43results about How to "Solve technical problems with low recognition accuracy" patented technology

The invention discloses an obstacle recognition method and device, a storage medium and an electronic device. The obstacle recognition method comprises the steps: in the automatic driving process of atarget vehicle, obtaining point cloud data, wherein the point cloud data are obtained by scanning the driving environment of the target vehicle through a sensor on the target vehicle; converting thepoint cloud data into a first image for representing a driving environment; and identifying a first object in the driving environment from the first image through a first neural network model, the first neural network model being obtained by training a second neural network model by using a second image marked with a second object, and the first object and the second object being movable obstacles. According to the invention, the technical problem of low obstacle recognition accuracy in the prior art is solved.

The invention provides a motion recognition method and a device based on human body posture, the methods include: the skeletal data are obtained by the improved clipping filtering algorithm, Angle features are obtained by improved angle calculation method, and the angle features of the classified classes are trained based on logical regression, and the trained classifier is obtained, and then therecognition result of static posture of human body is obtained through the classifier, and finally the human body motion is recognized by reverse order method according to the recognition result of static posture. The invention realizes the technical effects of improving the identification speed and the identification accuracy.

The invention discloses a non-invasive load identification method and device. The method comprises the steps of finding out a transient stage of a time sequence according to an obtained time sequenceat a power supply inlet, wherein the time sequence comprises an active power time sequence, a reactive power time sequence and a current time sequence; extracting load identification characteristic quantities at the transient stage according to the electric energy data at the power supply inlet, wherein the electric energy data include voltage, current and power, and the load identification characteristic quantities include steady-state power increment, active power characteristics, harmonic current increment and load time characteristics; and determining a load type according to the load identification characteristic quantities and a preset load library, wherein the preset load library has a corresponding relationship between the load identification characteristic quantities and the loadtype. According to the invention, the technical problems of limited steady-state characteristics and relatively low identification accuracy under the condition of complex actual working conditions ofexisting non-intrusive load identification are solved.

The invention discloses an image recognition method and device. The method comprises the steps that at least one face recognition algorithm is acquired; the corresponding weight is assigned to at least one face recognition algorithm, and the weight value of at least one face recognition algorithm is determined; the similarity value of at least one face recognition algorithm is determined by comparing the initial recognition result of preset image recognition of at least one face recognition algorithm and the preset recognition result; and fusion processing is performed on the initial recognition result of at least one face recognition algorithm according to the weight value and the similarity value of at least one face recognition algorithm so as to obtain the recognition result of presetimage recognition. The technical problem of low recognition accuracy in case of performing recognition of the face image by using multiple recognition algorithms can be solved.

The invention discloses a human body action reconstruction system and method based on multi-view vision and an action training system. The method comprises: capturing an action image of the target human body based on the at least two monocular cameras after correction; performing human body two-dimensional articulation point identification on the action image on a two-dimensional image; and reconstructing the three-dimensional motion of the target human body on the basis of the human body two-dimensional joint points. A user does not need to wear a specially-made optical capture garment or assemble a corresponding sensor, and the technical problem that an existing depth camera human body three-dimensional motion recognition method is high in application scene requirement and low in recognition accuracy in a complex application scene is solved.

The invention discloses a terminal device and a terminal device identification method and system. The method comprises the steps that the hardware information of the terminal device and basic service interface identification information and / or basic service interface attribute information are acquired, and match an information base which is corresponding to each terminal device and is stored by a cloud server; and if the acquired hardware information of the terminal device and the acquired basic service interface identification information and / or basic service interface attribute information do not match the information base which is corresponding to any terminal device and is stored by the cloud server, a user is prompted. According to the invention, the identification accuracy of the terminal device can be improved.

The invention provides a malicious information identification method, device and system. The method comprises that real-time session information of a session group is obtained; a malicious information model is used to determine malicious information in the real-time session information and further to obtain suspicious malicious information; a malicious information determination request for the suspicious malicious information is initiated for users of the session group, and determination results input by the users are received; and malicious information is identified from the suspicious malicious information according to the determination results and a preset determination method. According to the malicious information identification method, device and system, the malicious information is determined in the suspicious malicious information by the users of the session group, and the accuracy of malicious information identification is improved.

