61results about How to "Fully featured" patented technology

The invention discloses a hyperspectral image classification method based on fusion of multi-scale and multi-dimensional spatial spectral characteristics. The method comprises the following steps: (1)inputting a hyperspectral image; (2) preprocessing the hyperspectral images to be classified; (3) tracking neighborhood blocks; (4) generating a training set and a test set; (5) constructing a multi-scale spatial spectrum feature and multi-dimensional feature fusion network; (6) training a multi-scale spatial spectrum feature and multi-dimensional feature fusion network; and (7) classifying the test samples. The method provided by the invention can effectively solve the problems of too single feature and too single scale of the convolutional neural network during training, can solve the problem of low average classification precision AA during hyperspectral classification, can maintain the recognition capability of small sample number categories while realizing high classification precision, and is good in classification performance.

The invention relates to a CRISPR / Cas9-based sgRNA design method. The method is characterized by comprising the following steps of obtaining the value of enzyme digestion efficiency of sgRNA and corresponding Cas9; building a personalized sgRNA design model; measuring the quality of the built personalized sgRNA design model by applying an NDCG algorithm and updating a database; and designing sgRNA and giving out an assessment value of each sgRNA. Compared with the prior art, the method has the characteristics of high accuracy, complete characteristics, wide application range and wide data analysis range.

Disclosed herein is an information feedback method and system based on instant messaging relating to telecommunication techniques. An instant messaging system obtains a location of a mobile terminal from a mobile positioning system according to request information of the mobile terminal. An automatic information processing system generates location-based information according to the content of the request information and obtained location. The instant messaging system feeds back the location-based information to the mobile terminal. The request information that the mobile terminal sends to the instant messaging system includes an instant messaging identifier of the mobile terminal. The instant messaging system obtains a corresponding and relevant mobile identifier according to the instant messaging identifier. The mobile identifier indicates the information identifier of the mobile terminal in the mobile network system and is used for the mobile positioning system to obtain the location of the mobile terminal. In this way, the disclosed method and system provide a value-added technique relating to location with high serviceability and practicability.

The invention relates to an unmanned aerial vehicle with an umbrella. The unmanned aerial vehicle comprises an unmanned aerial vehicle body and the umbrella, wherein the unmanned aerial vehicle body comprises a fuselage; four wings and four feet are arranged on the fuselage; a propeller and a driving motor are arranged on each wing, and each driving motor drives the corresponding propeller to rotate so as to realize ascending, descending and turning of the unmanned aerial vehicle; a connecting block is arranged on the fuselage in an outwardly raised manner relative to the fuselage; the umbrella comprises an umbrella cover and a bracket for supporting the umbrella cover; through a connecting part, the umbrella is mounted on the connecting block of the fuselage. Through the umbrella, the balance of the unmanned aerial vehicle during flying is ensured, and a crash condition caused by out of control of the unmanned aerial vehicle is avoided; meanwhile, the unmanned aerial vehicle is suitable for being used in a rainy environment; in addition, characteristics of the unmanned aerial vehicle body are fully utilized, the umbrella plays the common umbrella function in air, and the unmanned aerial vehicle body and the umbrella cooperate with each other, so that use requirements of people can be met.

The invention discloses a radar radiation source recognizing method based on a VAE-ResNet network. The problems that network training cannot be completed under few signals, the radar radiation sourcefeature extraction is not sufficient and the recognition accuracy is low in the prior art are mainly solved. By means of the scheme, seven different types of radar radiation source signals are generated; the radar signals are output in the form of a sequence, then time-frequency converting is conducted on the signals, and the signals are output in the form of a time-frequency graph; the data output in the two forms is concentratedly marked with the signal types, and a training set and a testing set are prepared; the VAE-ResNet network is established; the VAE-ResNet network is trained through the training set; the testing set is sent into the trained network, and the network output is of a radar radiation source type. The radar radiation source recognition feature extracting completeness degree and recognition precision are improved, and the method can be used for radar signal recognition under a complicated electromagnetic environment.

