233results about How to "Improve recognition results" patented technology

The invention discloses a method for recognizing voice. The method comprises the following steps of: acquiring audio data; acquiring a Lattice result of the audio data, wherein the Lattice result comprises time point information, a plurality of pieces of candidate information and matching likelihood scoring information; acquiring confidence scoring information according to the plurality of piecesof candidate information and the matching likelihood scoring information; rearranging the plurality of pieces of candidate information by using a stronger voice model and providing the optimal recognition result; positioning a voicing position corresponding to the audio data and simultaneously displaying other candidate words; selecting or inputting a correct text to finish amendment and freezingthe amended text; and searching a related text training language model by using a search engine according to the amended text serving as a key word, interpolating to acquire an adaptive language model, and returning and newly recognizing the rest part of audio data by using the adaptive voice model. By using the technical scheme, the voice recognition rate can be improved, and the workload of manual checking can be reduced.

The invention relates to a multi-modal emotion recognition method based on attention feature fusion. According to the multi-modal emotion recognition method based on attention feature fusion, final emotion recognition is carried out by mainly utilizing data of three modes of a text, a voice and a video. The method comprises the following steps: firstly, performing feature extraction on data of three modes; in the text aspect, bidirectional LSTM is used for extracting text features, a convolutional neural network is used for extracting features in a voice mode, and a three-dimensional convolutional neural network model is used for extracting video features in a video mode. Performing feature fusion on the features of the three modes by adopting an attention-based feature layer fusion mode;a traditional feature layer fusion mode is changed, complementary information between different modes is fully utilized, certain weights are given to the features of the different modes, the weights and the network are obtained through training and learning together, and therefore the method better conforms to the whole data distribution of people, and the final recognition effect is well improved.

The invention discloses a medical record structured analysis method based on medical field entities, and the method comprises the steps of 1) building a medical entity and an attribute category tablefor a common medical record text, and carrying out the corresponding relation mapping; 2) identifying the medical entity in the medical record text by adopting a Bert _ BiLSTM _ CRF model; 3) segmenting the medical record text according to semantics to form events; 4) recombining the events; 5) constructing an attribute recognition model, and extracting the attributes in the segmented events; 6) connecting the medical entities of the events in the same sentence by utilizing the knowledge graph to obtain the relationship between the entities, and 7) customizing different attribute recognition models for different types of medical record text segments, and finally forming a final medical record structured analysis text according to the text sequence accumulating structured analysis results.

The invention discloses a speech classification method based on a deep neural network, and aims at solving different speech classification problems through a unified algorithm model. The method of theinvention includes the steps of S1, converting a speech into a corresponding spectrogram; segmenting the complete spectrogram along the frequency domain into blocks to obtain a local frequency domaininformation set; S2, taking the complete and local frequency domain information as inputs of a model respectively, and based on the different inputs, the convolutional neural network being capable ofextracting local and global features; S3, using an attention mechanism to fuse global and local feature expressions to form a final feature expression; S4, using tagged data to train the network by gradient descent and back propagation algorithms; and S5, using the trained parameters for an untagged speech and taking the classification of highest probability that the model outputs as a predictionresult. The method of the invention realizes a unified algorithm model for different speech classification problems, and improves the accuracy on multiple speech classification problems.

The invention discloses a complex table and a method for identifying handwritten numbers in the complex table. A complex template table is preprocessed by line detection and angular point detection to realize classified row and column ordering and obtain structural description of cells of the template table; after an electronic handwritten table is obtained, inclination of the electronic handwritten table is corrected, and the electronic handwritten table matches the template table to obtain cell position description thereof; each cell is processed by removing the edge and reserving characters in the cell as completely as possible; number images are extracted from the cells, and a trained classifier of a data set is used to identify the number images; and the handwritten characters are post-processed, and an identification result is filled into the template table. The method of the invention is easy to realize, and better in identification effect, and provides a good solution to automatic identification and table filling later.

