711results about How to "Low resolution accuracy" patented technology

The invention discloses a scene semantic segmentation method based on full convolution and a long-short term memory unit, relating to the technical field of image processing. The method includes the following steps of S1 constructing a deep neural network based on full convolution, a pyramid pooling module and long-short term memory unit module; S2 comparing a predictive image with a marked image, training by taking the Softmax loss as the objective function and the stochastic gradient descent as the optimization method, and updating the weight of the deep neural network obtained in step 1; S3 carrying out the S2 for many times, and completing the training until the loss is decreased to the limitation; and S4 inputting a new scene image to the trained deep neural network, and performing the bilinear interpolation to the original image resolution to obtain the semantic segmentation result of the scene. The method solves the problems that the current scene image segmentation is low in accuracy, and objects in the image are subjected to over-segmentation and under-segmentation.

The invention relates to a method and a system for discovering a user residence position based on mobile phone signaling. The method comprises the following steps: the obtained mobile phone signalingdata of the user is taken as the original trajectory data, and the original trajectory data is sequentially processed by data cleaning, data slicing, preliminary clustering, data verification and corecluster calculation, and finally the user staying position is obtained. Adopting a series of algorithms such as track slice aggregation, hierarchical clustering algorithm, kernel density clustering algorithm and machine learning algorithm, etc., considering mobile phone track point density, signaling time interval, track point moving direction and moving distance, etc., we can judge user 's staying position, staying time, entering and leaving staying point time. The algorithm can effectively eliminate the interference of base station drift on user location judgment, improve the identificationaccuracy of user dwell point position and dwell time, and objectively restore the user dwell position.

The invention discloses an audio and video processing method, a device, equipment and a medium, which relate to the computer technical field. The method comprisesseparating image frame information andaudio information from the video file; extracting image feature information and audio feature information from the image frame information and the audio information, respectively; fusing the image feature information and the audio feature information into video content feature information; determining a classification result corresponding to the video file according to the video content characteristic information. The invention combines the audio characteristic information in the video and the image characteristic information of the video frame for video classification, improves the accuracyand recall rate of the video classification, thereby reducing the labor cost of the video classification examination.

The embodiment of the invention provides a method and device for recommendation. The method specifically comprises the steps that at least one recommendation content is generated according to ecological historical behavior data of a user, wherein the user's ecological historical behavior data comprises at least one type among historical behavior data types including historical behavior data of the user generated in at least two applications of at least one terminal as well as historical behavior data of the user generated in at least one application of at least two terminals; and the recommendation content is recommended to the user. The method and the device provided by the embodiment of the invention can increase accuracy of the recommendation content.

The invention discloses an authentication method and device and belongs to the technical field of information security. The authentication method comprises the following steps: providing a prompting message, used for prompting to make an object authentication action, for a user, wherein the object authentication action comprises facial expression and/or head movement; acquiring at least one frame image; detecting whether authentication action information of the user in at least one frame image is matched with a reference action message when the object authentication action is made, and if the authentication action information of the user is matched with the reference action information, determining that authentication of the user passes. The authentication method and device can be used for solving the problem that accuracy of terminal during authentication is relative low as a photograph can be adopted for authentication in the prior art; and the effects of user participation in an authentication process and improving authentication accuracy are achieved.

The invention relates to a live broadcast previewing method and belongs to the field of video live broadcast. The method includes the following steps of obtaining broadcast starting data of an anchor client, obtaining live broadcast preview pictures from video data streams provided by the anchor client at preset time intervals according to the broadcast starting data, and providing the live broadcast preview pictures for an audience client. The live broadcast preview pictures are obtained from the video data streams provided by the anchor client at preset time intervals and are provided for the audience client, so that the problems that live broadcast preview pictures in the prior art can only reflect live broadcast content when the anchor client starts broadcasting, and the live broadcast preview pictures are low in accuracy after the anchor client starts broadcasting for a period of time are solved; the live broadcast preview pictures are obtained at a time at preset time intervals, and the effects that the live broadcast preview pictures can reflect live broadcast situations relatively in real time and the accuracy of the live broadcast preview pictures is improved are achieved through a small computation amount.

