127results about How to "Solve the low recognition rate" patented technology

The invention discloses a method for protecting mobile terminal privacy based on voiceprint recognition. The method comprises the following steps: analysis and extraction of voice message characteristics, model training and voiceprint recognition. In the method, the efficiency of the voiceprint recognition is high, and the performances and the recognition efficiency of the voiceprint recognition are not associated with the texts and the language. The voice length of a speaker is not required specially, and only voice messages in minutes are required so as to acquire the voice feature messages of the speaker, thereby carrying out the model training and the recognition. Applications of a voiceprint recognition algorithm in the intelligent mobile terminal have the advantages of high stability and recognition accuracy, and a voice feature model which is trained occupies small memory space of the intelligent mobile terminal. The method can be utilized to preferably protect the private information of a user in the intelligent mobile terminal and avoid the private information from being revealed and stolen when a mobile phone is lost.

The invention relates to an embedded type speech emotion recognition method and an embedded type speech emotion recognition device. The method comprises a feature extraction method, an emotion model training method, a Gaussian mixture model and an emotion recognition method. The method is as follows: parameters of a speech emotion recognition model can be adjusted in a self-adoption manner according to the recognition result of a speaker module, and an unspecific speaker speech emotion recognition question is transformed into a specific speaker speech emotion recognition problem. The device comprises a central processor, a power supply, a clock generator, a Nand Flash storage, a Nor Flash storage, an audio coding-decoding chip, a microphone, a loudspeaker, a keyboard, an LCD (Liquid Crystal Display) display and an USB (Universal Serial Bus) interface storage. According to the embedded type speech emotion recognition method and device, a speaker recognition model is added to the speech emotion recognition, and therefore, the problem that the speech emotion recognition is suddenly declined under an unspecific speaker condition is solved, and the identity recognition function is brought to the device.

The invention discloses a semi-supervised speech feature variable factor decomposition method. Speech features are divided into four types: emotion-related features, gender-related features, age-related features and noise, language and other factor-related features. Firstly, a speech is pretreated to obtain a spectrogram, speech spectrum blocks of different sizes are inputted to an unsupervised feature learning network SAE, convolution kernels of different sizes are obtained through pre-training, convolution kernels of different sizes are then respectively used for carrying out convolution on the whole spectrogram, a plurality of feature mapping pictures are obtained, maximal pooling is then carried out on the feature mapping pictures, and the features are finally stacked together to form a local invariant feature y. Y serves as input of semi-supervised convolution neural network, y is decomposed into four types of features through minimizing four different loss function items. The problem that the recognition accuracy rate is not high as emotion, gender, age and speech features are mixed is solved, and the method can be used for different recognition demands based on speech signals and can also be used for decomposing more factors.

The invention provides an identification method for a radar disoperative target based on a mixed model, which is used for solving the problem that the disoperative target is low in identification rate. The method comprises the following steps: establishing a standard body model library of a refined scattering point model; structurally decomposing the disoperative target according to the standard body model library to generate a first scattering point model; shielding the first scattering point model to obtain an effective scattering point model; calculating RCS intensity for the effective scattering point to obtain intensity information and combining the intensity information to generate a scattering point matrix; adding a statistic characteristic into the scattering point matrix to obtain a second scattering point model of the disoperative target containing coordinate information and the RCS intensity information; carrying out multi-scattering point radar return simulation on the second scattering point model to establish a high resolution one-dimensional range profile template library; and identifying the tested high-resolution one-dimensional range profile by adopting a K near neighbor classifier by virtue of the high resolution one-dimensional range profile template library. According to the method provided by the invention, the target identification performance of the radar system can be improved.

A speech recognition apparatus predicts, based on the occurrence cycle and duration time of impulse noise that occurs periodically, a segment in which impulse noise occurs, and executes speech recognition processing based on the feature components of the remaining frames excluding a feature component of a frame corresponding to the predicted segment, or the feature components extracted from frames created from sound data excluding a part corresponding to the predicted segment.

