118results about How to "Solve the low recognition accuracy" patented technology

The invention provides a method of human face vivo detection, which belongs to the field of biometrics recognition and solves the problem that the human face vivo detection method in the prior art is low in recognition accuracy. The method comprises steps: a first image and a second image containing a to-be-detected human face and with an acquisition time interval smaller than a preset time duration are acquired, wherein the first image is an image acquired in an environment of starting an active light source, and the second image is an image acquired in an environment of closing the active light source; then, to-be-detected areas in the first image and the second image are determined respectively, and a differential image for the to-be-detected areas in the first image and the second image is acquired; to-be-recognized features are extracted from the differential image; and finally, the to-be-recognized features are inputted to a preset classifier, and human face vivo detection is carried out, wherein the to-be-detected area uses the to-be-detected human face as a center, and comprises the to-be-detected human face and a partial background around the to-be-detected human face.

The invention discloses a method for fusing radio frequency identification and vehicle license plate recognition. The method comprises the step of license plate recognition and the step of license plate fusion. According to the license plate recognition, a radio frequency identification system conducts collection and recognition on motor vehicle license plate information through a reader-writer, wherein the motor vehicle license plate information is stored in vehicle-mounted radio frequency identification tags; a vehicle license plate recognition system conducts information collection and recognition on motor vehicle license plates through a camera device, and respectively records license plate information which is successfully recognized and time information at the moment of successful recognition. According to the license plate fusion step, a table of similarity among characters of the motor vehicle license plates is built, fusion recognition is carried out on the motor vehicle license plate information obtained from the radio frequency identification system and the vehicle license plate recognition system based on the table of similarity, and a final recognition result is output.

The invention belongs to information retrieval. And its database structure technology field, discloses a city planning dynamic monitoring system and method based on high-score remote sensing and unmanned aerial vehicle, which utilizes high-score satellite remote sensing data to improve the time frequency and spatial position precision of the change patch identification of the city planning area; UAV image acquisition technology and tilt photography modeling technology are used to improve the accuracy of land use types and features recognition and provide better image data source for urban planning monitoring. According to the accurate extraction technology of change patch, the change patch is connected with the planning approval system, and the matching degree between the planning contentand the construction content of the change patch is automatically identified by the technology of spatial superposition and attribute analysis, and the illegal construction behaviors such as construction without approval and construction not according to the planning approval content are automatically identified by the system.

The invention discloses a machine learning-based vehicle abnormal trajectory real-time recognition method and belongs to the field of vehicle trajectory anomaly recognition. The method includes the following steps that: collected data are cleaned, so that complete, non-repetitive, abnormal value-free training data are obtained; an unsupervised isolated forest method and training data are used to perform model training, so that an anomaly detection model is obtained; the anomaly detection model is put into a flow calculation engine for real-time prediction, and a prediction result is sent to avehicle owner; and the model is automatically updated and corrected according to the feedback information of the vehicle owner, and the updated model is put into the flow calculation engine for real-time prediction, and a prediction result is sent to the vehicle owner. According to the method of the invention, the vehicle information is periodically collected; the unsupervised isolated forest algorithm is adopted; real-time prediction analysis is performed on vehicle trajectories in the flow calculation engine; the probability value of the abnormal behavior of a vehicle is rendered; the modelis periodically adjusted according to the feedback data given by the vehicle user; the dynamic update of the model is realized; and the recognition accuracy of the model is improved.

The invention relates to an intelligent recognition method based on movement change. Through the identification and analysis of motion video images and based on the short video content input by the user, the sports category is identified based on the action semantics and the clothing coverage, after the error correction judgment, according to the between class distance, the correctness of the category is determined, then specific sports under the sports category are identified according to the characteristics of the sports motion, and then the method focuses on the facial expressions of the athlete for face identification and outputs the related sports information to the user. The method solves the problem that the accuracy of the motion recognition is not high, and solves the technical problems in the prior art that the relevant information of the sports category and the athlete cannot be acquired in real time, and effectively improves the timeliness and accuracy of the user obtainingthe motion information.

The invention is applicable to the technical field of computer vision and image processing, and provides a lane line detection method and device, terminal equipment and a readable storage medium. Inputting the road image into a trained neural network model for processing, and outputting a detection result of a lane line in the road image of the current scene; wherein the trained neural network model is obtained by training according to sample images in a training set and a semantic segmentation model, and the sample images in the training set comprise collected road images of a plurality of scenes and marked images corresponding to the road images of the plurality of scenes. According to the method and the device, the problems that most of deep learning models used for lane recognition atpresent are relatively large in calculation amount, relatively complex in model and unfavorable for meeting the requirement on real-time performance in an actual application scene of an automatic driving task can be solved.

