518results about How to "Increase the amount of information" patented technology

The invention discloses a domain-knowledge-based short text classification method and a domain-knowledge-based short text classification system used in the technical field of information. The method is used for overcoming the defect that the traditional text classification method cannot well classify short texts. Aiming at the characteristics that the short text description concept signals are relatively weak and the text features are seriously insufficient, the invention provides the short text data classification method and the text classification system suitable for commodity web page data. According to the embodiment, a commodity classifier with excellent classification effect is obtained by reforming the traditional classifier, introducing new elements and devoting to matching application of algorithm and data. The introduction of the new elements comprises the following steps of: introducing a concept of domain words and introducing the concept into the classifier so as to effectively increase the information quantity of the short texts; and performing different-lexical-item-set-based semantic analysis on the short text data, particularly the web page commodity data, and introducing the semantic analysis result into the classifier so as to introduce new information for the commodity data information and improve the accuracy of text classification.

The invention discloses an abstract generation method and apparatus, and a computer device, and belongs to the field of natural language processing. The method comprises the steps of obtaining a document D, wherein the document D comprises at least one sentence; extracting m candidate sentences from the document D through an extractive model; and outputting target words according to the m candidate sentences through a generative model, and according to the target words, generating an abstract. Firstly the m candidate sentences suitable to serve as the abstract are extracted through the extractive model, so that the text sequence length needed to be processed by the generative model is reduced; then the target words are generated or extracted according to the m candidate sentences through the generative model; and the abstract of the document is synthesized according to the target words, so that the readability and information quantity of the finally generated abstract are improved.

The invention discloses a gastrointestinal tumor microscopic hyper-spectral image processing method based on a convolutional neural network, comprising the following steps: reducing and de-noising the spectral dimension of an acquired gastrointestinal tissue hyper-spectral training image; constructing a convolutional neural network structure; and inputting obtained hyper-spectral data principal components (namely, a plurality of 2D gray images, which are equivalent to a plurality of feature maps of an input layer) as input images into the constructed convolutional neural network structure using a batch processing method, and by taking a cross entropy function as a loss function and using an error back propagation algorithm, training the parameters in the convolutional neural network and the parameters of a logistic regression layer according to the average loss function in a training batch until the network converges. According to the invention, the dimension of a hyper-spectral image is reduced using a principal component analysis method, enough spectral information and spatial texture information are retained, the complexity of the algorithm is reduced greatly, and the efficiency of the algorithm is improved.

The present invention identifies biomarkers that are diagnostic of nerve cell injury and/or neuronal disorders. Detection of different biomarkers of the invention are also diagnostic of the degree of severity of nerve injury, the cell(s) involved in the injury, and the subcellular localization of the injury.

The invention discloses a distance measurement method and system based on binocular stereo vision. The method comprises the following steps: acquiring a position relationship between two fixed points; constructing a distance measurement model by using one fixed point as the original point according to the position relationship; and acquiring unknown parameters through easy calibration according to the model so as to realize the distance measurement of a target. The method can enlarge the information amount of infrared-visible images and provide position information of the target by using a heterogeneous source consisting of a thermal imager and a visible CCD (charge coupled device) camera to measure the distance.

The invention discloses a method of sending a data package and incidentally sending a reverse interest package in content center network. The method includes the following steps : step 1, sending the interest package to the network to by a user A show that the user A is interested in the content; step 2, an intermediate node dealing with a request when the intermediate node in the network receives the interest package; step 3, combining a response data package and the interest package which needs to be requested by the user A to form a mixed data package by a user B when the user B wants to send a data request to the user A; step 4, the intermediate node splitting the mixed data package into a data package and a reverse interest package when the intermediate node in the network receives the mixed data package and respectively dealing with the data package and the reverse interest package; step 5, splitting the mixed data package by the user A when the user A receives the mixed data package, receiving the data package and delivering the data package to primary requested program, and meanwhile responding to the interest package by the user A and producing a corresponding data package. The method of sending the data package and incidentally sending the reverse interest package in content center network reduces transmission times of the package and improves information quantity and efficiency in once transmission.