The invention discloses an automatic identification method for a milking sow gesture on the basis of a depth image. The method comprises the following steps that: collecting original depth image data,carrying out preprocessing, and carrying out manual annotation to form a milking sow gesture identification dataset; designing and training a milking sow hybrid deformable component model based on animproved HOG (Histogram of Oriented Gradient) feature; constructing a milking sow gesture identification deep convolutional neural network, utilizing an annotation frame and annotated gesture category training set information, and combining with a random Dropout method to train the network; inputting the test set into the milking sow hybrid deformable component model to obtain the target area ofthe milking sow; and inputting a target area result into the milking sow gesture identification deep convolutional neural network to identity the milking sow gesture. By use of the automatic identification method for the milking sow gesture on the basis of the depth image, the problem that an RGB (Red, Green and Blue) image is likely to be affected by the changes of factors, including outside illumination, shades and the like is overcome, the problem that the milking sow gesture is difficult in identification at night is solved, and the practical application requirement of all-weather milkingsow gesture monitoring can be met.

The invention provides a picture recognition method and system. The method comprises the following steps of: obtaining a to-be-recognized picture; inputting the to-be-recognized picture into a pre-trained picture recognition model so as to obtain a predicted category label of the to-be-recognized picture, wherein the pre-trained picture recognition model adopts a convolutional neural network model, the convolutional neural network model comprises an input layer, a convolution layer, an attention branch, an element-wise operation layer, a pooling layer, a full connection layer and an output layer, the attention branch is used for determining a weight of each area of the picture according to global features of each area of the to-be-recognized picture, and the element-wise operation layer isused for weighting local features, of the picture, output by the convolution layer according to the weight of each area of the picture; and recognizing a category of the to-be-recognized picture according to the predicted category label of the to-be-recognized picture. According to the method and system, the problem that the efficiency of recognizing specific types of pictures such as pictures with too small main body areas or embedded / spliced pictures is relatively low is solved.

Disclosed are a character identification method and a character identification apparatus. The character identification method comprises: obtaining a threshold array; selecting a first threshold from the threshold array as a selected threshold; performing binarization processing on a character image by using the selected threshold to obtain a binary image of the character image; performing character identification on the binary image to obtain an identification result; calculating a confidence of the identification result; determining whether the confidence of the identification result is greater than a preset confidence value; if the confidence of the identification result is greater than the preset confidence value, using the identification result as an identification result of the character image; and if the confidence of the identification result is not greater than the preset confidence value, selecting a second threshold from the threshold array, and replacing the first threshold by using the second threshold as a selected threshold. By means of the present invention, the problem is solved that the conventional character identification method is only applicable for identifying an original copy with a high image contrast and the identification rate of an original copy with a low image contrast is low.

The invention provides a visible light image insulator identification method and solves problems of poor universality and a non-high identification rate existing in a visible light image insulator under the complex background in the prior art. The method comprises steps that a training sample data set is constructed according to an original image; according to the training sample data set, the original image is detected through employing a BING algorithm, and an insulator candidate window is outputted; the insulator candidate window is identified through a convolutional neural network after training, and an insulator fine extraction window is outputted; the insulator fine extraction window is calculated through a high overlapping degree window iteration weight merging algorithm, and an insulator target window is outputted. The method is advantaged in that the problem of poor universality existing in the visible light image insulator under the complex background in the prior art is solved through the BING algorithm, and the problem of the non-high identification rate is solved through the convolutional neural network and the high overlapping window iteration weight merging algorithm.

The invention belongs to the field of electronic reconnaissance identification, in particular to a radar emitter signal modulation identification method combined with multi-dimensional feature migration fusion, comprising the following steps of generating nine kinds of radar signals to form a radar signal set; transforming the radar signal into time-frequency image by time-frequency transform; transforming the time-frequency image so as to meet the input requirements of the pre-trained large-scale network; sending the pre-processed time-frequency image to LeNet 5 network for feature extraction, and using the feature extraction module from input layer to form C5 convolution layer to output the feature extraction module; selecting a dimensionality reduction mode for the data obtained from the extracting feature step and processing the dimensionality reduction mode. The invention adopts the method of time-frequency analysis, maps the one-dimensional time-domain signal to the two-dimensional time-frequency domain, analyzes and processes the radar signal in the time-frequency domain, and has better effect for the non-stationary radar signal. The self-training network adopted by the invention has simple structure, and can improve the reliability of the system under the condition of low signal-to-noise ratio.