The embodiment of the invention discloses an information interaction method, a device, a computer device and a storage medium. The method comprises the following steps of: generating a dialogue scenedatabase according to the actual conversation between users under a real scene; Acquiring the interaction problem of user input, and searching in the dialogue scene database according to the interaction problem, acquiring the scene problem that matches the interaction problem best as the target scene problem; The answer corresponding to the question of the target scene is obtained in the dialoguescene library and sent to the user. The technical proposal provided by the embodiment of the invention solves the problem of low identification accuracy caused by matching corresponding answers through a human-maintained semantic template in the prior art, establishes a dialogue scene database according to a dialogue actually occurring under a real scene, is closer to the interactive problem inputby a user, and improves the accuracy rate of problem matching.

The embodiment of the invention, which relates to the technical field of computer application, provides an object control method and device. The object control method comprises: carrying out credit scoring on historical data of a monitored object; predicting an abnormal probability of the monitored object; and on the basis of the credit scoring result and the abnormal probability, carrying out control processing on the monitored object. On the basis of credit scoring on the historical data of the monitored object, the abnormal probability of the monitored object is predicted and thus a cheating user can be identified effectively and accurately, so that a problem of low accuracy of cheating user identification in the prior art is solved.

The invention provides an identification method for reservoir fluid properties. The identification method comprises following steps: analyzing sandstone samples of a target reservoir segment to be identified in a single well in order to first carbon isotope data forming sandstone samples; processing first carbon isotope data in order to obtain second carbon isotope data in accord with the standard of carbon content; setting up an identification chart for reservoir fluid properties according to second carbon isotope data; and identifying fluid properties of the target reservoir segment of tight sandstone to be identified in the single well based on the identification chart for reservoir fluid properties. The technical scheme of the identification method for reservoir fluid properties helps to significantly improve accuracy rate for identification of fluid properties of the target reservoir segment to be identified and deepen the understanding of gas reservoir so that great application value is obtained.

The invention provides a diptera insect recognition method based on a deep convolutional neural network, and the method comprises the following steps: collecting a diptera insect image, and making a data set of the diptera insect image; performing data enhancement on the data set; constructing an improved RetinaNet target detection model; setting training parameters, and training the RetinaNet target detection model through the data set; and classifying and positioning the test set images of the diptera insects based on the trained target detection model. According to the diptera insect identification method based on a deep convolutional neural network, the RetinaNet target detection model is adopted, meanwhile, an improved convolution block attention module is added, a feature pyramid network is improved, a large amount of manpower and material resources do not need to be consumed, the problem of dependence on manual design features can also be solved, and the image acquisition methodis easy to operate.

The invention relates to a human behaviour recognition method, a human behaviour recognition device, a storage medium and electronic equipment. The method comprises the following steps: acquiring behaviour data which reflect a current human behaviour and are acquired through a sensing device; then, performing feature extracting on the behaviour data to generate a feature map of the behaviour data;next, acquiring a bag-of-visual-word model of the feature map; finally, determining the type of the current human behaviour by utilizing a classifier which is trained in advance and the bag-of-visual-word model of the feature map. According to the human behaviour recognition method, the human behaviour recognition device, the storage medium and the electronic equipment, the problem of low recognition accuracy caused by the fact that the feature of a human behaviour action cannot be fully expressed in the prior art can be solved, and human abnormal behaviour recognition accuracy can be improved.

The invention discloses a network traffic identification method, device and equipment and a computer storage medium, and the method comprises the following steps: collecting and labeling network traffic in different graph modes; preprocessing the acquired network traffic, and extracting feature information of each network session; generating a data flow diagram based on the feature information; training a graph neural network by utilizing the data of the data flow graph to generate a network flow identification model; converting unknown flow into data of a data flow graph and inputting the data into the network flow identification model, wherein the network flow identification model matches the data flow graph of the unknown flow with a graph mode learned by the network flow identificationmodel; and when the matching degree is greater than a preset threshold, judging a graph mode corresponding to the unknown flow, thereby determining a network application corresponding to the unknownflow. According to the invention, the problem of low accuracy of network traffic identification is solved, and the graph pattern of the network traffic and the corresponding network application are judged by using the graph neural network.

The embodiment of the invention provides a terminal state identification method and device and a storage medium. The method comprises the steps of acquiring a call request sent by a calling terminal to a called terminal, wherein the call request comprises a terminal identifier of the called terminal; obtaining audio data sent by the called terminal for the call request according to the terminal identifier; performing feature extraction on the audio data to obtain audio features; inputting the audio features into a preset state recognition model, and recognizing the terminal state of the called terminal, wherein the state recognition model is obtained by training a preset training model through an audio characteristic sample corresponding to the audio data sample and a terminal state mark sample corresponding to the audio data sample. Thus, the audio features are extracted from the audio data of the called terminal, the terminal state of the called terminal is determined by using the audio features, and the process of converting the audio data into the text data in the prior art does not need to be executed, so that the problem of low recognition accuracy in the prior art is solved.

