911results about How to "Improve recognition accuracy" patented technology

A global speech user interface (GSUI) comprises an input system to receive a user's spoken command, a feedback system along with a set of feedback overlays to give the user information on the progress of his spoken requests, a set of visual cues on the television screen to help the user understand what he can say, a help system, and a model for navigation among applications. The interface is extensible to make it easy to add new applications.

Calculates a word n-gram probability with high accuracy in a situation where a first corpus), which is a relatively small corpus containing manually segmented word information, and a second corpus, which is a relatively large corpus, are given as a training corpus that is storage containing vast quantities of sample sentences. Vocabulary including contextual information is expanded from words occurring in first corpus of relatively small size to words occurring in second corpus of relatively large size by using a word n-gram probability estimated from an unknown word model and the raw corpus. The first corpus (word-segmented) is used for calculating n-grams and the probability that the word boundary between two adjacent characters will be the boundary of two words (segmentation probability). The second corpus (word-unsegmented), in which probabilistic word boundaries are assigned based on information in the first corpus (word-segmented), is used for calculating a word n-grams.

The invention provides a barrier identification method and system of a laser radar. The method comprises that S1) original point cloud data, position data and attitude data are fused to obtain fused point cloud data; S2) the fused point cloud data is divided into multiple fused point cloud data segments according to the time sequence, and ICP registering is carried out on point cloud in each fusedpoint cloud data segment to obtain superposed point cloud data; S3) point clouds of each group of superposed point cloud data segment are clustered to obtain candidate barriers, and static information of the candidate barriers is extracted; and S4) according to the static information of the candidate barriers, static and dynamic barriers are identified from the candidate barrier, and dynamic information of the dynamic barrier is extracted. A graph can be made needless of using offline data of the laser radar, the method and system are suitable for detecting barriers in different complex application environments, the identification precision is high, and the speed is high.

A method for efficiently determining and displaying pertinent information determined from multiple input and calculated parameters associated with a videoconference call. The method for efficiently determining and displaying this personal information is performed using input from the user at an endpoint and calculated information throughout the videoconference to present personal information, about the currently speaking person, to all participants. Videoconferencing systems are typically used by multiple people at multiple locations. The method of this disclosure allows for more user interaction and knowledge transfer amongst the participants. By sharing information between the different locations participants are more aware of who is speaking at any given time and the importance to be applied to what that particular person is saying.

The invention belongs to the information extraction technology in the artificial intelligence field, and provides a semi-supervised Chinese named entity recognition method based on deep learning for the Chinese text in the specific field with only a small amount of labeled data and a large amount of unlabeled data. Specifically, the method includes the following steps: firstly, constructing a character-based deep learning named entity recognition model; secondly, design a scoring device, train a learning device and a scoring device by using that annotated data; thirdly, designing a semi-supervised learning framework to realize semi-supervised Chinese named entity recognition using unlabeled data. The invention can improve the accuracy and recall rate of the Chinese named entity recognitionin the specific field.

The invention discloses a rotating shaft geometric error identification method commonly used for a five-axis numerical control machine tool. The method comprises the steps that according to the type of the five-axis numerical control machine tool, a ball rod instrument measuring mode is determined; according to a numerical control machine tool geometric error model, a ball rod instrument sensitive vector is used for obtaining a ball rod instrument reading model; according to the ball rod instrument reading model, the ball rod instrument measuring mode, the property of a geometric error item and a horizontal moving shaft geometric error item, a machine tool rotating shaft geometric error item expression is obtained; proper mounting parameters are selected, the machine tool is operated, and corresponding ball rod instrument reading is obtained; and three horizontal shaft geometric error items of the five-axis numerical control machine tool are input, and according to the rotating shaft geometric error item expression and the ball rod instrument reading, 16 items of geometric errors of two rotating shafts of the five-axis numerical control machine tool are obtained. The method is suitable for different types of five-axis numerical control machine tools, all the 16 items of geometric errors of the two rotating shafts of the machine tool can be obtained, meanwhile, mounting is easy, measuring time is short, the identifying results accord with the property of the geometric error items, and measuring accuracy is high.

