44results about How to "Improve the recognition effect" patented technology

The invention discloses a method and a device for identifying cheat web-pages. The method comprises the steps as follows: obtaining a set of known webpage samples, wherein the known webpage samples are the webpage samples known whether to be the cheat web-pages or not; generating an initial support vector machine used for judging the cheat web-pages according to the set of known webpage samples; obtaining a set of a first preset amount of unknown webpage samples, wherein the unknown webpage samples are the webpage samples unknown whether to be the cheat web-pages or not; adjusting the model parameters of the initial support vector machine according to the set of unknown webpage samples; and judging whether the web-pages to be detected are the cheat web-pages by the adjusted support vector machine. By the method, the problem of poorer effect of identification of the cheat webpage identification method based on machine learning in related technology on novel cheat web-pages is solved and the effect of identifying the novel cheat web-pages is improved.

The embodiment of the invention provides a named entity recognition model training method and a named entity recognition method. According to the invention, the method comprises the steps: employing atraining set for training a BERT-CRF model to obtain a named entity recognition model trained in this round; marking the to-be-identified data set by using the named entity identification model trained in this round to obtain a weakly marked to-be-identified data set; selecting one part from the weakly marked to-be-identified data set and taking the part and the initial training set as a new training data set to continuously carry out the next round of training on the named entity identification model; therefore, the named entity recognition model uses the to-be-recognized data set to adjustthe model before recognizing the to-be-recognized data set, so the named entity recognition model has better generalization ability, and finally the recognition effect of the model on the to-be-recognized data set is improved.

The invention discloses a method for identifying a 3D model based on sparse coding. The method comprises the following steps: acquiring a set of 2D views of different 3D models to construct a database; selecting a checking 3D model to obtain low-layer features; using a sparse coding model to learn a filter function between low-layer features and primary high-layer features; obtaining low-layer features of a comparing 3D model from the database; based on the filter function, conducting convolution on the low-layer features of the comparing 3D model, obtaining high-layer features of the comparing 3D model through pooling; based on the filter function, conducting convolution on a 2D view of the comparing 3D model, obtaining high-layer features of the checking 3D model through pooling; using high-layer features of the 2D view, calculating similarity between the checking model and the comparing model; determining whether all 3D models in the database are taken as comparing 3D models; arranging similarities between the checking 3D model and all the comparing 3D models, taking the type of the 3D model which has the highest similarity as an identification result. According to the invention, the method greatly increases identification effect.

The invention belongs to the technical field of information processing. The technical problem that in the prior art, an event information extraction technical scheme is not high in accuracy is solved.The first aspect of the invention provides a text event information extraction method, and the method comprises the steps: carrying out the word segmentation of a text, carrying out the vector conversion of segmented words, obtaining a word vector, inputting the word vector into a neural network model, and outputting an entity; arranging segmented words and entities in the text blocks into structured text blocks according to grammar-defined corresponding mode rules on the basis of information types defined by the text format features; and performing event information extraction processing onthe structured text block, realizing keyword extraction by using a grammar-defined corresponding mode rule, and ooutputting keywords to a result template. Therefore, the event extraction model is configured in a mode of combining neural network deep learning and rules, and accurate extraction of text event information is realized.

The invention provides an automatic foreign name identification and control method based on context semantics in a natural language processing system by researching foreign name characteristics and combining a statistic probability model. The method is characterized by comprising the following steps of: a. analyzing a text to be identified and acquiring a candidate foreign name string set; b. correcting and screening the candidate foreign name string set by utilizing a foreign name rule set to acquire a first middle foreign name string set; c. further screening the first middle foreign name string set by utilizing probability statistics and the probability model to further screen the acquired identified foreign name set; and d. determining the unidentified foreign names according to the identified foreign name set. According to the system, the context characteristics of the names and the word characteristics of the foreign names are fully utilized, the identification error caused by word segmentation is greatly reduced, the condition that the other named entities are identified into names is well avoided, and the identification effect is improved.