The invention discloses a fighting behavior recognition method and device, a storage medium and an electronic device. The method comprises the steps: calculating optical flow data according to a (t-1)-th frame of image and a t-th frame of image in a target video, and determining whether candidate areas of fighting behaviors exist in the t-th frame of image or not through head and shoulder detection tracking data, wherein t is a positive integer greater than 1; under the condition that it is determined that the candidate area exists in the t-th frame of image, inputting the image sequence of the candidate area in the t-th frame of image and the image sequence of the same area of the L frames of images before the t-th frame of image into a target neural network, and outputting a recognitionresult; under the condition that the recognition result determines that the fighting behavior exists in the candidate area, transmitting early warning information, carrying out the pre-detection operation on fighting behaviors in a video, extracting candidate areas, and then carrying out the secondary filtering on the candidate areas to obtain final fighting areas, and feeding back alarm information, so the technical problem of relatively low accuracy of fighting behavior identification in the prior art is solved.

The invention discloses an image recognition method and system and a data processing method. The image recognition method comprises the following steps: acquiring a table image; processing the table image to obtain table data corresponding to the table image, wherein the table data comprises text information in cells in the table image and the structural relation between the cells; processing the table data by using a table recognition model to obtain keywords and structure types of the keywords in the table image, the table recognition model being used for obtaining the table data, inputting the table data to the coding network to obtain vectors corresponding to the table data, inputting the vectors to the feature extraction network to obtain features of the table data, and inputting the features of the table data into a classification network to obtain keywords and structure types of the keywords. According to the method and the system, the technical problem of relatively low recognition accuracy of keywords in table images in related arts is solved.

The invention discloses an image recognition module training method and device. The method comprises the steps of performing feature extraction on a specified medical image based on a convolutional neural network to generate a first image feature; processing the specified medical image based on an attention mechanism to obtain a second image feature and an extraction region; obtaining an extraction region image corresponding to the extraction region, namely, generating an extraction region image with a better training effect by secondary windowing in a progressive focusing mode; performing feature extraction on the extracted region image based on a convolutional neural network to obtain a third image feature of the extracted region image; and fusing the first image feature, the second image feature and the third image feature to obtain a fused image feature. According to the method, on the basis of global image learning, re-windowing can be carried out on the region of interest, input of the neural network is carried out in a progressive focusing mode, and the purpose of training the image recognition module through the multi-windowing image processing method based on the progressive focusing mechanism is achieved.

The invention discloses a method and system for target recognition by a laser composite imaging. The method comprises the following steps of: obtaining a distance image, an intensity image, an azimuthangle, and a pitch angle of each point of a target by a laser composite imaging; reconstructing a three-dimensional point cloud data from the distance image, the intensity image, the azimuth angle, and the pitch angle to obtain the three-dimensional point cloud data, wherein the three-dimensional point cloud data comprises amplitude, height, shape, spectrum, and scatter information of the target;filtering the three-dimensional point cloud data, and estimating a size, the shape, the height, the distance, the scattering, and the physical characteristic parameter of the surface layer of the target by using a multi-dimensional spatial transformation; identifying in a classifier according to an identification function determined by a large number of training samples, combined with a feature database of known typical targets. The method uses the distance image, the intensity image, the azimuth angle, and the pitch angle of the target as the basis of the target recognition to more accurately identify the target, and solves the technical problem that the recognition accuracy of the existing target recognition technology is not high.

The invention discloses a pig variety identification method, a pig variety identification device, and a computer readable storage medium. The method comprises the steps of obtaining a to-be-identifiedpig image; inputting the to-be-identified pig image into n preset convolutional neural network models with target functions as follows for feature extraction, and outputting n groups of feature vectors; inputting the output n groups of feature vectors into n preset classifiers, outputting n first classification results, and performing weighted voting on the output n first classification results to obtain an identification result; wherein losscan is a cross entropy loss function, wi is the ith weight, N is the total number of samples, alpha is a regularization term coefficient, r is the sum oftwo regularization terms and the proportion of the two regularization terms, and r takes n different values between [0, 1] and corresponds to n different models. According to the pig variety identification method provided by the invention, the technical problem that the identification accuracy of pig identification based on a traditional machine vision image identification method is not high is solved.