The invention discloses an automatic identification system of a number plate on the basis of a simplified convolutional neural network. The convolutional neural network comprises an input layer, a convolutional layer, a pooling layer, a hidden layer and a classification output layer and solves the problem of number plate identification under a daily background. The number plate identification comprises the following steps: positioning, segmenting and identifying. The invention puts forward a positioning method which extracts colorful edges by colorful edge information and colorful information. Since parameters in the method are set on the basis of color features, noise in the daily background can be effectively inhibited, and input images of different sizes can be subjected number plate extraction. The automatic identification system omits a front convolutional layer of a traditional depth convolutional neural network and only keeps one layer of convolutional layer and one hidden layer. As the supplementation of a missing convolutional layer and the strengthening of input features, a gray level edge image obtained by a Sobel operator is used as the input of a colorful image, i.e., coarsness features which are artificially extracted replace features extracted by multiple convolutional layers of the traditional convolutional neural network.

Techniques for generating, distributing, and using speech recognition models are described. A shared speech processing facility is used to support speech recognition for a wide variety of devices with limited capabilities including business computer systems, personal data assistants, etc., which are coupled to the speech processing facility via a communications channel, e.g., the Internet. Devices with audio capture capability record and transmit to the speech processing facility, via the Internet, digitized speech and receive speech processing services, e.g., speech recognition model generation and/or speech recognition services, in response. The Internet is used to return speech recognition models and/or information identifying recognized words or phrases. Thus, the speech processing facility can be used to provide speech recognition capabilities to devices without such capabilities and/or to augment a device's speech processing capability. Voice dialing, telephone control and/or other services are provided by the speech processing facility in response to speech recognition results.

The invention relates to a character detection and recognition method for boarding pass information verification, and belongs to the field of computer vision. The method comprises the following steps:S1, reading a boarding pass image, and obtaining a boarding pass test image and a training image; S2, positioning each text block through a text line detection method of a multi-task full convolutional neural network model based on a fuzzy region; S3, through text recognition model learning based on a CTC and a self-attention mechanism, realizing recognition of a text line, namely a positioned text block; S4, establishing a boarding pass common text library so as to learn an n-gram language model, and assisting in optimizing a text line recognition result. The boarding pass character information is automatically detected and recognized, Chinese and English mixed text line recognition is achieved, and more comprehensive personal information is obtained.

The invention discloses a license tag recognizing and vehicle speed measuring method based on videos. With the method, license tags from every frame of a video image are recognized, and the license tag recognizing results are calculated out according to the historical information, a higher recognition rate is realized. Meanwhile, the central position of characters can be positioned and tracked with the method to realize the accurate measurement of the vehicle speed. A vehicle head lower edge line detecting and tracking module positions a vehicle head lower edge line according to prospective motion and prospective edge detecting results, efficiently avoids the interference of shadows, captures the position of the vehicle head lower edge line, and tracks. After a trigger line is reached, the vehicle head lower edge line detecting and tracking module carries out space mapping on the historical tracking position to realize the accurate measurement of the vehicle speed. A vehicle lamp pairdetecting and tracking module eliminates the interference of the circumference and headlamps of the vehicle in a night mode, accurately detects the vehicle lamp pair, and tracks. After the trigger line is reached, the vehicle lamp pair detecting and tracking module carries out space mapping on the historical tracking position to realize the accurate measurement of the vehicle speed.

The invention provides a high-speed and high-resolution number collecting device of a banknote sorting machine and an identification method. The high-speed and high-resolution number collecting device of the banknote sorting machine comprises image sensors, a multichannel A/D (analog to digital) converter chip, a programmable logic chip, a digital signal processor chip, a dynamic memory chip and a communication structure chip, wherein the image sensors are respectively connected with the multichannel A/D converter chip, the multichannel A/D converter chip is connected with the programmable logic chip, and the programmable logic chip is respectively connected with the digital signal processor chip and the dynamic memory chip. The invention realizes the high-resolution collection to banknote images under the condition of high-speed operation of the banknotes. The contact type image sensors have the advantages of low cost, small volume and low requirements for light source performance. Positioning and cutting thresholds are not needed to be set in the identification method of banknote numbers, so that an accurate positioning and cutting position and ideal identification results can be obtained when the banknotes are depreciated and polluted.