The invention discloses a sand body thickness prediction method based on spectrum inversion and a sand body thickness prediction system thereof. The prediction method comprises the steps that seismic wavelet and stratum reflection coefficient are acquired from original seismic data; an objective function is established, and spectrum inversion is performed according to the objective function, the seismic wavelet and the stratum reflection coefficient after parity decomposition so as to acquire an even component reflection coefficient sequence; new seismic data are synthesized according to the seismic wavelet, the even component reflection coefficient sequence and the objective function; and the sand body thickness is predicted according to the new seismic data. The spectrum inversion result resolution is better than that of the conventional inversion so that the thin layer geological target identification effect can be enhanced, the operation cost is low and the application effect is great.

According to an EOF-based unmanned aerial vehicle signal identification system and method provided by the invention, the problem that similar unmanned aerial vehicle signals are difficult to identifyis solved, the identification effect of unmanned aerial vehicles with similar signals can be effectively improved, and better support is provided for unmanned aerial vehicle identification managementand control. The method provided by the invention comprises the following steps: 1) firstly, carrying out preliminary identification on a received unmanned aerial vehicle signal through a characteristic parameter matching method; 2) carrying out FFT (Fast Fourier Transform) processing on the unmanned aerial vehicle signal which cannot be recognized through characteristic parameter matching, and then generating a corresponding time-frequency analysis graph; 3) adopting an empirical orthogonal function analysis method to decompose a video analysis matrix corresponding to the video analysis graphinto an EOF matrix reflecting signal common characteristics and a characteristic coefficient matrix reflecting signal difference characteristics; and 4) performing feature data extraction and BP neural network training on the feature coefficient matrix, constructing a corresponding unmanned aerial vehicle type classifier, and outputting an identification result.

The application provides a bilingual hybrid speech recognition method, device and equipment and a storage medium. The method comprises the following steps: obtaining target speech, wherein the targetspeech is hybrid speech of a first language and a second language; recognizing the target speech by using a pre-established bilingual hybrid speech recognition model, and obtaining a plurality of candidate recognition texts and scores respectively corresponding to the plurality of candidate recognition texts, wherein the bilingual hybrid speech recognition model can determine a text classificationprobability and a language classification probability of each text unit in each candidate recognition text, and the score corresponding to any candidate recognition text is determined according to the text classification probability and the language classification probability of each text unit in the candidate recognition texts; and determining a target recognition text of the target speech fromthe plurality of candidate recognition texts according to the scores corresponding to the plurality of candidate recognition texts. The bilingual hybrid speech recognition method, device and equipmentand the storage medium in the application have a good recognition effect on bilingual hybrid speech, and the user experience is good.

The invention discloses an identification method of a GIS local discharge mode. The method comprises the following steps: identifying the local discharge mode by using a probability neural network, and performing dimension reduction processing on a high-dimension sample by using a linear judgment method; mapping a high-dimension sample to the optimal authentication vector space, classifying and compressing the sample dimension, wherein various mapped samples have the maximum inter-class distance and the minimum intra-class distance in the new sample space, and the identification effect can bewell improved; furthermore, the smooth factor of the probability neural network is optimized by using a genetic algorithm, the sample processing speed of the local discharge mode of the GIS is greatlyimproved, and the identification effect on the GIS local discharge mode by the probability neural network is greatly improved. The method disclosed by the invention has good promotion and use value.

The invention relates to a ceramic cultural relic intelligent identification system and method. Based on a multi-dimensional characteristic matrix, a deep learning model of ceramic cultural relic identification is built, so that the problem of insufficient modeling precision of a conventional method is solved. Moreover, the ceramic cultural relic identification is performed by fusing a multi-dimensional characteristic library of a genuine ceramic library and the deep learning model, so that the system running efficiency is improved and the identification effect is improved; and a self-feedbackupdate mechanism further ensures the identification precision of the identification system.