The invention discloses a network traffic classification method. The method comprises the following steps: receiving an input real data set and a to-be-tested data set; performing data processing on the real data set to obtain a training data set; training the preset model by using the training data set to obtain a network traffic classification model; and classifying the to-be-tested data set byusing the network traffic classification model to obtain a classification result. According to the invention, data processing is carried out on the received real data set; according to the technical scheme, the training data set is acquired through the network traffic classification model, so that the training data set has all characteristics of the real data set, the network traffic classification model trained by using the training data set can classify the to-be-tested data set according to the characteristics of the real data, the acquired classification result is more accurate, and the network traffic identification precision is improved. The invention also provides a network traffic classification system and device, and a computer readable storage medium, which have the above beneficial effects.

The invention provides a method for improving speech recognition accuracy on the basis of voice leading end noise elimination for large-scale isolate word speech recognition. The method solves the problem that the recognition accuracy rate is low because of speech endpoint detection errors in the MFCC extraction process with noise existing. CASA is used for the leading end of speech recognition, and compared with traditional denoising methods such as noise reduction and speech enhancement, noise can be effectively separated from the speech with the noise by simulating an auditory nerve system of a human ear. The method for improving the speech recognition accuracy on the basis of the voice leading end noise elimination is adopted for recognizing 10240 pieces of speech with noise, the accuracy rate of recognition is improved to 95.5 percent from 83 percent compared with the method without leading end noise processing.

The embodiment of the invention discloses a speaker recognition method and device and electronic equipment. The method comprises the following steps: separating an audio file and an image file from a to-be-identified video file; performing voice segmentation on the audio file to obtain start-stop time information and speaker identification information corresponding to at least one voice segment, and performing face recognition on the image file to obtain face recognition results corresponding to different time; performing alignment processing on the speaker identification information and the face recognition result in time, and determining a speaker corresponding to at least one voice segment. According to the scheme, the accuracy of speaker recognition can be improved.

The present invention discloses a parameter identification method for a multi-nonlinear parameter coupling system and an identification device thereof. The parameter identification method for the multi-nonlinear parameter coupling system obtains an estimated nonlinear parameter values, namely a nonlinear parameter identification value through processing such as the Fourier transform, the inverse Fourier transform, the equal division processing and acquaintance assessment, etc. The parameter identification method for the multi-nonlinear parameter coupling system and the identification device thereof can realize the simultaneous identification of all the nonlinear parameter values included by nonlinearity to be identified so as to fundamentally solve the problem that the parameter identification precision is low caused by the transferring errors generated in the parameter identification process of the multi-nonlinear parameter coupling system. The parameter identification method for the multi-nonlinear parameter coupling system and the identification device thereof does not restrict the system orders, the nonlinearity type, the nonlinearity position and the number of nonlinear parameters included in the nonlinearity and can realize the simultaneous identification of all the nonlinear parameters included by any nonlinearity at any position of a large-scale nonlinear system.

The invention discloses a CNN-based handwritten Chinese text recognition method. According to the method, single handwritten Chinese recognition and a character segmentation algorithm are combined, automatic recognition of handwritten Chinese texts is achieved, and the method comprises the following steps that graying and binarization processing are conducted on text pictures, and then histogram projection is used for segmenting the Chinese texts; First, single-row characters are segmented through transverse scanning, and then single characters are segmented through longitudinal scanning. Carrying out scanning processing on a single Chinese picture, carrying out ortho-rectification on the Chinese picture, enabling the Chinese picture to be located in the middle of the picture, and leaving10 blank pixels in the upper, lower, left and right directions respectively; a convolutional neural network comprising four convolutional layers, four pooling layers and two full connection layers isconstructed based on a TensorFlow framework, and a training set is used for training; and inputting a to-be-tested picture, and performing recognition according to the constructed convolutional neuralnetwork.

The invention discloses an emotion detection method and device based on baby cry and relates to the technical field of big data. The emotion detection method comprises the following steps: acquiring ato-be-detected baby cry signal, and windowing the baby cry signal. The baby cry window signals in each time window are respectively subjected to fast Fourier transform so as to generate a spectrogramcorresponding to the baby cry signal. The spectrogram is input into a well-trained emotion detection model so as to determine the emotion corresponding to the baby cry signal. Therefore, the baby crysignal is converted into a corresponding sound frequency spectrum image through Fourier transform, and a capsule network is used for carrying out emotion recognition on the sound frequency spectrum,thereby improving the recognition accuracy of the baby cry. According to the technical scheme provided by the embodiment of the invention, the problem of low recognition accuracy of the baby cry in the prior art can be solved.