A face recognition that is robust to external illumination variations is provided. The face recognition apparatus includes a face localizer extracting a predetermined number of feature points from a facial region extracted from an input image, a Gabor filter unit applying a set of Gabor filters with a plurality of orientations and frequencies to each of the facial feature points in order to obtain filter responses and generates a one-dimensional (1-D) vector consisting of the filter responses, a training data set storage storing the 1-D vector when the gaber-filtered 1-D vector is obtained from a training image, and a binary classifier generating a binary determination criterion using the stored 1-D vector, applying the binary determination criterion to the 1-D vector when the gaber-filtered 1-D vector is obtained from an image to be recognized and determining the identity of the image to be recognized.

The invention discloses a detecting system and a detecting method for fatigue driving of a driver, and belongs to the technical fields of image processing and pattern recognition. The system comprises a human face localization module, a human eye state judging module, a mouth state judging module and a fatigue determining module, wherein the human face localization module is connected with the human eye state judging module and the mouth state judging module for communication; the determining result is sent to the fatigue determining module. The detecting method comprises the following steps: 1, localizing and detecting the human face; 2, judging the human eye state; 3, judging the mouth state of the driver; 4, judging the mental state of the driver. The detecting system and the detecting method combine two obvious facial characteristics of the driver, namely the human eyes and the human mouth, for complex judging; compared with recognition detection using a single method, the method provided by the invention has a higher recognition accuracy; through determining the mental state of the driver, the driver under fatigue state is reminded to stop the vehicle for rest, so that traffic accidents are effectively reduced, and the safety of life and property is strongly guaranteed.

The invention provides a semi-supervised-transfer-learning character recognition method based on a convolutional neural network. The method includes: using batch character image samples, which are ina target domain and have no class label, as a test sample set to input the same to the convolutional neural network after semi-supervised transfer learning, and identifying character images of the test sample set, wherein the convolutional neural network after semi-supervised transfer learning is obtained by joint training of batch character image samples which are in a source domain and have class labels, batch character image samples which are in the target domain and have class labels, and batch character image samples which are in the target domain and have no class label, and recognitionprecision is improved. According to the semi-supervised-transfer-learning character recognition method based on the convolutional neural network provided and a semi-supervised-transfer-learning character recognition system based on the convolutional neural network provided by the invention, the large number of source-domain samples with the class labels, the small number of target-domain sampleswith the class labels and the relatively easily obtained target-domain samples without a class label can be utilized for semi-supervised transfer learning, and adaptation ability of the convolutionalneural network on the target-domain samples is improved.

The invention discloses a method for identifying a Web named entity based on a statistical model. The method comprises the following steps of: representing multiple characteristics of the Web named entity with structure and text characteristics; combining a statistical method with a rule method and adopting an improved MR-GHMM (MR-Generalized Hidden Markov Model) to increase the training efficiency; marking the entity with the improved GHMM, and marking each named entity to realize entity identification; and processing a Web complex named entity identifying process on two layers and performing complex nested entity identification by taking a marking result of a first layer as the input of second layer processing. Compared with an original identifying algorithm, the method has the advantages that: the identifying accuracy of an algorithm used in the method is increased, and the time complexity of model training is lowered greatly. By representing multiple characteristics of the Web named entity and modifying entity characteristics in different fields, named entities in different fields on Web can be identified.

The invention relates to a regional convolutional neural network-based efficient gesture detection and recognition method. The method includes following steps that: Chinese character gesture alphabetsample images are preprocessed; a gesture image data set is constructed and strengthened; a regional convolutional neural network-based Faster R-CNN is utilized to perform gesture detection and recognition: a feature extraction network extracts gesture features and divides an extracted feature graph into two parts, the first part directly enters a Fast R-CNN for deep convolution, the second part enters an RPN, a regional proposal is generated, the regional proposal is inputted into the Fast R-CNN, the regional proposal and a feature graph obtained in the first part together enter a RoI poolinglayer and then enter a full joint layer, position regression and gesture category scores are obtained, and finally, gesture detection and recognition can be realized; and the network model is trained, Chinese character gesture alphabet detection and recognition can be realized. The method of the invention can improve recognition speed and accuracy.