The invention provides a method and device for recognizing intelligent equipment users. The method for recognizing the intelligent equipment users comprises the steps of extracting network behavior data of a user and confirming a characteristic variable in the network behavior data; obtaining a first variable value and a second variable value, wherein the first variable value comprises a variable value of the characteristic variable of the to-be-detected equipment user and the second variable value comprises a variable value of the characteristic variable of a predetermined positive sample; calculating the distance information between the first variable value and the second variable value; and recognizing the intelligent equipment user according to the distance information. By adopting the method, the effect of recognizing the intelligent equipment users can be increased.

The invention discloses a method for recognizing a component feature. The method comprises the following steps: acquiring a component picture to be recognized; marking the component picture as one having the component feature to be recognized; selecting a feature extraction algorithm with the highest recognition rate for the component feature from a database, wherein the database records the recognition rates of various feature extraction algorithms for the component feature; extracting an algorithm feature value of the component picture according to the feature extraction algorithm; recognizing whether the component picture has the component feature or not according to the algorithm feature value. Moreover, the invention further discloses a system for recognizing the component feature. Through the adoption of the method and the system provided by the invention, the feature extraction algorithm with the highest recognition rate for the component feature can be selected adaptively, so that effectiveness and accuracy are improved for recognition each time.

The invention relates to a font recognition method and device based on a picture, electronic equipment and a storage medium. A to-be-processed picture is processed to obtain two kinds of pictures with different resolutions; the two pictures with different resolutions are input into two independent convolutional neural networks for feature extraction; pictures with different resolutions can be processed at the same time; the network also has good resolving power for pictures with lower resolutions; the method improves the recognition effect on the fonts in the blurred picture, finally improves the recognition accuracy of the blurred picture, obtains the feature information of the fonts in the picture through the pre-trained deep learning network, captures the local key information of the fonts in the blurred picture, and improves the recognition accuracy of the fonts in the blurred picture.

The invention discloses an automatic water sprinkling and dust suppression device for mining and relates to the technical field of mining environmental protection facilities. The device comprises three operation ways of touch control, sound control and automatic spraying. When a touch sensor is used, as long as pressure is imposed on the touch control sensor, a solenoid valve can be timely opened,so water sprinkling and dust suppression are performed. When it is inconvenient for a worker to carry out touch control in a certain environment, the touch control sensor is switched to a sound control sensor, so as long as sound of certain decibel is given out, the solenoid valve can be timely opened, and water sprinkling and dust suppression are performed. In case of much more coal dust, by useof sound control, health will be seriously affected, so the sound control can be switched to a dust concentration sensor at this moment. When dust concentration reaches a certain value, the solenoidvalve can be timely opened, so water sprinkling and dust suppression are performed. The whole automatic water sprinkling and dust suppression device can be used in multiple working scenes, and requirements in many situations are met.

The invention provides a text recognition method and device and a storage medium, and relates to the technical field of computers, and the method comprises the steps: carrying out character recognition of an original image containing a to-be-recognized character, and obtaining a character sequence; judging whether a recognition confidence coefficient of the characters in the character sequence issmaller than a preset confidence coefficient threshold value or not; if yes, determining that the character is an unrecognized character, and generating a to-be-recognized word vector based on the character sequence and the unrecognized character; obtaining similar word vectors matching the word vectors to be recognized in a word vector library, and determining unrecognized characters based on thesimilar word vectors. According to the method, the device and the storage medium disclosed by the invention, for words which cannot be identified by an optical character identification technology, the corresponding words are obtained by querying the word vector library, and the word vector library can be expanded and perfected, so that the identification effect of the optical character identification technology can be better improved, the working efficiency is improved, and the competitiveness is enhanced.