The invention discloses an intelligent wheelchair voice drive controller and an identification and control method. The controller includes an audio circuit module, a voice processing circuit module, a MAX232 communication module, a DSP processor, a joystick circuit module, a speed regulation circuit module, Power supply circuit module, motor drive circuit module, fault detection circuit module, voltage and current detection circuit module, brake circuit module. Speech recognition adopts a specific human speech recognition method based on fuzzy support vector machine, which improves the recognition rate and enhances the anti-noise ability. The DSP processor controls the coordinated operation of the motor by outputting PWM waves. The invention proposes a master-slave intelligent wheelchair voice drive control scheme with a single-chip microcomputer as a voice information processor and a DSP as a drive controller. The advantage of the present invention is that it can be controlled manually or automatically by voice, and it is convenient to use.

The embodiment of the invention discloses a data processing method and device and a storage medium, wherein the method comprises the steps of separately obtaining the sample object data from multiplepieces of multimedia data, and extracting the object feature information of the sample object data; clustering the plurality of sample object data according to the object feature information to obtaina plurality of first object feature clusters with different cluster labels; cleaning the heterogeneous object data in each first object feature cluster, and determining the cleaned first object feature cluster as a second object feature cluster; performing similar cluster merging among the plurality of second object feature clusters, generating a plurality of third object feature clusters with different cluster labels, and updating the object labels of the sample object data into cluster labels of the third object feature clusters to which the object labels belong; and based on the object label and the object feature information of the sample object data in each third object feature cluster, training an object recognition model. According to the invention, the accuracy of the face recognition can be improved.

The invention discloses a plant identification method and device with a high identification rate. The method includes the specific steps that firstly, a plant organ digital image is collected through an image collection unit and serves as a test sample; secondly, the sample is preprocessed to obtain a gray level image; thirdly, feature extraction is carried out on the gray level image through a pulse-coupled neural network so as to obtain an entropy sequence capable of reflecting the gray level image; a high plant identification rate can be obtained through a support vector machine classifier with the entropy sequence as a main feature and the morphological features as auxiliary features. The plant identification method and device can achieve simple and accurate identification of large sample data (with a large number of plant species to be distinguished) and have a high identification rate and strong adaptability; tests on existing databases verify that the accuracy rate of the plant identification method and device reaches more than 98%.

Embodiments of the invention provide a telecommunication fraud recognition method and data processing equipment, and aims at solving the technical problem that electronic equipment is relatively low in fraud event recognition rate in the prior art. The telecommunication fraud recognition method comprises the following steps of: obtaining user behavior information from terminal equipment connected with the data processing equipment, wherein the user behavior information is used for indicating an operation behavior carried out on to-be-assessed information in the terminal equipment by a first user; processing the user behavior information on the basis of at least one assessment index so as to obtain an assessment probability, wherein the at least one assessment index is determined on the basis of a history fraud event and is used for representing sensitiveness, for the history fraud event, of the first user, and the assessment probability is used for representing a matching degree between an operation indicated by the user behavior information and a history operation related to the history fraud event; and if the assessment probability is greater than a preset probability, determining the to-be-assessed information as telecommunication fraud information.

The invention discloses a pipeline safety event identification and knowledge mining method based on an HMM model, belonging to the pipeline safety event monitoring field. The method comprises the following steps: (1) extracting the multi-domain features of signals collected by each space point and obtaining the feature vector sequence of the signals; 2, inputting a feature vector sequence into anHMM model for off-line training to complete that construction of a typical event HMM model library; 3, after obtaining that current feature vector sequence of the signal to be identified through the step 1, inputting the signal to be identified into a typical event HMM model library for identification and outputting an event judgment type, and calculate an optimal hidden state sequence as the information of the evolution process of the event state sequence for output, and completing knowledge mining; The HMM model of the invention is analyzed and identified based on the characteristic timing sequence, and the event identification rate is effectively improved. At the same time, knowledge mining is realized in the evolution process of event state series, which can be used for short-term prediction.