The invention provides a radar target recognition method based on micro-Doppler feature extraction and deep learning. The new method can be divided into three stages of step 1, a micro-Doppler featureextraction stage: using an MSTFT-WVD algorithm to obtain a micro-Doppler feature time-frequency graph of a target; step 2, a classifier model training phase: dividing the time-frequency graph of theknown category information obtained in the first stage into a training data set and a test data set, utilizing an original deep learning model for training, and continuously adjusting parameters to obtain the optimal deep learning model for radar target recognition; step 3, an identification phase of an unknown target: using the MSTFT-WVD algorithm to obtain a micro-Doppler feature time-frequencygraph of the unknown target, inputting the time-frequency graph to the trained deep learning model in the second stage to obtain the category information of the unknown target. The radar target recognition method can achieve higher recognition accuracy rate.

The invention discloses an identification method of fake-licensed cars, and has the advantage that fake-licensed cars can be found by processing in the invention just when more than two cars with the same license plate numbers appear on roads at the same time no matter the distance is far or near after fake-licensed cars set off along roads because of the processing of all cars moving on roads that are considered as an integral; the identification precision is high; as the invention is realized by a time matrix, an input buffer area, a window concordance list and a window data storage area positioned in a vehicle data information processing center, the vehicle information obtained by a video capturing device is transmitted to the input buffer area in real time; and the fake-licensed cars are identified by on-line processing of buffer data in the input buffer area, thereby greatly improving the identification speed. In addition, the on-line identification processing of the method in the invention is realized in real time by setting the time matrix, the input buffer area, the window concordance list, the window data storage area, and the like in the existing vehicle data information processing center, so the investment is few.

The invention provides a setup method of a network flow classification model. The method includes that identification indexes are selected and extracted from the stream data packets of the caught network flow; an initialization cluster center is selected under the set-up cluster numbers according to the identification indexes; a clustering operation is taken on the identification indexes and a new cluster center is newly computed; the normalized mutual information values of all clustering operation results are computed respectively under the set-up cluster numbers and one of the clusters is selected as the best cluster number according to the normalized mutual information values; the clustering operation result under the best cluster number and the new cluster center are selected as the network flow classification model according to the obtained best cluster number. The invention verifies if the selection of the cluster numbers is right by the computation of a NMI value and gives better clustering feature and higher identification precision by the selection of the best cluster number.

The invention discloses a feature image extraction method based on a deformable convolutional layer. The method comprises the steps that a target image is acquired; and pixel values are extracted from the target image through the sampling points of the convolutional kernel in the convolutional layer so as to obtain a feature image, wherein the actual coordinate values of the sampling points are the actual coordinate values calculated according to the preset initial coordinate values and the deviation variable trained in advance. According to the method, one deviation variable can be additionally arranged for each sampling point in the convolutional kernel so that a convolutional neural network is enabled to realize the capacity of learning image space geometric deformation, the convolutional kernel is enabled to randomly sample near the present position, the adaptability of the convolutional layer for the deformed image in extracting the feature image can be increased and the identification accuracy for the deformed object can be increased. The invention also discloses a feature image extraction device based on a deformable convolutional neural network. The feature image extraction device also has the beneficial effects.