The invention relates to a ring tone recognition method and system for a call center system. The method comprises the steps of firstly, calling the input telephone number; then, initiating a fax request to a called user, and preliminarily recognizing a fax machine via SDP (Session Description Protocol) analysis; recognizing a ring tone state, wherein based on an audio fragment within 5 seconds after calling, the ring state before the called user answers being standard ring-back tone or color ring tone is recognized; and after the called user hooks off, recognizing fax machine off-hook, automatic responder off-hook, natural person off-hook and get-through nobody answer based on voice analysis of a voice fragment after off-hook answer. By adopting the method, the number state and the terminal type can be accurately recognized, the calculation quantity of ring tone state recognition is reduced, the recognition result is quickly given, and the recognition efficiency is improved.

The invention provides a fuzzy domain characteristics based optical fiber vibration signal identifying method and mainly aims to solve the problem that existing identifying methods are low in identifying efficiency under the conditions of low sampling rate and similar intrusion events. The fuzzy domain characteristics based optical fiber vibration signal identifying method includes steps of (1) subjecting optical vibration signals to wavelet denoising; (2) subjecting the denoised signals to de-meaning and energy normalization; (3) calculating the normalized signal fuzzy function and sectioning the fuzzy function to be fuzzy domain characteristics; (4) reducing dimensions of sections to construct a signal characteristic set; (5) partitioning the signal characteristic set into a training set and a test set; (6) training an SVM (support vector machine) classifier by the training set; (7) classifying the test set by the trained SVM classifier. The fuzzy domain characteristics based optical fiber vibration signal identifying method effectively extracts the fuzzy domain characteristics of optical fiber vibration signals, has the advantages of high identification rate and wide applicability as compared with the prior art, and can be used for a signal processing subsystem of an optical fiber perimeter security system.

The invention discloses a dynamic time sequence convolutional neural network-based license plate recognition method. The method comprises the following steps of: reading an original license plate image; carrying out license plate angle correction to obtain a to-be-recognized license plate image; inputting the to-be-recognized license plate image into a previously designed and trained convolutionalneural network so as to obtain a feature image and time sequence information, wherein the feature image comprises all the features of the license plate; and carrying out character recognition, inputting the feature image into a convolutional neural network of a long and short-term memory neural network layer on the basis of time sequence information of the last layer so as to obtain a classification result, and carrying out decoding by utilizing a CTC algorithm so as to obtain a final license plate character result. According to the method, vision modes are directly recognized from original images through using convolutional neural networks, self-learning and correction are carried out, the convolutional neural networks can be repeatedly used after being trained for one time, and the timeof single recognition is in a millisecond level, so that the method can be applied to the scenes needing to recognize license plates in real time. The dynamic time sequence-based long and short-termneural network layer is combined with CTC algorithm-based decoding, so that recognition error problems such as leak detection and repeated detection are effectively avoided, and the algorithm robustness is improved.

The invention relates to a facial expression image recognition method based on an expert system. According to the method, inference and recognition of the facial expression of a preprocessed image are carried out through the expert system established on the basis of an expression image processing method and the function of a traditional computer program. The method comprises the following steps: (1) capturing an image in a video, acquiring user information in the video, then carrying out identity verification through image processing and image characteristic extraction, acquiring characteristic parameters of the expression image of a user, determining a user expression library, and establishing the expert system for facial expression recognition; (2) carrying out imaging processing and image characteristic extraction on the image captured in the video, acquiring the characteristic parameters generated when the degree of the expression of the user is maximized, comparing the characteristic parameters with parameters, determined in the step (1), of image training samples in the user expression library, and finally outputting the statistical result of facial expression recognition through an inference engine of the expert system. Compared with the prior art, the method has the advantages of being high in recognition speed and the like.