The application discloses a prediction method and device of internet resource request values used for improving the accuracy of predicting the internet resource request values. The method comprises the steps of obtaining the historical request data, and determining the time series data according to the historical request data, wherein the time series data comprises the request values of the target resources within a plurality of unit time quanta; according to the time series data, determining the time series trend information, and determining the preliminary prediction values of the target resource requests according to the time series trend information; obtaining the historical exchange rate data of the target resources and the own resources; according to the historical exchange rate data, determining the exchange rate trend information, and determining an adjustment coefficient according to a preset rule; according to the preliminary prediction values and the adjustment coefficient, predicting the request values of the target resources within the unit time quanta. The application also discloses a prediction method of the internet transaction amount.

The invention relates to a method and device for processing channel state information. The method comprises the steps that UE receives the high-level configuration information of N CSI-RS resources of a network side; the UE receives corresponding CSI-RS signals according to the high-level configuration information and carries out channel measurement to obtain a channel measurement result; the UE confirms M1 first kind of channel state information and M2 second kind of CSI according to the channel measurement result and takes the M1 first kind of channel state information and the M2 second kind of CSI to report to the network side, wherein the M1 is an integer larger than or equal to 1, M2 is an integer larger than or equal to 0, the first kind of CSI is obtained according to the CSI-RS signals on one CSI-RS resource, and the second kind of CSI is obtained according to the CSI-RS signals on two or three CSI-RS resources. The method and device can adapt to various COMP transmission modes, and the reduction of the CSI accuracy rate cannot be caused.

The invention provides a roadside parking space detection method and relates to the communication field. By using a current geomagnetic sensor detection parking space mode, a detection result is easy to generate a deviation because of an external influence so that accuracy is low. By using the method of the invention, the above problem can be solved. The method comprises the following steps: burying a geomagnetic sensor and a wireless communication amplification module in a center position of a parking space, and collecting vertical geomagnetic data and horizontal geomagnetic data of the parking space and RSSI data of the parking space, wherein the vertical geomagnetic data and the horizontal geomagnetic data are acquired by the geomagnetic sensor; the RSSI data is acquired by the wireless communication amplification module; the vertical geomagnetic data is the geomagnetic data which is vertical to a ground direction; the horizontal geomagnetic data is the geomagnetic data which is parallel to the ground direction; comparing the collected vertical geomagnetic data of the parking space with a preset stable vertical geomagnetic data value. A technical scheme provided in the invention is suitable for traffic management and accurate parking space detection is realized.

The invention provides a front end interface code generation method, and belongs to the technical field of computers. The problem that in the prior art, a front end interface code generation method islow in accuracy is solved. The method comprises the steps that front end interface design data is exported from a front end interface design layout, information collection and conversion treatment are performed on the front end interface design data, control information included in the front end interface is determined, and on the basis of input configuration information and the control information, front end interface codes are generated. According to the method, the control information in the front end interface is exported on the basis of the design layout, the front end interface codes are automatically generated according to the exported control information and configuration information, the interface design information in the design layout is sufficiently reserved, the defect that generated interface codes are not accurate due to the fact that obtained interface information is not sufficient when interface codes are directly generated from screenshots is overcome, and the accuracy of the automatically generated front end interface codes is effectively improved.

One embodiment of the invention provides a question answering device and method for vehicles, belonging to the field of vehicles. The device comprises a receiving module for receiving questions brought out by users; an application scene classification module for performing application scene classification on the questions brought out by the users, wherein the application scene comprises one or many of the followings: fault type, general knowledge type, handbook type, car purchasing type and insurance type; a semantic comprehension module for semantic recognition of the questions brought out by the users according to the classified application scene to obtain feature information; an answer assuring module for determining answers corresponding to the questions brought by the users according to the feature information; and an answer outputting module for outputting the answers corresponding to the questions brought by the users. The method and device effectively solve the problem of low accuracy of FAQ-based question answering mode.