The present invention discloses a non-intrusive resident load identification method based on S_Kohonen. The method comprises the steps: the step 1: determining a switching event according to the changing of an active power at a home electric power inlet, and when the switching event happens, collecting current samples of electric appliances having generation of the switching events at the home electric power inlet; the step 2: performing frequency-domain analysis of the collected electric appliance current samples, extracting the frequency-domain harmonic amplitudes of the collected electric appliance current samples as load features of each electric appliance, and forming a load feature base; the step 3: designing an S_Kohonen neural network suitable for resident load identification, and determining the number of nerve cells of the input layer and the output layer of the S_Kohonen neural network and the scale of the competition layer to determine a selection mode for obtaining the nerve cells and a learning algorithm of weight regulation; the step 4: performing parameter initialization; the step 5: performing training of the S_Kohonen network through a training set and performing test of the training set; and the step 6: regulating the network parameters to realize optimal network performances.

The invention discloses a blind identification method of coding parameters of an RS code of an error-tolerant code, which relates to the technologies of intelligent communication, communication reconnaissance and information security in the communication field. In the method, the characteristic that continuous zero roots are generated on the same specific positions after the coding parameters of the same RS code undergo Fourier transformation in Galois field is adopted, blind identification of the coding parameters of the RS code is realized by the communication contents and the aims of intelligent communication and communication reconnaissance are achieved. The invention is characterized by simple algorithm, stable and steady performance, fast identification, high accuracy and the like and is especially suitable for the channel code identification algorithm in such fields as intelligent communication, communication reconnaissance, wireless electronic detection and communication countermeasure, etc.

The invention discloses a vehicle fault detection system and method based on a vehicle-mounted CAN network. According to the system, a CAN bus control module is used for generating fault information when fault occurrence is judged and sending the fault information and whole vehicle data in a preset period through a fault transport protocol; a computer is used for analyzing the fault information and the whole vehicle data according to the fault transport protocol and determining the fault reason with the maximum association possibility to the whole vehicle data and a solution corresponding to the fault reason according to the fault information, the whole vehicle data and prestored fault analysis data. By means of a CAN bus processing device and the computer of the system, vehicle fault detection can be continuously, automatically, promptly and effectively achieved in real time, the fault source can be quickly found and positioned, the possible reason of the fault and the solution are found in an assisting mode, the processing speed and the identification accuracy rate are high, safety and reliability are achieved, timeliness is high, manpower and material resources are greatly saved, and cost is saved.

The invention relates to an image identification method and an image identification device. The method comprises the following steps: receiving an input to-be-identified image; acquiring global features and local features of a target object in the to-be-identified image; and acquiring to-be-identified information of the target object according to the global features and the local features. According to the technical scheme, the global features and the local features of the to-be-identified image can be acquired when the to-be-identified image is received, and then, the to-be-identified information of the target object displayed in the to-be-identified image is identified accurately according to the global features and the local features. The identification accuracy of the to-be-identified information is improved. Identification error due to high similarity and small difference between the to-be-identified information and the information of other objects in the category to which the target object belongs is avoided as much as possible.

The embodiment of the invention discloses a vehicle recognition method and device and a storage medium According to the embodiment of the invention, the method comprises the steps: carrying out the combination of each two collected vehicle sample images, so as to build sample pairs; merging each sample pair into one multichannel image, and then adding the multichannel images to a training sample set; training a preset twin neural network model according to the training sample set, and obtaining a trained model, so as to achieve the recognition of a to-be-recognized image based on the trained model during the vehicle recognition, such as the recognition of a target vehicle from mass images, and so on. According to the scheme of the invention, the method can achieve the recognition of a vehicle through the building of a model. Compared with a conventional scheme based on human eyes or simple matching, the method can improve the recognition efficiency and accuracy.

The invention provides an outdoor scene recognition method fused with sensor information. By using the method, an outdoor scene can be quickly recognized and the recognition correct rate is high. The method comprises the specific steps of 1, a mobile terminal collects an image to be recognized of the current scene, and gravity information which is produced when the mobile terminal collects the image is acquired; 2, the image to be recognized and the gravity information are packed into a compressed file and the compressed file is sent to a server; 3, the server calculates the gravity direction theta g of the image to be recognized according to the gravity information in the compressed file; 4, lines in the image to be recognized are detected; 5, wrong lines in horizontal and vertical lines are removed; 6, horizontal vanishing points and vertical vanishing points are calculated; 7, perspective distortion correction is carried out on the image to be recognized; 8, image matching is carried out, and the most similar sample image is searched; and 9, the scene information corresponding to the sample image is returned to the mobile terminal, so as to realize scene recognition.

