloses an identification method for truck brake shoe breakdown. The identification method for the truck brake shoe breakdown comprises extracting segmentation characteristics of three angles from a current image; conforming a characteristic region of a truck brake shoe according to the segmentation characteristics of the three angles; extracting truck brake shoe characteristics from the characteristic region of the truck brake shoe in the current image; utilizing the identification method of a support vector machine (SVM) for the truck brake shoe breakdown and calculating characteristics values of the truck brake shoe; judging whether the truck brake shoe failures or not according to the characteristic values and the presetting fault indentification values. The invention discloses an identification device for the truck brake shoe breakdown. Manual-acting faults can be avoided by utilizing the identification device for the truck brake shoe breakdown. The identification rate of the breakdown can be guaranteed. The accidents can be prevented timely. Therefore, operation safety can be guaranteed.

The invention discloses a method for speech recognition and an electronic device. The speech recognition method is applied to the electronic device which includes a display unit. The speech recognition method comprises the following steps: a first prompt message of at least one prompt message in a prompt database is displayed in the display element; based on the first prompt message, a first voice message is obtained; matching is carried out between the first voice message and the at least one prompt message; when the matching between the first voice message and the at least one prompt message does not succeed, based on a first instruction set which carries out operation on the electronic device, identification is carried out on the first voice message, and a first identification result is obtained; and when the first identification result displays that the identification result is not correct, a second instruction set is employed to carry out identification on the first voice message to obtain a second identification result, wherein the second instruction set is an instruction set different from the first instruction set.

A voice processing method and a terminal. The method comprises: retrieving voice information, of which the usage frequency reaches a designated value in a server end, and locally saving the voice information (S11); after input information is received, recognizing voice feature information from the information and converting same into a voice instruction (S12); and according to the voice instruction, performing matching in the locally saved voice information, and if no corresponding voice information is matched, sending the voice instruction for matching to the server end (S13). The method and terminal can increase voice recognition flexibility, controllability, expansibility and resource reusability on the premise of guaranteeing the recognition rate.

The invention relates to a facial expression recognition algorithm based on discriminative component analysis, and particularly relates to a facial expression recognizing method based on an improved DCA (discriminative component analysis). The method comprises the steps of firstly obtaining a sample subset in accordance with principles of a maximum type inner distance and a minimum type inter-distance; secondly calculating a projection matrix of the sample subset; and finally multiplying samples with the projection matrix, and judging the types of the samples through a nearest neighbor method so as to accomplish a facial expression recognition task. With the adoption of the method, not only is the DCA algorithm migrated to the field of facial expression recognition, but also an improved algorithm based on the DCA is provided according to practical conditions; the effectiveness of the algorithm is proved through the experiment result; and 95.71% of average recognition rate is ultimately achieved in a JAFFE (Japan female facial expression) image library.

The invention discloses a system and method for transmitting data using barcode. The system comprises a first terminal and a second terminal; each of the first terminal and the second terminal is provided with a HD digital interface, and the first terminal and the second terminal are connected through a HD video wire, wherein the first terminal further comprises a barcode conversion module for acquiring to-be-transmitted data from the first terminal, converting the data into the barcode capable of bearing data and data identification, and transmitting the barcode to the second terminal; the second terminal further comprises a video acquiring card and a barcode analysis module, the video acquiring card is used for acquiring the barcode and sending the barcode to the barcode analysis module; the barcode analysis module is used for analyzing the barcode, and transferring the analyzed barcode to the second terminal. Through the adoption of the system disclosed by the invention, the data exchange can be performed between the computers in network isolation.

The invention provides a weight loss feather extraction method based on an acceleration transducer, comprising the following steps: (1) receiving the measuring data of an acceleration transducer; (2) according to the measuring data, vertically correcting accelerated speed to obtain net accelerated speed after removing gravity in the vertical direction; and (3) according to the net accelerated speed obtained in step (2), detecting the state of weight loss, and extracting weight loss feather data.