The invention discloses a monogenic multi-characteristic face expression identification method based on sparse fusion. The method comprises the following steps of: 1, carrying out monogenic filtering on an expression image after pre-processing, and obtaining monogenic amplitude information, monogenic direction information, monogenic phase information, transverse conversion information and longitudinal conversion information in three dimensions; 2, utilizing five kinds of monogenic information to extract a monogenic binary mode histogram characteristic, a monogenic direction histogram characteristic and a monogenic phase histogram characteristic of the face expression image, wherein the three kinds of characteristics respectively construct corresponding sparse dictionaries; and 3, utilizing a l1 regularization least square method to optimize weights of the three sparse dictionaries, and realizing face expression identification by means of weighted fusion. According to the invention, the texture, shape and direction characteristics of the face expression image can be fully extracted, and the expression identification rate is improved.

The invention discloses an arc welding seam forming accurate prediction method based on deep learning. The arc welding seam forming accurate prediction method comprises the steps that a correspondingwelding process is designed according to welding equipment, a welding material, a structural form and a welding method; designing a corresponding process experiment according to a welding process, andimplementing in batches; according to different combinations of welding equipment models, material marks, structural forms and welding methods, weld joint forming sub-databases are established respectively; constructing a deep neural network taking the process parameters as input and taking the weld joint forming size and the weld joint section profile as output, and training and evaluating the deep neural network by utilizing data of the sub-database to obtain a weld joint forming prediction model; and real-time technological parameters in the welding seam forming process are collected and input into the welding seam forming prediction model, the corresponding welding seam forming size and welding seam section contour are output, and accurate prediction of arc welding seam forming is achieved. According to the invention, accurate prediction of weld joint forming can be realized.

An external cavity diode laser system includes a thin film filter. The output facet of the laser diode is coated with a partial reflection coating and the cavity side facet is coated with an anti-reflection coating. A roof-top prism or corner cube retroreflector serves as the laser cavity end reflector and provides stability of the wavelength of the output over time. The laser cavity lies between the partial reflective coated facet of the laser diode and retroreflector. A collimating lens and a thin film filter are located between the end reflector and the laser diode cavity side facet. The lasing wavelength can be adjusted during either manufacture or operation by tilting the filter. Also included is thermal compensation in the mounting for the retroreflector to compensate for thermal movement of the laser system cavity.

The invention provides a side-scanning sonar image region-of-interest extracting method based on region growing, which comprises the following steps of: obtaining the seed region of a side-scanning sonar image sound shadow region by adopting an L-G operator, obtaining the seed region of a side-scanning sonar image sound reflecting region by utilizing a neighborhood gray-scale mean value, then considering the similarity of pixel points relative to the seed regions to be used as a growing condition, and respectively obtaining the extraction results of the sound shadow region and the sound reflecting region through region growing to obtain the region-of-interest extraction of a side-scanning sonar image. From a processing result, the method has better effect on the region-of-interest extraction of the side-scanning sonar image.

The invention discloses a traffic sign recognition method based on spare self-encoding and sparse representation. The traffic sign recognition method comprises the steps of: collecting pictures containing traffic signs, and dividing the pictures into l types manually; extracting blocks of each type of the pictures containing the traffic signs, and preprocessing the image blocks; training l sparse auto-encoders respectively by taking the preprocessed image blocks as training samples, and denoting weight dictionaries as D1, D2 , ... , Dl; extracting a block of a test picture with an unknown traffic sign, and constructing a test image block; calculating a sparse coefficient of the test image block under each weight dictionary by utilizing a sparse representation principle and adopting an OMP algorithm; and solving F norms between a reconstructed image block and the original image block, and selecting the dictionary type with the minimal F norm as a traffic sign recognition result of the test picture. The traffic sign recognition method adopts spare self-encoding to extract sufficient features of traffic signs automatically, completes recognition through calculating distance between a reconstructed sample and the test sample, and can achieve high recognition accuracy.

A snowman toy includes a housing, a body mounted on the housing and rotatable in right and left directions, and a head mounted on the top of the body and swayable in right and left directions. The body and the head respectively are a hollow case which is formed by melting and joining outer edges of a plurality of EVA granules. The EVA granules are transparent and a great quantity of apertures are defined between the EVA granules. A plurality of LEDs and a music player are disposed in the interior of the snowman toy. The snowman toy of the present invention is made by EVA material with good translucence and reflection of light, thereby fully representing the sparkling and crystalline features of a traditional snowman. Furthermore, the snowman toy has functions of playing music, emitting colorful light, rotating the body and swaying the head, which increase the entertainment effect.