The invention discloses a container contour positioning method based on angular point detection. Images at the two sides of a container below are acquired through a pick-up head, by use of a container lockhole coarse positioning and tracking method, coarse positioning areas of upper and lower lockholes of the images are obtained and images I1 and I2 in the areas are obtained; mask matrixes m1 and m2 in the same size as the images I1 and I2 are newly established, according to the upper and lower lockhole course positioning images I1 and I2 and the corresponding mask matrixes m1 and m2, segmentation images are obtained by use of a Grabcut algorithm, then according to the segmentation images, minimum external connection rectangles of foregrounds of the images are obtained by use of an algorithm of obtaining minimum external rectangles, and by taking one summit of each rectangle as an angular point of a container contour, based on a binocular stereo visual technology, pixel coordinates of inflection points are converted into world coordinates and are sequenced to form a quadrangle, i.e., the container contour. The method can effectively solve the disadvantages of light interference, non-obvious lockhole foreground and background discrimination and the like, prevents the problem of low recognition rate under the condition of insufficient light and realizes accurate positioning of the container contour.

The invention discloses a modulation signal identification method based on course learning, and mainly solves the problem of low identification rate caused by signal noise in the prior art. Accordingto the scheme, the method comprises the steps of acquiring a trained modulation signal sampling sequence and corresponding mark data, and preprocessing the sampling sequence; constructing a deep residual network; taking the preprocessed sampling sequence as the input of a deep residual network, taking the mark data of the sampling sequence as the modulation type corresponding to the maximum component in the output vector of the deep residual network, and training the constructed deep residual network by utilizing a training strategy of course learning to obtain a trained network; and taking the modulation signal grey-scale map to be identified as the input of the trained network, wherein the modulation type corresponding to the maximum component in the network output vector is the identified modulation type. According to the method, the training speed is increased, the influence of too strong signal noise on the recognition rate is reduced, the modulation recognition performance in a strong noise environment is improved, and the method can be used for electronic countermeasure and radio management.

The invention belongs to the field of pattern recognition, and particularly relates to a robust face recognition method based on dictionary decomposition and sparse representation. The method comprises the steps of designing a dictionary decomposition model to extract class specific information in a face image from a given face image training data set, then calculating a mapping matrix to describe a mapping relation between the class specific information and original training data, correcting the tested image according to the calculated mapping matrix, then reducing the dimensionality by using principal component analysis (PCA), and finally performing recognition classification via a sparse representation classifier (SRC). The method can effectively avoid the problem that the recognition rate is greatly reduced in the SRC recognition process because the training data is polluted or shaded or missing, and can achieve a high and stable recognition effect.

The present invention relates to an apparatus for iris recognition using multiple iris templates which are registered for each individual, and to an apparatus for performing the method. The method comprises the steps of: acquiring single or multiple eye images of a person to be registered; generating a registered iris template group from the acquired single or multiple eye images, and registering the group in a database; photographing an eye with a camera multiple times and acquiring images thereof for authentication; generating multiple iris templates for authentication from the photographed and acquired eye images to configure an iris template group for authentication; comparing the iris template group for authentication with the registered iris template group for each registered person stored in the database; and performing authentication or identification using the result of the comparison.

The invention discloses a driver lane-change depth warning method for a high-speed driving environment. The driver lane-change depth warning method comprises the steps that rear side images of a vehicle are captured by using a camera, and the captured rear side images of the vehicle are transmitted to a computer; through establishment of a deep learning network for driver lane-change depth warning, an end-to-end mapping between input of images of expressway vehicles of the rear side of the vehicle captured by the camera and output vehicle types is completed; vehicle type identification and recognition frame labeling are completed at the same time by using the deep learning network, and distance of the images of output recognition vehicle types is calculated; and then vehicle operating conditions are determined through a vehicle ECU, if a vehicle type is detected in a 0-100m detection range, a driver is given pre-warning to achieve the driver lane-change depth warning safety in the high-speed driving environment. The driver lane-change depth warning method reduces the error rate of recognition, improves the recognition accuracy, and realizes real-time monitoring; and real-time automatic identification, distance measurement, and the pre-warning of vehicle types at the rear side of the vehicle are realized.