The invention discloses a laparoscopic surgery stage identification method and system based on dual-granularity time convolution. The method comprises the following steps: 1) constructing a laparoscopic surgery data set; 2) performing preliminary feature extraction on the picture sequence by using a dual-granularity time convolution module of the dual-granularity time convolution network, and outputting an initial prediction result of each frame of image; 3) correcting an initial prediction result output by the dual-granularity time convolution module by using a single-granularity time convolution module of the dual-granularity time convolution network; and 4) mapping a prediction result to an interval of (0, 1) to obtain a final operation stage identification result. The laparoscopic surgery stage recognition is realized by using the dual-granularity time convolution network, higher precision and better generalization ability under different backgrounds are realized, different types of surgery stages can be accurately detected, and the visual and time sequence information of the surgery video is utilized to realize the recognition of the laparoscopic surgery stage. The problem that in the deep learning field, operation stage categories can be recognized, but stage transition frames are difficult to distinguish accurately can be solved.

The embodiment of the invention provides a text recognition method and device, electronic equipment and a readable storage medium. The method relates to the field of artificial intelligence technology and cloud. The method comprises the steps of obtaining a to-be-recognized text image, wherein the to-be-recognized text image comprises a curved text area; extracting a text feature map of the to-be-recognized text image; determining a to-be-recognized text area in the to-be-recognized text image and area boundary information of the to-be-recognized text area based on the text feature map, wherein the to-be-recognized text area comprises a curved text area; correcting the to-be-recognized text region according to the region boundary information to obtain a corrected rectangular text region; and performing character recognition on the rectangular text area to obtain a text recognition result of the to-be-recognized text image. In the embodiment of the invention, the curved text region can be corrected into the rectangular text region, the background noise in the text region is reduced, and the recognition effect is improved.

The invention relates to a face recognition ATM with a good recognition effect, and the ATM comprises a housing, a shading mechanism and two cameras. The interior of the housing is provided with an adjusting mechanism and two sliding rails. The shading mechanism comprises a lifting assembly, a reciprocating component and two shading plates. The lifting assembly comprises a positioning rod, a driving unit, a telescopic frame, a connecting plate, two sliding blocks, two supporting plates and two transmission blocks. The adjusting mechanism comprises a first motor and two adjusting assemblies, and each adjusting assembly comprises a first gear, a rotating disc and a connecting rod. The driving unit comprises a second motor, a second gear and two transmission units. The ATM achieves the shading of strong light through the shading mechanism, and reduces the impact on the face recognition of the ATM from strong light, thereby improving the recognition effect of the ATM. The shooting angles of the cameras are adjusted through the adjusting mechanism to collect the face data at different angles for comparison, so that the face recognition precision of the ATM is higher.

The invention belongs to the technical field of computer vision, image processing and convolutional neural networks, and particularly relates to an optical character recognition method in a patent text scene. According to the method, a CNN and an LSTM are combined, the advantages of the CNN and the LSTM are achieved at the same time, the problem that the CNN is weak in sequence correlation processing is solved, and the defect that the LSTM is insufficient in image feature extraction is overcome. According to the present invention, a new loss function calculation method CTC is combined to solvethe problem that the sample data is difficult to align during the text recognition process in a manner without alignment.

The invention discloses a method and a device for pre-processing block Arab characters, belonging to the field of optical character recognition. The method comprises the following steps of: step 1, performing a gravity centre analysis on the character images obtained after a segmentation, and calculating the deviation degrees of the gravity centres of the character images; if the deviation degrees of the gravity centres of the character images are less than an appointed threshold, then turning to step 3; or else, turning to step 2; step 2, adjusting a confidence frame according to the obtained deviation degrees of the gravity centres, and determining the character images in the confidence frame; and step 3, loading the character images in the confidence frame on a recognition core to recognize. In the invention, a confidence frame is selected via a gravity centre analysis before a character recognition is performed, and the character images are adjusted by the confidence frame, then the character images at the gravity centre parts in the confidence frame are recognized, so that the character images cannot be influenced by an improper selection for segmentation points while segmenting characters, thereby avoiding the influence of elongation characters in Arabic and improving the recognition effect of the character recognition.