The invention provides a wireless network cross-boundary coverage detection method and device and a communication system. The detection method comprises the steps of acquiring location information and signal information of each measurement report (MR) sampling point in a cell; calculating cross-boundary coverage parameters of the cell according to the location information and the signal information of each measurement report MR sampling point, wherein the cross-boundary coverage parameters comprise whether the cell is a cross-boundary coverage cell or not and a boundary-crossed coverage cell covered by the cell if the cell is a cross-boundary coverage cell. According to implementation of the method, firstly the location information and the signal information of all the MR sampling point in the cell to be detected are acquired, and the cross-boundary parameters of the cell are calculated according to the information. The method has nothing to do with a mobile station, thereby avoiding interference brought about by the mobile station, and being high in accuracy. Furthermore, the method needs to calculate according to multiple parameters, so that the accuracy is further improved, and a problem that an existing cross-boundary coverage judgment method is low in accuracy is solved.

The invention discloses a sound signal time delay estimation method and device and belongs to the technical field of audio processing. The sound signal time delay estimation method comprises the steps of obtaining two sound signals, conducting coherence matching on the two sound signals through short-time Fourier transformation of the two sound signals to obtain a first matching result, wherein the first matching result includes first matching positions and first matching degrees of the two sound signals; conducting coherence matching on the two sound signals according to spectral correlation of power spectrums of the two sound signals to obtain a second matching result, wherein the second matching result includes second matching positions and second matching degrees of the two sound signals; calculating the time delay of the two sound signals according to the first matching result and the second matching result. The problem that a relevant time delay estimation method is low in accuracy is solved. The sound signals are matched from two aspects of frequency domain distribution and power spectrums, the two matching results are integrated to determine a final matching result, and matching accuracy and time delay estimation accuracy are improved.

The invention discloses a voice end node detection method and device. The voice end node detection method comprises the following steps: acquiring to-be-detected audio frequency in targeted environment; utilizing a voice training model generated in advance to mark the audio clips in the to-be-detected audio frequency; marking the voice clips in the to-be-detected audio frequency as first pre-arranged identification, wherein the voice training model is a model obtained from training of the voice data in the targeted environment; searching the whole segment of audio frequency provided with the first pre-arranged identification in the marked to-be-detected audio frequency, wherein the whole audio frequency is the whole segment of voice marked with the first pre-arranged identification; taking the whole segment of audio frequency provided with the first pre-arranged identification as a segment of voice; taking the starting node and the end node of the whole segment of the audio frequency provided with the first pre-arranged identification as the end nodes of the voice. According to the invention, the problem in the prior art that the voice end node detection accuracy is relatively low under the condition of high noise is solved; the voice end node detection accuracy is improved.

The embodiment of the invention discloses a three-dimensional image acquisition method and device, a three-dimensional image positioning method and device, equipment and a storage medium. The three-dimensional image acquisition method comprises the steps of acquiring a target two-dimensional image and point cloud data corresponding to the target two-dimensional image; endowing three-dimensional image points in the point cloud data with the feature vectors of the two-dimensional image points in the target two-dimensional image according to a coordinate corresponding relation between the three-dimensional coordinates corresponding to the point cloud data acquisition equipment and the two-dimensional coordinates corresponding to the target two-dimensional image acquisition equipment; and obtaining a three-dimensional scene image corresponding to the target two-dimensional image according to the three-dimensional image points with the feature vectors in the point cloud data. According to the technical scheme provided by the embodiment of the invention, the scene reconstruction accuracy based on the two-dimensional image is improved; accurate three-dimensional reconstruction can be carried out on a large-scale scene, almost no scale difference exists between a three-dimensional image obtained through reconstruction and a real scene, and the shooting equipment attitude correspondingto the two-dimensional image can be simply, conveniently and rapidly obtained through a small calculation amount.