The invention discloses an interactive advertisement management method and an interactive advertisement management system. The method comprises the following steps: monitoring a web page by using a client, identifying a display area of an advertisement when the advertisement in the web page is monitored, and acquiring the attribute information of the advertisement; receiving operation indication information aiming at the advertisement in the display area of the advertisement by using the client, extracting an operation track corresponding to the operation indication information, generating track data, and transmitting the track data and the attribute information corresponding to the advertisement to a server; searching advertisement logo data corresponding to the attribute information in a database according to the attribute information by using the server, matching the logo data and the track data to generate matching result information, and generating an advertisement control instruction to manage and control the advertisement according to the matching result information. Through the method and the system, the problem of low advertisement information identification rate because the conventional internet advertisement lacks of interactivity is solved.

The invention discloses an automatic band card number recognition device based on digital image processing. The device comprises a box, a computer and a display. A black light absorption material is arranged on the inner side of the box, a camera and an LED background light source are installed in the box, a recognition window with a card support is formed in the outer portion of the box, card surface image information of a bank card is extracted by a camera through photographing, the camera is connected with the computer via a data line and transmits the image information, and processed card number data is displayed by the display connected with the computer for outputting. The automatic band card number recognition device based on digital image processing provided by the invention adopts an optical non-contact type bank card number extraction mode, and is more convenient and durable compared with an existing bank card number recognition system based on magnetic stripe reading. The processing mode is based on universal computing software, so that an advantage of intelligently recognizing a bank card number is provided. The device of the invention can be applied to the social financial aspect such as automatic money depositing and withdrawing of banks and card number statistics of enterprise financial departments.

The invention relates to a special embedded type two-dimensional code recognition method. The method includes the first step of generating a two-dimensional code image suitable for a special domain, the second step of carrying out image graying, image binaryzation, edge extraction and initial locating and geometrical cutting on the two-dimensional code image, searching three view finding graphs and carrying out rotating processing, the third step of starting to determine the version number after image rotating, the fourth step of establishing a sampling grid by combining the specific positions of the three view finding graphs, sampling data and converting the image to a data matrix, the fifth step of decoding, the sixth step of verifying and the seventh step of displaying decoded decoding data. The special embedded type two-dimensional code recognition method has the advantages that the generated two-dimensional code image is fixed to a certain version or several versions, so reduction of the recognition rate is avoided, and the work needing to do to adapt to many versions is avoided at the same time; when the two-dimensional code image is generated, the error correction grade should be improved to the greatest extent, and the recognition rate is further improved. The verification is carried out when the two-dimensional code image is generated, and the introduction of the technology further improves the recognition rate.

The invention relates to the field of acoustic fish identification, in particular to a fish identification method with multi-feature and multidirectional data fused. The method comprises the steps that acoustical signals are emitted to the underwater, and fish body multidirectional acoustic scattering signals are acquired; the acquired multidirectional acoustic scattering signals are normalized and filtered; the preprocessed signals are subjected to multi-feature extraction; the preprocessed multidirectional acoustic scattering data are subjected to orthogonal transformation, envelopes are extracted, the wavelet packet coefficient singular value features, the time domain mass center features and the frequency domain mass center features of envelope information are extracted, and feature fusion and dimension reduction processing are conducted. The multidirectional data acquiring method is simple and easy to implement. Based on the extracted multiple features, the multidirectional acoustical scattering features are subjected to collaboration fusion, the fusion degree is high, fusion is compact, and the problems that identification is not clear, and correct identification even cannot be achieved when only single-dimension acoustical scattering information is classified can be effectively solved.