The invention relates to a gesture recognition method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining a to-be-recognized video; performing human body detection on the video to obtain the number of first objects included in the video; and recognizing the gesture in the video according to the gesture recognition mode corresponding to the number of the first objects to obtain a gesture recognition result. According to the embodiment of the invention, the gesture recognition effect of a multi-person scene can be improved.

The invention discloses a text recognition method, device and system. The method comprises the steps of obtaining text data; obtaining a word vector corresponding to the text data; and recognizing theword vector by using a recognition model to obtain a recognition result of the text data, the recognition model being used for recognizing whether illegal content exists in the text data, and the recognition model being obtained through adversarial training. According to the invention, the technical problem of low recognition accuracy of a text recognition method in the prior art is solved.

The invention provides a rainstorm identification method and device based on a multi-model fusion convolutional network, and relates to the technical field of rainstorm early warning. The method comprises the steps: obtaining sample Doppler radar base data of a to-be-identified region, and determining the attribute information of a storm body contained in the sample Doppler radar base data based on the sample Doppler radar base data; obtaining observation data of the to-be-identified area sent by the ground observation station, and constructing a data set based on the observation data and the attribute information; inputting the data set into a multi-model fusion convolutional network, and training and optimizing the multi-model fusion convolutional network to obtain a rainstorm identification model; and obtaining the current Doppler radar base data of the to-be-identified area, and and determining whether the rainstorm occurs in the to-be-identified area by using the current Doppler radar base data and the rainstorm identification model. Thus, the technical problem of low identification accuracy of an existing rainstorm identification method is solved.

The invention discloses a video content recognition method and device, a storage medium and electronic equipment. The method comprises the following steps: fusing multi-dimensional features extracted from video content of an object video to be identified, so as to obtain a multi-modal video feature vector corresponding to the object video; according to the multi-modal video feature vector and a first weight set determined based on the first recognition tag, obtaining a first grade recognition parameter set; and according to the multi-modal video feature vector and a second weight set determined based on a second recognition tag, obtaining a second grade recognition parameter set, and according to the first grade recognition parameter set and the second grade recognition parameter set, determining a target content quality grade matched with the object video. According to the invention, the technical problem of low video content recognition accuracy is solved.

The invention provides a business card identification method and device. The method comprises the following steps of: obtaining a candidate business card area in a to-be-identified image; identifyinga plurality of candidate outline key points in the candidate business card area according to a preset image identification algorithm, calculating a plurality of candidate contour key points accordingto a convex hull algorithm to obtain a plurality of candidate edges, selecting a plurality of groups of candidate edges meeting a preset condition from the plurality of candidate edges, constructing acandidate quadrangle corresponding to each group of candidate edges; calculating the confidence coefficient of each candidate quadrangle according to a preset algorithm, determining a target candidate quadrangle according to the confidence coefficient, determining the outline key points of the candidate business card areas according to the target candidate quadrangle, determining the target business card areas in the candidate business card areas according to the outline key points, and identifying the business card information in the target business card areas. According to the method, the technical problem that the identification accuracy is low when the business card image with the complex background is identified in the prior art is solved, and the business card identification precision and accuracy are improved.

The invention discloses a data identification method and device, a storage medium and an electronic device. The method comprises the following steps: obtaining a display record of media resources provided by resource providing equipment which are requested to be displayed; according to the display record, determining a use record for displaying the target media resource provided by the resource providing equipment under the condition that the third-party equipment does not obtain the use authorization of the resource providing equipment; obtaining feature data matched with the target media resource according to the use record, the feature data being used for indicating display features of the target media resource in the current period; and identifying the hot media resources with the feature data meeting a predetermined condition from the target media resources. The technical problem that an existing data identification method is low in identification accuracy is solved.