The invention discloses a method and a device for fusing infrared and visible light images based on spectral wavelet transformation. The method comprises the steps of performing spectral wavelet decomposition on an infrared image and a visible light image according to four decomposition scales to obtain a low-frequency sub-band coefficient and high-frequency sub-band coefficients, fusing the obtained low-frequency sub-band coefficient and the high-frequency sub-band coefficients at different decomposition scales according to a fusion rule, and performing spectral wavelet transformation on the fused low-frequency sub-band coefficient and high-frequency sub-band coefficients to obtain a fused image. The invention also discloses a device for fusing infrared and visible light images based on spectral wavelet transformation. According to the method and the device for fusing infrared and visible light images based on spectral wavelet transformation, the images of irregular images can be expressed better; a weight of image fusion is obtained by virtue of bilateral filtering so that the definition, the contrast ratio, the amount of information and the like of the fused image can be effectively improved, and consequently, a better-fused image can be obtained.

An image coding, recording and reading apparatus generates a local decoded image which is used for motion compensation prediction at the same time when coding an original image to be coded so as to generate an original image coded stream, acquires a difference image which is a difference between the local decoded image and the original image to be coded, codes the difference image so as to generate a difference image coded stream, records the original image coded stream and the difference image stream generated thereby into a storage memory, and adds an original image and a difference image which are obtained by decoding both the above-mentioned coded streams read out of the storage memory, respectively, so as to generate an addition image.

The invention provides a method, equipment and system of media negotiation of a multithreading meeting. The method comprises the following steps: sending a media declaring message, wherein the media declaring message has the information of two media data objects, and at least one information is the meeting place real-time picture including media view information; receiving a media selecting message, wherein the media selecting message has the information of the media data object selected via a second media entity; determining corresponding media data object based on the information of the media data object selected via the second media entity; and establishing a media transmission channel to the second media entity, so as to transmit the corresponding media data object to the second media entity through the media transmission channel. With the adoption of the method, the equipment and the system, more media data flows can be displayed; the accuracy and information content of the media data flow can be improved.

The invention discloses an intelligent vehicle positioning and road mark identification system based on a camera. The system comprises an image collecting module, an FPGA core control coding module and a data processing module, and the image collecting module is connected with the FPGA core control coding module which is connected with the data processing module; the image collecting module is used for collecting image information around a vehicle; the FPGA core control coding module is used for conducting lossless compression on the information collected by the image collecting module and transmitting compressed video to the data processing module; the data processing module divides the video to form image frames, target areas in the image frames are extracted, and identification is conducted on vehicle, pedestrian and road mark information; the distance of a target and the vehicle is judged according to the physical property of the target. The system is compact in structure, capable of adapting to vehicles of various types and conducting further depth development according to corresponding video information and high in expandability.

The invention relates to a video classification processing method and device, computer equipment and a storage medium. The method comprises the steps of obtaining a target video; extracting multi-modal data of the target video; performing classification prediction on the data of each mode to obtain probability vectors which respectively correspond to each mode and are used for representing the probability that the target video belongs to each preset category; combining the probability vectors corresponding to the modalities; and predicting the final category of the target video according to the combined vectors. According to the scheme, the accuracy of video classification can be improved.

The invention discloses an infrared and visible light image fusion method based on a self-attention mechanism, and belongs to the field of image fusion. The method mainly solves the importance problemthat subjective feelings of human eyes are ignored during image fusion. The method comprises the following steps: 1) establishing a deep self-coding network structure based on a self-attention mechanism, extracting features in an end-to-end manner, and simultaneously learning a fusion rule and a reconstruction rule; 2) inputting the feature maps of different branches of the coding layer into a self-attention mechanism to obtain an attention map, and adopting a mean value fusion strategy to obtain an output feature map; 3) designing a content loss function and a detail loss function which arerespectively used for highlighting infrared target information, sharpening edges and better utilizing texture details in the source image; 4) training a neural network, and visualizing a self-attention mechanism to adjust a network structure and a loss function. According to the method, attention can be distributed in an optimal mode by learning an attention graph, image key information is collected, the visual effect is improved, and the quality of a fused image is improved.