The invention discloses a method for recognizing a cloud based on a mie-scattering laser radar with equalized value distribution. The method comprises the steps that a mie-scattering laser radar signal is input; semi-discretisation data processing based on noise estimation is carried out on the mie-scattering laser radar signal; data distribution equalization is carried out on the mie-scattering laser radar signal; the background base line of the mie-scattering laser radar signal is generated; the mie-scattering laser radar signal is layered and recognized based on the background base line. Through the method, the importance of a smoothness problem in the cloud recognition process can be weakened, meanwhile, the background base line adapting to signal changes by itself is found out, and therefore the problem that a current algorithm overly depends on smoothness or missing detection happens to weak cloud echo signals is solved, the recognition rate of cloud signals in the observation process is guaranteed, and the data quality of cloud observation is improved, and more accurate observation data are provided for weather and climate study and forecasting.

The invention relates to the technical field of image mode recognition, and provides a road region detection method and a system, wherein, the method comprises the following steps: S1: an image of a detected road is obtained; S2: the obtained image data is processed for determining the smooth region; S3: the smooth region with the biggest pixel area in the smooth region is determined; S4: the road candidate region is determined according to the position relationship between other smooth regions and the smooth region with the biggest pixel area; and S5: the road candidate region is modified for obtaining the final road region. The road region detection method and the system solve the problems that the prior road region method which is based on color segmentation has the problems on the treatment of shadow, water trace and uneven road color, the road region detection method and the system can not only be applied to the structural road for solving the problem of difficult boundary extraction, but can also be applied to the non-structural road. The real-time property and the robustness of the detection are enhanced on the basis of ensuring the recognition rate, and the method is simple, rapid and effective.

The invention discloses an authentication optimization method of a voiceprint identity authentication device. The authentication optimization method comprises the steps that the Mel-frequency cepstral coefficients corresponding to registration voice signals are extracted and preset number binding is performed on the Mel-frequency cepstral coefficients; the Mel-frequency cepstral coefficients act as an input layer and the bound numbers act as an output layer to perform differentiated deep belief network training and acquire the parameter space; the Mel-frequency cepstral coefficients are inputted to the differentiated deep belief network to acquire the hidden layer output to act as the feature vectors; all the feature vectors act as the input to construct a Gaussian mixture model; and the corresponding Mel-frequency cepstral coefficient of any registration voice signal is inputted to the differentiated deep belief network to acquire multiple hidden layer outputs, and the hidden layer outputs of which the degree of distinction is higher than the preset threshold are selected to act as the training data to update the Gaussian mixture model. The following spontaneously changed voice signal of the registrant acts as the raining data to update the Gaussian mixture model so as to be more adaptive to the present sound production state of the registrant, and the recognition rate can be guaranteed.

The invention discloses a KAZE feature-based hand back vein recognition method, which comprises the following steps of: 1, preprocessing an acquired hand back vein image, and carrying out feature point extraction on the preprocessed image based on a KAZE algorithm; step 2, performing coarse matching on the feature points extracted from the two images by using a nearest neighbor ratio method to obtain coarse matching feature points; and step 3, performing fine matching on the coarse matching feature points again by using an RANSAC algorithm to obtain matching points. The invention provides a KAZE feature-based hand back vein recognition method, which can significantly reduce the false recognition rate and the recognition time, and has good recognition accuracy and real-time performance.

The invention discloses a method for obtaining an APP service feature library and a corresponding device. The method comprises the steps of obtaining an APP installation package, current network service data and simulation service data; analyzing the APP installation package, the present network service data and the simulation service data respectively to obtain APP names to which the APP installation package, the present network service data and the simulation service data belong respectively, and generating a learning data set; performing feature extraction on the current network service data and the analog service data to obtain a service feature tree containing at least one service feature; and carrying out feature matching on the business feature tree and the learning data set, determining an APP name to which each business feature in the business feature tree belongs, and generating a business feature library. The existing network service data has certain complexity, a data blind area can be eliminated, and the recognition rate is ensured; the service feature library is constructed from multiple feature dimensions, so that the service data with lower identification degree can be effectively identified, the feature identification accuracy is improved, and manpower can be effectively solved.

The invention discloses an action living body recognition method based on attitude estimation and action detection, and relates to the technical field of biological recognition. The method comprises the steps: recognizing a real person and an attack by enabling a user to cooperate to do an action, wherein an action instruction set is random, so whether the user is a living body or not can be judged more accurately; before a living body is made, in order to improve the one-time passing rate of a real person, carrying out the light recognition, face recognition in a circle, face alignment screenrecognition and face occlusion recognition on a user; estimating a depression angle, a deflection angle and a rotation angle of the user by adopting a posture estimation PSECNN model, and performinghead shaking, nodding and head raising recognition on the user, so as to accurately recognize the head action of the user; when blinking and mouth opening actions are identified, adopting an MCNN model and an ECNN model to replace mainstream detection based on feature points, so the blinking and mouth opening identification rate is ensured in precision; in vivo detection, action instructions are random, actions and action sequences need to be accurate, anti-copying recognition is carried out, and safety is higher.