The invention provides a super-large field-of-view polarization camera based on a bionic curved surface compound eye. The problems that an existing polarization imaging system is complex in structure,large in installation and adjustment difficulty and small in imaging range, and position information, intensity information and target material information of a target cannot be accurately obtained are solved. The camera comprises a curved-surface compound eye unit, an optical relay image transferring subunit and a large-area-array plane detector which are arranged in sequence, wherein the curved-surface compound eye unit comprises a supporting shell, a sub-lens array and a polarizing film array; a plurality of mounting holes are formed in the surface of the supporting shell; the sub-lens array comprises a plurality of curved-surface sub-lenses, and the curved-surface sub-lenses are arranged from the center to the periphery in a hexagonal mode; the polarizing film array comprises a plurality of polarizing films, and the polarizing films are located on an incident light path of the curved-surface sub-lens; the optical relay image transferring subunit is used for guiding the curved-surface image information acquired by the curved-surface compound eye unit to the large-area-array plane detector; and the large-area-array plane detector is used for collecting image information in different polarization directions.

The invention discloses a pavement disease detection method and device based on an edge detection neural network, and the method comprises the steps: carrying out the pavement disease recognition of an inputted pavement image through employing a first edge detection neural network, and outputting a first disease probability matrix of the pavement image; performing pavement disease identification on the pavement image by using a second edge detection neural network, and outputting a second disease probability matrix of the pavement image; calculating a final disease probability matrix of the pavement image according to the first and second disease probability matrixes; recognizing pavement diseases from the pavement image according to the final disease probability matrix, wherein the firstedge detection neural network and the second edge detection neural network are obtained by pre-training pavement images with common pavement diseases and complex pavement diseases respectively. By applying the method and device, the pavement disease identification accuracy in the actual pavement acquisition image can be improved, the calculated amount in the identification process is reduced, andthe identification efficiency is improved.

The invention discloses an image recognizer with high intelligent degree, which comprises a base, a preformed groove with a first motor fixedly mounted at the top is formed in the middle end of an inner cavity of a shell, and an output shaft of the first motor is fixedly connected with the shell of which the upper end of the left side of the inner cavity is fixedly connected with a temperature sensor. The first motor is arranged; the purpose of driving the image collector to rotate in the horizontal direction can be achieved; a second motor is arranged, the inclination angle of the image collector can be driven to change; a GPS positioner is arranged, the position information of the recognizer can be detected; a wireless signal emitter is arranged; a background worker can conveniently carry out remote control on the recognizer; a temperature sensor and an exhaust fan are arranged, the purpose of automatic heat dissipation can be achieved, a photoresistor and a light supplementing lampare arranged, the recognition effect of the recognizer can be improved, an infrared tracker is arranged, and the purpose of tracking movement of a target object can be achieved by controlling the first motor and the second motor.

The present invention relates to a preparation method of a polyethyleneimine / chiral peptide modified glassy carbon electrode and an application thereof in tryptophan enantiomer selective recognition. The method comprises following steps: preparing a polyethyleneimine modified glassy carbon electrode, and immersing the polyethyleneimine modified glassy carbon electrode in a chiral peptide (BGAc) solution for further modification, to obtain the polyethyleneimine / chiral peptide modified glassy carbon electrode, for electrochemical identification of a tryptophan enantiomer. The method has beneficial effects that: the method has simple preparation process of the polyethyleneimine / chiral peptide modified electrode, is environmentally friendly, low in costs and obtains the modified electrode having obvious recognition effect on the tryptophan enantiomer.

The invention discloses an electrocardio identity recognition method and system based on local constraint non-negative matrix factorization. The method comprises the steps that an electrocardio signalto be recognized is obtained; heart beat division is carried out on the to-be-recognized electrocardiosignals to obtain a plurality of single-cycle electrocardiosignals; the method also includes projecting each single-cycle electrocardiosignal by using a pseudo-inverse matrix of a pre-acquired basis matrix, and projecting the electrocardiosignal from a high-dimensional space to a low-dimensionalspace to obtain feature representation after dimension reduction, wherein the pre-acquired pseudo-inverse matrix of the basis matrix is acquired by processing the electrocardiosignal training data setthrough a non-negative matrix decomposition mode based on local constraints; and inputting the feature representation after dimension reduction into the trained classification model to obtain an identity recognition result corresponding to the current to-be-recognized electrocardiosignal.