The invention provides an indoor positioning method based on multipath energy fingerprints, and belongs to the technical field of indoor wireless positioning and wireless communication. The method comprises the following steps that S1, in an off-line signal collecting stage, collected data is subjected to wireless signal estimation to obtain multipath parameters of k channels corresponding to each base station, in addition, RSSI (received signal strength indication) is calculated, and meanwhile, the corresponding position is measured to obtain a multi-dimension space fingerprint base; S2, in the on-line period, through actual measurement signals, the current position channel multipath parameters of users are obtained through estimation by using a channel estimation algorithm, in addition, the RSSI is calculated, and the actual measurement RSSI and the RSSI of the fingerprint base are compared to obtain fingerprint points with the 10 percent similarity; S3, the fingerprint points screened out in S2 are subjected to channel parameter comparison to obtain the average coordinate of several most similar points as the user position estimation value. The method provided by the invention has the advantages that the indoor channel multipath parameter information is perfectly merged into the indoor positioning algorithm, and the positioning precision and the stability can be more obviously improved.

The present invention discloses a video record method and device, terminal and a storage medium, belonging to the technical field of the Internet. The method comprises the steps of: when a video record command is received, collecting images of a target object in real time, and displaying an interactive special effect in a record interface, wherein the interactive special effect at least comprisesthe icon of the interactive object which performs interaction through a target motion; in the interaction process, detecting the motion information of the target object in real time; according to themotion information of the target object, determining a result special effect corresponding to the target motion, wherein the special effect result is configured to indicate an interaction effect of interaction of the target motion and the interaction object; displaying the result special effect on the record interface; and according to the collected image in real time and the interactive special effect and the result special effect displayed in the interaction process, generating a video file. The interaction process is increased to improve the user activeness and allow the video file to comprise a plurality of wonderful interaction moments so as to enrich the video content and improve the video interestingness.

The invention belongs to the field of infrared image processing, and relates to an image enhancement method based on long-wave infrared polarization feature extraction and fusion method. The method comprises the steps of (S1) collecting an infrared polarization image, and preprocessing the collected infrared polarization image; (S2) performing Stokes polarization state calculation on the preprocessed image in the step (S1), and calculating a polarization degree and a polarization angle according to a Stokes polarization state calculation result; (S3) carrying out completely polarized light decomposition calculation by using the polarization degree and the polarization angle; (S4) defining a polarization orthogonal difference value, setting a weighting coefficient, and extracting a polarization feature image, wherein the polarization feature image comprises a polarization parallel component image and a polarization vertical component image; and (S5) fusing the polarization feature imageand an infrared intensity image by using a non-downsampling shear wave algorithm. The method can effectively extract the polarization information of a target scene, has a good polarization feature extraction effect, performs fusion on the extracted polarization feature image and the infrared intensity image and significantly enhances the information content of the target scene.

The invention relates to the technical field of road traffic image recognition, and discloses a lane line multi-task learning detection method based on road segmentation, and the method comprises thesteps: constructing a multi-task learning network, carrying out the recognition and processing of an inputted road image, and outputting road segmentation data and lane line detection data; constructing a link coding structure, and connecting the two sub-networks; and performing alternate fine adjustment training on the two sub-networks, correcting and improving the lane line detection data precision by adopting a cross entropy loss function, and finally outputting corrected lane line detection data. The detection scale is diversified, the robustness is improved, the detection precision in a complex scene is improved, and the speed is high. The two sub-networks are connected through the link coding structure, and the information amount obtained by the feature map is increased through parameter hard sharing of the two sub-networks.