The invention relates to a gesture track recognition method and device. The method comprises the following steps: coordinates of gesture key points in the video are detected through a target key pointdetection model; a coordinate sequence of a first length is obtained, wherein the coordinate sequence of the first length is used for indicating a gesture track of a hand displayed in the video, thetarget key point detection model is obtained by training an initial key point detection model by using a key point detection sample, and the key point detection sample is a sample image marked with coordinates corresponding to gesture key points; the coordinate sequence of the first length is matched with a template coordinate sequence of a second length, and the template coordinate sequence of the second length is used for indicating a target gesture track; and under the condition that the coordinate sequence of the first length is successfully matched with the template coordinate sequence ofthe second length, it is determined that a target gesture track is recognized in the video. The technical problem that the gesture track recognition accuracy is low is solved.

The invention discloses a session information transmission method and device, a storage medium and an electronic device. The method comprises the steps that of detecting target operation executed on atarget client side is detected, andwhrein the target operation is used for establishing a target session between the target client side and a target server corresponding to the target client side through a target router; obtaining session information of the target session; and sending the session information to the target router, with the session information being used for indicating the target router to preferentially forward the data carrying the session information. According to the invention, the technical problem of low identification accuracy of the service traffic in the prior art is solved.

The invention provides a scene recognition method and device based on indoor opportunity signal enhancement. By using a convolution neural network trained by transfer learning to obtain the scene-level image features of indoor localization points, and describing the positioning characteristics of the positioning point through the positioning information of the positioning point and the scene bottom map and the positioning error and then using the location feature to expand and enhance the image features, and using the depth learning method to fuse the image features and the location features to predict the scene recognition, a technical effect of achieving higher accuracy of scene recognition is realized.

The invention relates to a face recognition system. The face recognition system comprises a fixing unit, a collecting unit, an image preprocessing unit, an image processing unit, a matching unit and an analyzing unit, wherein the fixing unit is used for fixing the head of a user; the collecting unit is used for collecting face images of the user fixed in the fixing unit; the image preprocessing unit is used for preprocessing the face images collected by the collecting unit; the image processing unit is used for processing images preprocessed by the image preprocessing unit, so as to obatin corresponding image information; the matching unit is used for matching the image information with image information in a system database, so as to obtain matching results; and the analyzing unit is used for analyzing the matching results, so as to obtain recognition results. The face recognition system effectively solves a technical problem that recognition accuracy is lower in the existing face recognition system, and further achieves a technical effect of accurately performing face recognition.

The invention discloses a network flow identification method and device. The method comprises the following steps: acquiring a data packet of target network traffic to be identified; enhancing the data packet through wavelet transform and histogram equalization; the enhanced data packet is input into a recognition model, the type of the target network traffic is output by the recognition model, the recognition model comprises a residual module and an attention module, the recognition model is formed by training multiple sets of training data, and each set of training data comprises the input data packet of the network traffic and the type of the network traffic. According to the invention, the technical problem of low identification accuracy of the traditional network flow identification method in the related technology is solved.

The invention discloses a voice recognition method and system. The method comprises the following steps: acquiring a sound signal; performing feature extraction on the sound signal to obtain acousticfeature information of the sound signal; and identifying the acoustic characteristic information by using an acoustic model and a language model to obtain an identification result of the sound signal,wherein the acoustic model comprises an LC-BLSTM model and a DFSMN model. According to the invention, the technical problem of low recognition accuracy of a voice recognition method in a complex environment in the prior art is solved.

The invention provides a distorted two-dimensional code recognition method and device, a storage medium and an electronic device. The method comprises the following steps: determining a first group ofkey points in a target template pattern as a group of control points of thin plate spline fitting operation; obtaining a group of target values of the thin plate spline fitting operation according tothe position coordinates of the group of control points in the two-dimensional code image; performing thin plate spline fitting operation according to the group of control points and the group of target values to obtain a target value of a target parameter in a thin plate spline interpolation function; determining mapping coordinates of a central point of a module in a coding region in the targettemplate pattern in the two-dimensional code image according to the thin plate spline interpolation function to obtain a group of mapping coordinates; and identifying the two-dimensional code image according to the group of mapping coordinates. According to the method and the device, the technical problem of relatively low recognition accuracy of the distorted two-dimensional code in an uncertainform in related technologies is solved.