The invention discloses a method of recognizing ear speech in a normal speech flow under the condition of a small database, comprising the following steps: building a speech recognition and classification system which comprises a digital speech input module, a feature extraction module, and a deep neural network composed of a deep belief network and a Softmax output layer, wherein the deep belief network is composed of restricted Boltzmann machines stacked from bottom to top, spectrum feature is the input feature of the deep belief network, and the Softmax output layer correspondingly outputs a target value of the deep neural network; acquiring training data set samples and processing the samples in at least one of the following ways: (1) artificially extending the data set 8 to 24 times by repetition; and (2) scrambling the data set, and training the speech recognition and classification system; and using the trained classification system to recognize a to-be-recognized speech flow. The recognition performance of the system can be improved in a small database, and the recognition rate of normal speech is guaranteed while ear speech recognition is realized.

The invention belongs to the technical field of invoice information extraction, and particularly relates to an invoice information extraction method combining two-dimensional code recognition, which comprises the following steps: S1, carrying out bill text recognition; s2, carrying out bill two-dimensional code identification; and S3, selecting an identification result: after output information ofbill text identification and bill two-dimensional code identification is obtained, analyzing and accepting or rejecting the two groups of information, deriving an approximate position of a two-dimensional code according to a text identification result, and performing positioning identification on the two-dimensional code by using a deep learning technology. The two-dimensional code identificationrate is far higher than that of a traditional method. According to the method, text recognition and two-dimensional code recognition results are accepted or rejected according to the confidence coefficient, the recognition accuracy is further guaranteed, meanwhile, the text and the two-dimensional code are recognized, and when information of one party is defective, the result of the other party can supplement the final result advantageously. The method can adapt to more complex and changeable scenes, and has higher robustness.

The invention provides a two-dimensional code beautifying and anti-fake method, which relates to the technical field of two-dimensional code, and comprises the steps of reducing the area of a black block in an original two-dimensional code; reducing the pixel of the two-dimensional code; adjusting the transparency of the black block and white block in the two-dimensional code; obtaining the area and pixel of the black block of the two-dimensional code, and the minimum distinguishable value of the transparency of the black block and white block; merging the two-dimensional code with a high-definition background image, and adjusting the RGB value to the optimum value that can be most easily identified. The invention is advantageous in that with the adoption of the high-definition backgroundimage and the foreground two-dimensional code, the covering proportion and image transparency value can be reduced, and the beauty degree can be guaranteed to the maximum extent; the best parameter combination is selected, and the recognition rate of the two-dimensional code is guaranteed; the image processing algorithm is difficult to copy, and thereby the two-dimensional code is hard to counterfeit; the covering rate and pixel of the foreground two-dimensional code can be reduced, and thereby the two-dimensional code image is hardly distinguishable after shooting, and no cheating can be conducted under specific scene.

The invention discloses a Spark big data platform-based neighborhood density imbalance data mixed sampling method, and relates to a computer information acquisition and processing technology. According to the invention, the data is stored in the RDD through the Spark; carrying out normalization processing; The method comprises the following steps: dividing RDD into a positive domain space, a negative domain space and a boundary domain space according to neighborhood density in combination with a three-branch decision theory, sampling data of a boundary domain by adopting an SMOTE algorithm, sampling data of a negative domain by adopting a mixed sampling algorithm, and finally merging the data in the three domains to obtain a final data set. By dividing each piece of data into different domains and processing the data according to the characteristics of the different domains, a small number of types of data can be properly added, and meanwhile, most types of data are properly reduced. And finally, calling an MLLib algorithm library, and evaluating the effect by using a machine learning classifier. According to the method, the problem of inter-class proportion imbalance of unbalanceddata can be effectively alleviated, and the precision of the algorithm is improved.