Systems, methods and apparatus for generating, distributing, and using speech recognition models. A shared speech processing facility is used to support speech recognition for a wide variety of devices with limited capabilities including business computer systems, personal data assistants, etc., which are coupled to the speech processing facility via a communications channel, e.g., the Internet. Devices with audio capture capability record and transmit to the speech processing facility, via the Internet, digitized speech and receive speech processing services, e.g., speech recognition model generation and/or speech recognition services, in response. The Internet is used to return speech recognition models and/or information identifying recognized words or phrases. The speech processing facility can be used to provide speech recognition capabilities to devices without such capabilities and/or to augment a device's speech processing capability. Voice dialing, telephone control and/or other services are provided by the speech processing facility in response to speech recognition results.

Disclosed are a method and a device for real-time mark for a high-energy X-ray dual-energy imaging container inspection system in the radiation imaging field. The method comprises the steps of emitting a first main beam of rays and a first auxiliary beam of rays having a first energy, and a second main beam of rays and a second auxiliary beam of rays having a second energy; causing the first and second main beams of rays transmitting through the article to be inspected; causing the first and second auxiliary beams of rays transmitting through at least one real-time mark material block; collecting values of the first and second main beams of rays that have transmitted through the article to be inspected as dual-energy data; collecting values of the first and second auxiliary beams of rays that have transmitted through the real-time mark material block as adjustment parameters; adjusting the set of classification parameters based on the adjustment parameters; and identifying the substance according to the dual-energy data based on adjusted classification parameters. The method according to the invention simplifies the mark procedure for a substance identification subsystem in a high-energy dual-energy system while improves the stability of the material differentiation result of the system.

The invention discloses an incremental learning and face recognition method based on locality preserving nonnegative matrix factorization (LPNMF), and relates to the technical field of pattern recognition. The incremental learning and face recognition method is essentially a face recognition method based on on-line learning. The method comprises the step (a) of face image preprocessing, the step (b) of initial sample training, the step (c) of incremental learning and the step (d) of face recognition. The incremental learning and face recognition method can be applied to a linear face recognition system and a non-linear face recognition system, keeps local structures of original space of face images, greatly improves recognition rate, and can be actually applied in an on-line mode.

A system and method for assisting with analysis and recognition of ink is described. Analysis hints may be associated with a field. The field may receive electronic ink. Based on the identity of the field and the analysis hint associated with it, at least one of analysis and recognition of ink may be assisted.

The invention discloses a face recognition method based on extraction of multiple evolution features. The method comprises the steps as follows: (1), classification of initial samples: the initial samples are divided into three parts, including training samples for feature extraction, training samples for weight evolution and test samples respectively; (2), feature extraction of the training samples: the training samples are subjected to feature extraction with a multiple seed space method, such as PCA (principal component analysis), LDA (linear discriminant analysis), LPP (locality preserving projection) or the like; and (3), multiple feature fusion evolution: features obtained with different feature extraction methods are fused according to a form that Phi is equal to the sum of Omega 1 Phi 1, Omega 2 Phi 2, ..., and Omega n Phi n, and the like, wherein Omega is a weight coefficient. An optimal weight coefficient is obtained with a genetic algorithm, so that fused features have better recognition effects than prior features. The face recognition method has the advantages that the principle is simple, the method is unique, the application is easy, and the like.

The invention provides a mammary gland mass image recognition method and device. The method comprises the steps that a mammary gland magnetic resonance image to be recognized is acquired; Inputting the mammary gland magnetic resonance image to be identified into a constructed mammary gland mass identification model to obtain a mass identification result in the mammary gland magnetic resonance image to be identified; Wherein the lump recognition model adopts a deep full convolutional neural network model; wherein the coding process of the deep full convolutional neural network model adopts a basic convolutional module in the U-shaped convolutional neural network model to perform feature extraction, and the decoding process of the deep full convolutional neural network model adopts dense connection to perform size unification on to-be-fused feature maps and then perform fusion. According to the embodiment of the invention, automatic recognition of the breast lumps is realized, and the recognition accuracy of the breast lumps is improved.