The invention discloses an identification system and method for handwritten character strings in air. The system comprises a gesture identification module, a finger track extraction module, a pretreatment module and a character string identification module, wherein the gesture identification module is connected with the finger track extraction module and used for recognizing and identifying gestures for starting and ending writing; the finger track extraction module is connected with the pretreatment module and used for performing finger track extraction and record on input handwriting actions; the pretreatment module is connected with the in-air handwritten character string identification module and used for pre-treating finger track data; and the character string identification module is connected with the pretreatment module and used for performing character string identification on the finger track data. The invention further discloses the identification method for the handwritten character strings in air. According to the identification system and method, the supported writing mode is novel and convenient, the identification is accurate and fast, a more humanized and intelligent handwriting input mode other than the traditional handwritten mode is provided for people, and the system and the method can be widely applied to human-computer interaction systems such as game operation, television control, a teaching system and the like.

A hardware trojan identification method based on a single classification supporting vector machine belongs to the detection and identification field of hardware trojan chips. The invention solves the problems of huge error and low efficiency in the prior art because the technology of utilizing chip side channel information to identify the chip hardware trojan requires to observe an image manually. The method provided by the invention comprises the following steps: I, preprocessing, and acquiring a side channel information matrix; II, selecting a backward analysis chip to perform backward analysis, and determining whether a hardware trojan is contained; III, dividing the backward analysis chip not containing the hardware trojan into a train sample and a train optimizing sample; IV, utilizing the side channel information matrix of the train sample to establish the characteristic space of the chip; V, acquiring the side channel characteristic data matrix of a chip to be detected; VI, performing normalization processing; VII, picking up the normalized data of the train sample and the train optimizing sample; VIII, training the single classification supporting vector machine to constitute a minimum hypersphere; and IX, if beyond the minimum hypersphere, then determining that the chip to be detected is a hardware trojan chip.

The invention discloses a method for acquiring a number of a paper currency or a financial bill. The preset position on a conveying passage is provided with a number acquisition device according to the layout position of the number. The method comprises the steps of: acquiring a regional image where the number may appear when the paper currency and the financial bill to be detected reaches the preset position; detecting the regional image, and judging whether the regional image contains the number of the paper currency; and extracting the number from the image. The invention also provides a device for acquiring the number of the paper currency or the financial bill, which comprises an image acquisition unit, a detection unit and a number extract ion unit. The invention also provides a corresponding method and a corresponding device for identifying the paper currency or the financial bill. According to the invention, only a region which may contain the number needs to be extracted, which can ensure that the processed redundant information amount is minimum, the resolution of the region to be identified is highest, the useful information amount is increased, and the identification accuracy is improved, so that the number of the paper currency or the financial bill can be quickly and accurately acquired and can be identified and managed.

The invention discloses an intelligent voice service development cloud platform and method. The intelligent voice service development cloud platform comprises an application optimization database, a content service optimization module, a voice recognition module and a semantic comprehension module, wherein the content service optimization module is used for receiving an input sentence pattern of a user customized according to a built application and output data corresponding to the input sentence pattern and storing the input sentence pattern and the output data into the application optimization database; the voice recognition module is used for obtaining the input sentence pattern from the application optimization database and updating a voice recognition model according to the obtained input sentence pattern; the semantic comprehension module is used for obtaining the input sentence pattern and the output data from the application optimization database and updating a semantic comprehension model according to the obtained input sentence pattern and the output data. Thus, optimization and customization for applications are realized, all optimization operations are done by users on line without participation of background technical support personnel of the platform, the development period for optimization and customization is shortened, a developer can rapidly implement optimization conveniently, and a special intelligent voice interaction application is customized.