Embodiments of the invention disclose a two-dimensional code generation method and device. The two-dimensional code generation method comprises the following steps of: obtaining an original two-dimensional code and a plurality of self-defined two-dimensional code elements, wherein the original two-dimensional code comprises a plurality of two-dimensional code elements, and the self-defined two-dimensional code elements have first shape features; recognizing a two-dimensional code element combination in the original two-dimensional code, wherein the two-dimensional code element combination comprises at least two two-dimensional code elements which have the adjacency relation and has a second shape feature; matching the first shape features of the plurality of self-defined two-dimensional code elements with the second shape feature of the two-dimensional code element combination; and if a target self-defined two-dimensional code element with the matched shape feature, replacing the two-dimensional code element combination in the original two-dimensional code by the target self-defined two-dimensional code element so as to obtain a target two-dimensional code. According to the methodand device, data safety of two-dimensional codes is enhanced.

The invention belongs to the field of wireless sensor networks, and relates to a node optimal deployment method suitable for a wireless positionable sensor network. The method aims at processing optimal deployment of wireless positionable sensor network nodes, and comprises the following steps: acquiring a sensor node antenna gain estimation model in a discrete state; constructing a double-dipoleantenna radiation gain model and a field intensity estimation model in a simultaneous state; obtaining an optimization objective function through communication among a mobile charger, a sensing node and a common node; and optimizing a target function by adopting a multitask evolutionary algorithm based on an information positive migration mechanism to obtain an optimal deployment mode of the sensing nodes. The method has the characteristics that the energy utilization rate of the charger is effectively improved, the charging inactivation time is shortened, and meanwhile, the requirements of high-precision positioning and large-range coverage of the whole system are met.

The embodiment of the invention discloses a weight precision configuration method and device, equipment and a storage medium. The method comprises the following steps: determining a current recognition rate threshold value from at least two candidate recognition rate threshold values smaller than the target recognition rate threshold value, reducing and adjusting the weight precision correspondingto each layer in the neural network based on the current recognition rate threshold; and training the reduced and adjusted neural network to adjust the weight parameter value of each layer, the training target being to improve the recognition rate of the reduced and adjusted neural network, and determining the final configuration result of the weight precision of each layer according to the relationship between the current recognition rate and the target recognition rate threshold. By adopting the technical scheme, the resource utilization rate in the artificial intelligence chip bearing theneural network can be improved, the chip performance can be improved and the chip power consumption can be reduced under the condition of ensuring the identification rate of the neural network.

The invention provides an advertisement image detection method and device. The advertisement image detection method comprises the steps of extracting images at reference positions in advertisement video stream and at least one image before and after the reference positions; taking pixels with high identification rates in the selected images as identification points for matching images; segmenting the extracted images in a mode of dividing the images at the reference positions and the at least one image before and after the reference positions of the extracted images into segments, thereby forming image segments; dividing the image segments into multiple parts, thereby forming image groups; carrying out parallel similarity matching on the identification points of the reference position images in each image segment of the image groups and the identification points of the at least one image before and after the reference positions, thereby obtaining an advertisement segmentation time vector of each image group; and obtaining start time and end time of each piece of advertisement in the advertisement video stream according to the advertisement segmentation time vector of each image group.

The embodiment of the invention provides a pavement repair information detection method, device and equipment and a storage medium. The method comprises the steps of acquiring multiple frames of pavement sample images; training a plurality of convolutional neural network models according to the plurality of road surface sample images to obtain a plurality of trained convolutional neural network models; determining a target convolutional neural network model in the plurality of trained convolutional neural network models; and inputting a to-be-detected pavement image into the target convolutional neural network model to obtain pavement repair information in the to-be-detected pavement image. The method provided by the embodiment of the invention can solve the problems that the identification rate of repair information cannot be ensured and time and resources are wasted in the prior art.

The invention relates to the field of workpiece identification, in particular to a shielded workpiece identification method based on template matching. The method comprises: preprocessing a color image; then manufacturing a plurality of standard templates; and finally, judging whether a workpiece exists or not through standard template matching. The system comprises an image alignment module, a preprocessing module, a standard template module and a template matching module. According to the recognition method, the workpiece in the non-shielding state and the workpiece in the shielding state can be effectively recognized, the recognition rate can reach 94.4%, and the workpiece in the non-shielding state and the workpiece in the shielding state can be effectively recognized.