The invention discloses a second-generation ID card identification method with high efficiency and robustness. The second-generation ID card identification method with high efficiency and robustness includes the following steps: 1, inputting: 1) reading the storage information in a second-generation ID card chip; and 2) CIS-scanning two front and back side images of the ID card; 2, processing: 1) key element identification; 2) identification photo identification; and 3) character identification; and 3, outputting: gathering all the identifications in the step 2 and obtaining the authenticity information of the ID card. The second-generation ID card identification method with high efficiency and robustness has the advantages of 1) using Chinese character comparison and not using Chinese character identification, thus avoiding the problem that the Chinese character identification rate is low; and 2) not using face identification, but using certificate comparison, thus avoiding the problem that people just looks like the photo, and practically the people is not corresponding to the photo; and 3) being quick and efficient, and being high in robustness.

The invention discloses a multi-task vehicle component identification model, method and system based on deep learning. The method comprises the following steps: establishing a vehicle component database based on a vehicle image database and marking the vehicle component, performing image data enhancement on the vehicle component database to obtain a vehicle component training set; training a deepresidual network by using the vehicle component training set to obtain the vehicle component identification network; counting concurrence probability of different types of multiple vehicle componentsto obtain the joint probability of multiple vehicle components, and establishing a data set for the multi-task vehicle component identification and the corresponding multiple labels based on the jointprobability of multiple vehicle components; and training the vehicle component identification network to obtain the multi-task vehicle component identification model. The to-be-detected vehicle imageis identified by using the multi-task vehicle component identification model, thereby obtaining the probability of each vehicle component in the to-be-detected vehicle image. The network disclosed bythe invention is simple in training, easy to converge, easy to acquire data and high in identification accuracy rate.

The invention discloses a BP neural network-based steel seal character recognition method, which belongs to the technical field of image recognition. The method comprises the following steps of: photographing a workpiece steel seal through an industrial camera arranged in an industrial field, and acquiring an image; performing threshold segmentation on the image through a machine learning clustering algorithm. A good segmentation effect is achieved; the problem that features and character backgrounds cannot be accurately segmented through traditional single threshold segmentation for steel seal pictures is solved. Meanwhile, a clustering algorithm is applied to character segmentation, automatic segmentation of characters in the image is achieved, normalization processing of the image solves the problem that the size of the image is changed due to the fact that position deviation possibly exists in the moving process of the workpiece, and the accuracy of steel seal recognition is improved; meanwhile, training of the steel seal recognition model is achieved through the neural network, and the model has a good effect in a test set.

The invention discloses a cross-age face recognition method and system based on ternary constraints, and belongs to the technical field of computer vision. According to the scheme, the method comprises steps of obtaining face sample data of different age spans, constructing an estimation model of age attributes, and estimating age attribute information for a training sample set, dividing three time units, respectively performing feature extraction on subsets of different age groups, taking the center features of different age groups of each identity object as a basis, continuously performing close constraint on the center features through features obtained by ternary sample set training, so that the feature aggregation effect of different age groups is improved, the final output result can be directly applied to a face recognition system. The features of different age groups are aggregated to obtain face features with finer granularity, and the problem that an existing face recognition method is low in recognition rate in cross-age-group face recognition can be effectively solved.

A shielding door obstacle detection method is characterized by firstly judging whether that scene is changed significantly after the image is taken; if the scene is changed significantly, the detection area in the taken image is compared with the detection area in the train sample image to determine whether an obstacle exists or not; if the scene is not changed significantly, the obstacle recognition model is used to identify whether an obstacle exists or not. The invention also relates to a shielding door obstacle detection system, comprising a camera arranged above a gap between the shielding door and a vehicle door, a data processing unit electrically connected with the camera, a data storage unit electrically connected with the data processing unit, and a background computer communicatively connected with the data storage unit through a network communication interface. The screening door obstacle detection method and the system adopt different recognition schemes according to the scene change situation, thereby effectively avoiding the influence of the scene change on the foreign object recognition and improving the accuracy of the obstacle recognition.