The invention discloses a vehicle detection method based on a multi-branch cyclic self-attention network and cyclic frame regression. The vehicle detection method comprises the following steps: constructing a trunk network for vehicle detection; predicting a candidate box by using an RPN network, extracting an instance feature map according to the candidate box, circularly predicting the self-attention map and calculating a new instance feature map to obtain a final instance feature map; circularly selecting a feature map output by the basic network according to the predicted detection frame by using circular frame regression; and expanding the computing network to multiple branches by using a multi-branch network structure, and fusing a multi-branch detection result. Compared with a traditional vehicle detection method based on deep learning, the vehicle detection method based on the multi-branch cyclic self-attention network and the cyclic frame regression has the advantages that more accurate features of the vehicle can be obtained, the confidence coefficient of a detection frame of the vehicle is improved, a more accurate detection frame can be obtained, and the detection accuracy is improved.

The embodiment of the invention discloses a driving state recognition method and device, a terminal and a storage medium, and belongs to the field of information processing. The method comprises the steps: performing feature extraction on collected driving state data, and obtaining driving state features; processing the driving state feature data through an RNN model to obtain a first output feature; performing attention calculation on the first output feature through an attention model to obtain an attention weight vector, performing scaling processing on the attention weight vector, and performing weighting processing on the first output feature according to the processed attention weight vector to obtain a second output feature; performing probability conversion on the second output feature to obtain a first probability matrix; based on the first probability matrix, determining a driving state of the mobile terminal, the driving state being used for indicating a driving state of walking of the subway or the user. According to the invention, the driving state of the subway or the walking of the user can be detected through the mobile terminal, the flexibility is high, and the identification accuracy is high.

A sheath for use with a scalpel cutting device including a sheath body configured to cover a scalpel blade at a distal end of a scalpel cutting device. The sheath advantageously may include a plurality of slots formed therein. Providing two or more slots provides additional functionality and / or ease of manufacturing. A single sheath configuration including a plurality of slots (e.g., two) may be manufactured so as to be used with scalpels including somewhat different blade configurations (e.g., which may require a slightly different slot location). Such a universal sheath allows a single sheath design to be used with multiple scalpels, easing manufacture and reducing costs. In addition, depending on the particular location of the slots and the configuration of the scalpel blade, the different slots may each intersect the scalpel blade and provide different characteristics for cutting a suture.

The invention discloses a text intention matching method oriented to intelligent questions and answers and based on internal correlation coding, and belongs to the field of artificial intelligence. Inorder to solve the technical problem of how to accurately judge whether a text intention is matched or not, the adopted technical scheme is as follows: a text intention matching model consisting of amulti-granularity embedding module, an internal correlation encoding module, a global reasoning module and a label prediction module is constructed and trained to realize deep encoding of informationof different granularities of a text, and meanwhile, a soft alignment attention mechanism is used for obtaining internal correlation information between different granularities; a representation of the text and a multi-granularity representation between the texts are generated through global maximum pooling and global average pooling; similarity calculation is performed on the representations ofthe two texts, and a similarity calculation result is combined with the multi-granularity representation between the texts to obtain a final interaction information representation of the text pair; and the text pair intention matching degree is calculated to achieve the purpose of judging whether the text pair intention is matched or not.

A branched conjugated diene-aromatic vinyl copolymer (C) being a random copolymer, wherein the aromatic vinyl bond content in the conjugated diene-aromatic vinyl copolymer (C) is from 30 to 38 mass%, the vinyl bond content in the total bond units of the conjugated diene is from 30 to 43 mol%, the weight average molecular weight (Mw-C) in terms of polystyrene obtained by gel permeation chromatography (GPC) of the conjugated diene-aromatic vinyl copolymer (C) is from 700,000 to 1,000,000, the ratio of the weight average molecular weight (Mw-C) to the number average molecular weight (Mn-C) ((Mw-C) / (Mn-C)) is from 1.7 to 3.0, and the Mooney viscosity (ML-C) and the Mooney relaxation ratio (MSR-C) measured at 120 DEG C satisfy the relationship represented by formula (1): {214-(ML-C)} / 300 = (MSR-C) = {260-(ML-C)} / 300 (wherein 100 = (ML-C) = 140).