The invention discloses a system for monitoring the boiler efficiency of a thermal power station on line based on identification of calorific capacity of fire coal in the technical field of information processing. The system comprises a data acquisition module, a data preprocessing module, a coal and smoke characteristic module and a boiler efficiency computation module. The method comprises the following steps of: connecting an object linking and embedding (OLE) for process control (OPC) interface and a distributed control system server to acquire the required actual operating data of a unitand storing the data in a real-time database of an upper computer; preprocessing the real-time data in an upper computer system, performing dynamic energy balance of a smoke side and a working mediumside on all heating sides of a boiler, and identifying the actual calorific capacity of fire coal in a hearth by a soft measurement method to acquire the corrected actual lower heating value of the fire coal; correcting fire coal elementary composition involving efficiency calculation by using everyday coal quality industrial analysis data; and finally, monitoring the boiler efficiency on line.

The invention discloses a video face identification algorithm on the basis of multi-instance learning. According to the algorithm, each face video is used as a packet; normalized face frame images in the videos are used as instances in the packets; a blocked local binary pattern cascaded histogram on the basis of weighting is used as an instance feature; in a multi-instance feature space of a training set, a multi-instance learning algorithm is adopted to obtain a classifier so as to realize classification and prediction on a tested sample. By related experiments in a face video database, the algorithm obtains high identification accuracy; meanwhile, the algorithm has excellent robustness for illumination variation, expression variation and the like and effectiveness of the algorithm of verified.

The invention discloses a video object identification method and a relevant system based on support vector machines; with structure training samples and according to resolution that is selected from the training samples, utilize a method combined with wavelet outline description symbol, shape factor and invariant torque, to describe outline characteristics of the training samples; gain a support vector machine model according to the outline characteristic training, and meanwhile, determine decision-making function parameters at the optimum category aspects in the support vector machine model; then, extract outline characteristics from the video object to be identified; the support vector machine model after training can follow the video object outline characteristics that is input, in order to perform category for the acquired video objects with a decision-making function operation for the optimum category face. The invention has advantages of high calculation speed, high identification accuracy, reliable arithmetic performance and multi-category identification; and moreover, with increase of objects to be identified, the identification performance can still be kept stable, and the identification speed can still meet real-time monitoring demands.

The invention provides a subway driver emotion identification method and device based on deep learning. The method comprises the following steps: obtaining a driver emotional state database, includingface images and corresponding emotional states; constructing a deep learning network model used for emotional state identification; using the driver emotional state database to train the deep learning network model, and solidifying the structure and parameters of the trained deep learning network model to obtain an emotional state identification model; and collecting a real-time working video ofa subway driver, extracting multiple continuous frames of images according to a preset frame rate, and inputting the multiple continuous frames of images to the emotional state identification model toobtain the emotional state identification result of the subway driver. The emotional state of a subway driver is monitored in real time by using a deep learning network model based on a 3D convolution neural network and a stacked auto-encoder. The identification precision is high. The abnormal working state of a driver can be found early, and the safety of driving can be ensured.

The invention relates to an internet application flow rate identification method based on message sampling and application signing, comprising the following steps: firstly, message sampling capture: in accordance with sampling strategy and sampling rate the message is captured and decoded; secondly, decoding: the flow information and application data of the message is analyzed by decoding the message; thirdly, flow classification: according to the flow information of the message, a flow state table is found and maintained; fourthly, flow state distinguishing: the signature is matched if the application type of the flow state found through the flow classification is unknown; finally, signature matching: according to the application signature bank, the application data of the message is matched, if matched successfully, the application type of the flow state is updated, and the flow information and application type of that data stream is output. The method is of high accuracy in identification, high efficiency in processing, good expandability, high possibility in realization, and is applicable not only for message processing, but also for flow data analysis. The invention can be achieved in not only the network equipment, but also the network analysis system.