The embodiments of the application disclose a text image correction method and device, a storage medium and equipment, and belong to the technical field of image processing. The method comprises the following steps of: acquiring a text image to be corrected, wherein the text image contains a text; performing binarization processing on the text image to obtain a binarized region; filtering the binarized region corresponding to a non-text region in the text image through a character classification model, and generating a binarized image according to the binarized region obtained after filtering,wherein the binarized region obtained after filtering is a binarized region corresponding to the text region in the text image; projecting the binarized image to obtain an angle of inclination of thetext; and correcting the text image according to the angle of inclination. The text image correction method provided by the embodiment of the application can improve the accuracy rate of correction of the text image.

The invention discloses a fish fine-grained classification method based on deep learning. The method comprises the steps of preprocessing an acquired image; carrying out feature extraction by using adeep neural network; constructing a feature pyramid network for regional proposal; carrying out cutting and feature extraction on a proposed area, then carrying out primary classification by utilizingextracted features on one hand, inputting the classification accuracy into an area proposal network as a supervision signal on the other hand, fusing the features and full-graph features, sending thefused features into a full-connection layer for classification, and outputting a final classification result. According to the invention, when an existing object classification technology carries outa fine-grained classification task, a fine-grained classification task is carried out; due to the problems of complex environment, fine inter-class difference between classes and low accuracy causedby large intra-class difference, when the method is used for performing fine-grained classification on the fish images under the complex background, the identification time is short, the identification accuracy is high, extra label information is not needed, and the method is suitable for popularization and application.

The invention discloses a method and a device for identifying irises. The method comprises the following steps of rotating an extracted image matrix of upper eyelid boundary images from a first iris image according to a plurality of rotation angles, so as to obtain a plurality of rotation matrixes; obtaining a first number of crossing points of a first parabola matched with each rotation matrix and the upper eyelid boundary determined by each rotation matrix; using the rotation angle corresponding to the maximum value of the first number as the rotation angle of eyes in the first iris image; according to the rotation angle of the eyes in the first iris image, determining the initial angle for normalizing the first iris image; using the initial angle to normalize the first iris image; using the extracted image feature from the normalized image to search the feature information in a database, so as to identify the iris of the first iris image. The method has the advantages that the problem of low accuracy of iris identification in the prior art is solved, and the effect of improving the iris identification accuracy is improved.

The invention discloses a method and a device for acquiring meteorological data. The method comprises the steps of acquiring a grid point meteorological data set in a target area, wherein the grid point meteorological data set comprises meteorological data of a plurality of grid points, and the meteorological data of each grid point at least comprise wind power data, air temperature data and precipitation amount data; determining a plurality of electric power points in the target region, wherein the electric power points at least comprise a transformer substation and tower points, the plurality of tower points are connected to form a power transmission line; calculating the data in the grid point meteorological data set by adopting a preset algorithm, and determining the meteorological data of each electric power point, wherein the preset algorithm at least comprises a weighted interpolation method and a linear interpolation method. By means of the method and the device, the problem that the accuracy of meteorological data of a transformer substation and a power transmission line tower point is low in the related technology is solved.

The invention provides a speech and semantic recognition method of a service robot. The method comprises: system word library modeling is carried out; to be specific, a word library database is established and then a semantic frame database is established; a lip language mode base is established; a system uses a speech recognition system to collect a speech instruction inputted by a user and a face video and speech and semantic recognition and video segment lip language recognition are carried out respectively; and according to a combined recognition result, a determined Chinese semantic information is displayed at a display interface. Therefore, a noun database and a semantic frame database are established for different service fields; and on the basis of the word library database, the semantic frame database, and the lip language mode base, the robot understands the Chinese meaning expressed by the speech instruction, so that speech and semantic recognition becomes accurate.

The invention discloses a method and device for detecting network attacks. The method comprises the following steps: receiving a network request from a sender; when determining that the type of the network request is an attack request by adopting an attack detection rule set, forwarding the network request to a receiver, and acquiring a network response corresponding to the network request; and detecting the network response by adopting the attack detection rule set, and selecting a processing mode of the network response according to a detection result. By adopting the method and device disclosed by the invention, the technical problem that a one-way network attack detection method adopted in related technologies is relatively low in accuracy can be solved.