The invention discloses an object matching method, a model training method, a product matching method and a storage medium. The method comprises the steps of obtaining an object search request and description information of a target object; and performing semantic matching on the plurality of retrieval keywords and the description information by utilizing a semantic matching model to obtain a target matching result of the object search request and the target object. According to the method and the device, the anti-interference capability of the model on the redundant information of the keyword stacked text is improved by means of adversarial training in the training process of the model, and moreover, the distinguishing of the contrast learning enhancement model on positive and negative samples is fused in the adversarial training process, so that the characterization capability of the object search request and the description information of the target object is improved; therefore, the recognition capability of the semantic matching model on the fine semantics can be improved, the robustness is improved, the technical effect of improving the model recognition accuracy is further achieved, and the technical problem that in the prior art, the recognition accuracy of the semantic matching model is low is solved.

The invention discloses an advertisement position anomaly identification method and device. The method comprises the steps of obtaining a device unique identifier of a target device accessing a targetadvertisement on an advertisement position and an Internet protocol address used by the target device; under the condition that the device unique identifier does not belong to a device identifier database or the Internet protocol address belongs to a blacklist address database, increasing the abnormal access number of the target advertisement by 1, wherein the device identifier database records the identifier identifier of a real identifier, and the blacklist address database records the abnormal Internet protocol address; obtaining the total access number of the advertisement position and obtaining the abnormal access number of the target advertisement; and under the condition that the ratio of the abnormal access number of the target advertisement to the total access number is greater than a first threshold, determining that the advertisement position where the target advertisement is located is abnormal. According to the method, the technical problem of low identification accuracyof whether the advertisement position is abnormal is solved.

The invention discloses a dynamic target rapid sensing method and system based on deep learning. The method comprises the following steps: carrying out environment sensing based on sensing system joint calibration, and collecting data from different data sources; based on the collected data, scene acquisition is carried out, and map data and pose information are acquired; performing normalized data preprocessing of multivariate features based on the obtained map data and pose information, and performing feature matching on the data after the normalized data preprocessing through superpixel segmentation and semantic analysis; and based on data obtained by feature matching, sensing the dynamic target by using a sparse convolutional network and motion estimation fusion. According to the invention, the technical problem of low dynamic target identification accuracy in the prior art is solved.

The invention discloses a virus file identification method and device, a storage medium and an electronic device. The method comprises the steps that a sample record is acquired, and characteristic values of a program file which initiates access on an intelligent terminal on multiple characteristic dimensions are recorded in the sample record; sample records are clustered to obtain a sample set, and sample records of the same type are stored in the sample set; searching a target sample record with the same feature value on a target feature dimension in the sample set, the plurality of featuredimensions including the target feature dimension; and under the condition that the similarity between the program file recorded in the target sample record and the known virus file of the target typereaches a first threshold, taking the program file recorded in the target sample record as the virus file of the target type. According to the invention, the technical problem of relatively low virusidentification accuracy in related technologies is solved.

The invention relates to an iteration method and device for a recognition model, and the method comprises the steps: obtaining a first training sample which comprises data and annotation information which have a corresponding relation and are wrongly recognized by a first recognition model, wherein the first recognition model is used for executing a data recognition task; training an initial model corresponding to the first recognition model by using the first training sample to obtain a second recognition model; adding the first training sample to a second training sample so as to obtain a third training sample, wherein the second training sample is used for training a sample of an initial model corresponding to the first recognition model; and training the second recognition model by using a third training sample to obtain a third recognition model which is used for continuing to execute the data recognition task. According to the method and the device, the technical problem that the model recognition accuracy obtained after iteration is performed on the recognition model is relatively low is solved.