The present application discloses a human body image recognition method and apparatus. The method comprises: by combining the depth value with the gradient information, obtaining a video image containing a tracking object through capturing; reading a depth image of the video image; determining a depth value d1(x) of each pixel point x in the depth image; determining gradient information f [theta](1, x) around each pixel point x according to the depth value d1(x) of each pixel point x; and determining a human body depth image in the depth image according to the gradient information of each pixel point x. According to the technical scheme of the present application, the purpose of accurately recognizing the depth image of the human body is achieved, rapid recognition of the bone is realized,the problems such as the target is lost and is difficult to be recaptured during the recognition process are avoided, and the recognition effect can be effectively improved.

The invention discloses a face authentication method, which comprises the steps of receiving a face authentication instruction sent by an upper computer, and starting an infrared camera according to the face authentication instruction; controlling an infrared stripe projector corresponding to the infrared camera according to the configuration instruction of the upper computer and / or the illumination condition of the environment where the infrared camera is located, and an infrared picture frame containing a human face pattern is acquired through the infrared camera; and extracting human face features in the infrared picture frame, performing human face authentication based on the human face features, and feeding back an authentication result to the upper computer. The technical problems that an existing face authentication scheme is prone to interference in a complex illumination environment, the recognition effect is poor, and the control complexity is high are solved. The beneficialeffects of improving the recognition effect and reducing the control complexity are achieved.

The invention relates to a preposition phrase identification method based on a multi-model advantage complementation strategy. The method includes the following steps of classifying preposition phrases, classifying prepositions according to the characters of contexts, and conducting analysis and summary on the preposition phrases according to the sentence constituents and position characters of the preposition phrases to obtain position characters corresponding to different types; selecting different position character combinations according to different types, adopting a conditional random field model to identify the preposition phrases, and selecting training corpus characters; according to the selected training corpus characters, training multiple preposition phrase identification models, combining the identification results of each preposition in sentences, and obtaining a final result. According to the characters of the prepositions and the prepositional phases, a prepositional phase identification strategy based on multi-model advantage complementation is put forwards; through a 10-fold cross validation method and a contrast experiment, the effectiveness and applicability ofthe method are proved, and the identification effect on the preposition phrases is further improved.

The invention provides a face detection model training method and device, an identification method and device, a medium and equipment, and relates to the technical field of artificial intelligence. The face detection model training method is obtained through training based on sample images of the front face and the non-front face of a target animal, and the front face and the non-front face of the target animal in a to-be-detected image can be comprehensively recognized. Therefore, the technical problems that in the prior art, an existing pet face detection model is good in pet front face recognition effect and not ideal in non-front-face image recognition are solved, and the pet face recognition effect is improved from the two dimensions of comprehensiveness and accuracy of pet face recognition.

The invention discloses a vehicle logo recognition method and system based on feature enhancement, and belongs to the technical field of image recognition, and the method comprises the steps: carryingout the preprocessing of a vehicle logo image, obtaining a processed image, and extracting the overall features and self-symmetric similarity features of the processed image; processing the processedimage to obtain a saliency map, obtaining local saliency blocks according to the saliency map, and extracting the local saliency blocks to obtain local saliency block features; fusing the features toobtain final features; and recognizing the final features to obtain a vehicle logo recognition result. According to the system, a preprocessing unit and a feature fusion unit are connected with a feature extraction unit, and the feature fusion unit is connected with an identification unit. The method aims to overcome the defect that in the prior art, the vehicle logo recognition accuracy is low,the vehicle logo recognition accuracy can be improved, efficient vehicle logo recognition can be further achieved under the condition that the number of training samples is small, and a better vehiclelogo recognition effect is achieved.