The invention discloses a plug-in identification method, device and system, and belongs to the field of network broadcasts. The method comprises the following steps that: a server transmits multimedia data streams to a user client, wherein the multimedia data streams include multimedia data packets and camouflaged data packets, and the camouflaged data packets carry feature information being different from the multimedia data packets; the server detects whether feedback information is received or not, wherein the feedback information is transmitted after the user client obtains the camouflaged data packets by decoding; and if the feedback information is not received, or the received feedback information does not conform to a predetermined condition, the server determines that the user client is a plug-in client. Through adoption of the plug-in identification method, device and system, the problem of low identification accuracy in an existing plug-in identification method is solved, and the effects of accurately identifying the plug-in client and preventing cheating in the ways of virtual machine technology, hardware information forging and the like are achieved.

The embodiment of the invention provides an image fusion method and device, a storage medium and an electronic device. The method comprises the steps of obtaining an initial pixel image and an initial radar image of a target scene obtained after synchronous shooting of the target scene; performing clustering calibration processing on the initial radar image to obtain first area information of the target object in the initial radar image; determining a mapping area of a target object included in the initial pixel image in the initial radar image, and determining the mapping area as second area information; determining a target association sequence between the initial pixel image and the initial radar image based on the first region information and the second region information; and fusing the initial radar image and the initial pixel image according to the target correlation sequence. According to the invention, the problem of low image recognition precision in the prior art is solved, and the effect of improving the image recognition precision is achieved.

The invention relates to a data analysis technology, and discloses a virtual resource allocation method, which comprises the following steps of: obtaining a service data set, and performing word segmentation processing on the service data set to obtain a word segmentation set; performing feature screening on the word segmentation set by using a feature screening model to obtain a feature word segmentation set; constructing a service feature vector of each piece of service data in the service data set by utilizing the feature word segmentation set; performing feature extraction on the obtained personnel information to obtain personnel features; calculating a distance value between the personnel feature vector and the service feature vector, and determining the service data corresponding to the service feature vector of which the distance value is smaller than a distance threshold as to-be-allocated service data; and allocating the to-be-allocated service data to the target service personnel. In addition, the invention also relates to a blockchain technology, and the service data set can be stored in a node of a blockchain. The invention further provides a virtual resource allocation device, electronic equipment and a computer readable storage medium. According to the invention, personalized virtual resource allocation can be realized based on personnel information so as to improve the overall efficiency of service execution.

The invention discloses a day-night switching detection algorithm based on video sequence image characteristic analysis. The day-night switching detection algorithm is characterized in that (1) during a day-night switching detection process, image pixel point luminance values, which are used to represent an illumination intensity change characteristic, are selected to be determining objects; (2) pixel points in a road surface range are used as observed objects, and an average luminance value of a plurality of luminance detection areas is acquired; (3) conditions of illumination changes are counted and calculated, and when the duration of the conditions reaches the requirement, last determination of a current environment condition is carried out. The day-night switching detection algorithm is advantageous in that the monitoring environment is effectively identified, and then different detection modes are adopted according to different environments in order to achieve the best effect; the efficiency of identifying the day mode and the night mode of the video monitoring system is improved, and then the problem of the low accuracy of the monitoring system caused by the fast change of the illumination intensity before dawn and at dusk is solved, and therefore the robustness of the video monitoring system is greatly improved.

The present invention discloses a Fast-Flux Botnet detection method and system in a high-speed network, which belongs to the field of network security detection technology. Firstly, a DNS data packetis obtained, and the captured DNS data packet is parsed and filtered, to obtain a suspicious Fast-Flux Botnet. A global association mapping relationship between a domain name and an IP address of thesuspicious Fast-Flux Botnet is analyzed, and global features are extracted by using the global association mapping relationship. Time-based local features of the suspicious Fast- Flux Botnet are extracted. Finally, the global features and the local features in offline data are trained by using a machine learning algorithm to obtain a trained machine learning model, and global features and local features in unknown data are detected by using the trained learning model to obtain a classification result of the suspicious Fast-Flux Botnet. The present invention can improve detection accuracy, reduce a false alarm rate, and improve processing efficiency.

The invention provides an intention identification method and apparatus. The method comprises the steps of extracting key elements from words input by a user, wherein the key elements are used for indicating semantics expressed by the words input by the user; according to priority information of intention dimensions which the key elements belong to, identifying user intentions indicated by the keyelements, wherein the intention dimensions are used for classifying query intentions expressed by the key elements according to a preset rule, and the priority information is used for indicating priorities of the intention dimensions; and according to an intervention correction rule, correcting the intention dimensions which the key elements belong to, and determining a target user intention. Byadopting the scheme, the problem of low identification accuracy of user intention identification in related technologies is solved; and the identification accuracy of the user intention identificationis improved.