The invention discloses a wave front correction method based on defocused image training, and a device used by the method comprises an image acquisition module, a neural network module, a voltage conversion module and an actuating mechanism. The method comprises the following steps that the incident light is converged into defocused light spots through an optical prism; the image acquisition module is responsible for acquiring defocused light spots; after a large number of defocused light spot image training, the neural network module fits corresponding Zernike coefficients according to the image characteristics transmitted by the image acquisition module; the voltage conversion module establishes the relationship between the Zernike coefficient and a driving voltage; and finally, an execution mechanism realizes the correction of the wave front distortion generated by the atmospheric turbulence according to the fitted Zernike coefficients. The wave front correction method based on defocused image training does not need to use a Hartmann wave front detector and the like, and improves the utilization rate of light beam energy. The loop iteration process is not needed, the bandwidth is high, and the requirement of real-time performance is met. Only a single defocused image is needed to be collected for neural network training, so that the workload of image collection is reduced, and the information quantity of sample light spots is increased.

The invention proposes a big data platform based user behavior analysis system. The system comprises a user behavior information acquisition interface, a basic data layer module, a user behavior information analysis module, a user behavior information relevance determination module, a user behavior information organization module, a relevant data group representation module, a user feature analysis module, an aggregation module and an application module. According to the system, a relevant data group between user behavior information (between user behaviors and/or between the content of behavior objects) is established by utilizing interrelations of the user behaviors and the behavior objects, the data group describes effective information contained in the relation of the user behaviors, starting points and ending points of a series of the interrelated user behaviors can be accurately positioned, quantizable user behavior information can be accurately metered based on the relevant data group, and a series of the interrelated user behaviors can be subjected to comprehensive statistical analysis.

The invention discloses a two-dimensional code compiling method. The two-dimensional code compiling method comprises the following steps of: on the basis of a self-adaptive code compiling extended black and white code compiling technology and a code compiling scheme applicable to different multimedia, introducing an interweaving technology into an error-correcting code compiling technology by using a data compression technology with an open structure during compression code compiling; and adopting an advanced morphologic image pre-processing technology based on multi-scale structural analysis. The two-dimensional code compiling method has the advantages that a two-dimensional code is superior to the conventional two-dimensional code standard in terms of information bearing capacity, robustness, safety, reliability and adaptability; and coding and decoding bottom-layer supporting software provides technical support for various types of application of the two-dimensional code to the field of the internet of things in an open interface mode, the market monopoly position of the conventional foreign two-dimensional code standard can be broken, the chaotic situation of the conventional domestic two-dimensional code application is changed, and a firm foundation is laid for the development of the internet-of-things industry.

The invention provides a time sequence anomaly detection method and device, a server and a storage medium. The method comprises: extracting original features of a time sequence; processing the original features of the time sequence to obtain abstract features of the time sequence; performing anomaly detection on the abstract features of the time sequence based on a target anomaly detection model to obtain an anomaly detection result of the time sequence, wherein the target abnormity detection model is obtained through training in a supervised learning mode. The time sequence detection method and device are more universal in time sequence detection and higher in accuracy.

The invention relates to medicine and can be used for the disclosure of diseases in early stages. An object of the present invention is simplification and increase of accuracy of diagnosis of the state of human organism. According to the invention in the method of diagnosis of the human organism state consisting in obtaining visual images of structure of gas discharge streamers of fingers in electromagnetic field when contacted with electrode and in estimating streamer parameters, the visual images of structure of gas discharge streamers of fingers are divided into sectors, corresponding to various organs or systems of human organism, each sector is projected along the contour of the silhouette of human body, forming a single image at that, after which the human organism state is diagnosed by way of estimating parameters of gas discharge streamers of the obtained single image; closed curves are determined at that, corresponding to the borders of zones of contact of fingers with the electrode, cross points of these lines with the lines dividing the images of structure of gas discharge streamers of fingers into sectors are determined, and the single image is formed by these points along the silhouette of human body.