The invention relates to a partial discharge pulse feature extraction method and device, computer equipment and a medium. The method comprises the following steps: acquiring a partial discharge signal; based on an adaptive signal time-frequency processing method, decomposing the partial discharge signal to obtain an intrinsic mode function component; based on the statistical property of the noisesignal, randomly sorting the intrinsic mode function components to obtain a reconstructed signal; accumulating and averaging the reconstructed signals to obtain denoised partial discharge signals; decomposing the denoised partial discharge signal into a plurality of product functions based on a local mean decomposition method; and extracting pulse features according to the plurality of product functions. According to the partial discharge pulse feature extraction method provided by the invention, the pulse features can be extracted more completely, so that the partial discharge signals can beanalyzed accurately.

The invention discloses a bearing fault diagnosis method and system based on a sparse auto-encoder and Softmax, and the method comprises the steps: introducing local constraints into an original sparse auto-encoder, and obtaining an improved sparse auto-encoder; local constraints of an improved sparse self-encoding model are reflected in normalization of an encoder weight matrix, only k activation units with the maximum cosine similarity of a hidden layer are reserved to form a local subspace of an original sample, the local subspace is correspondingly characterized by k neighbors of a sample x, and a decoder reconstructs input through the reserved k coding units. Vibration signals of a rolling bearing in different running states are collected as a training set, the training set is used for training a local sparse auto-encoder model and a Softmax classifier model, model parameters are obtained, and therefore building of a fault diagnosis classification model is completed, due to the fact that local features of the vibration signals are considered, features learned by the local sparse auto-encoder are more complete, and the accuracy of the trained model is higher.

This invention discloses a conversion method for the favorite TV picture color space including the following steps: first of all, selecting the armed color sphere including a brightness sphere, a tone sphere and a saturation sphere, then setting corresponding target color sphere to test the color information of each pixel in the input image signals and convert it to the corresponding target color sphere when it belongs to the armed color sphere then to output the determined processed image signals.

The invention discloses a local feature characterization method based on a face expression image. The local feature characterization method is characterized by comprising the following steps: 1, utilizing a Haier detector to divide a face expression image into an eyebrow sub-image, an eye sub-image and a mouth sub-image according to the relation that the nose accounts for one-third of the face in the longitudinal direction and the eye counts for one-fifth of the face in the transverse direction; 2, obtaining a sufficient vector triangular code; 3, utilizing the sufficient vector triangular code to conduct local feature analysis on the eyebrow sub-image, an eye sub-image and a mouth sub-image. By means of the local feature characterization method based on the face expression image, local features of the face expression image can be effectively presented, the computing complexity is reduced, and timeliness and precision of feature extraction are improved.

The invention provides a wind turbine blade crack diagnosis method and device, and relates to the technical field of wind turbine fault diagnosis, and the method comprises the steps: obtaining a plurality of blade images, continuously collected by binocular camera equipment at a preset frame rate, of wind turbine blades under the application of preset excitation; arranging a plurality of marks onthe wind turbine blades; identifying a mark on the leaf image and determining a space coordinate of the mark so as to generate space vibration data of the mark; intercepting continuous preset frame data from the marked spatial vibration data to serve as initial data; wherein each frame of leaf image corresponds to one frame of data; correspondingly mapping the vibration data of each mark in the initial data in the X-axis direction, the Y-axis direction and the Z-axis direction to R, G and B components on the image so as to generate a fused image fused with the vibration information of each mark; and diagnosing the crack of the wind turbine blades according to the fused image. The wind turbine blade crack diagnosis method has a good effect in wind turbine blade crack diagnosis.

The invention belongs to the field of image processing and target detection, and discloses a target classification and attitude detection method based on a deep convolutional neural network. The core content of the method includes: 1. a method for making a sample set; 2. a method for generating candidate windows; 3. a method for feature extraction; 4. a method for network training; After the image to be detected gets the deep convolution feature through the convolutional neural network, the candidate window with the attitude angle is mapped to the feature layer to obtain the directional regional feature vector, and the final detection result is obtained by classifying and predicting the feature vector . This method can extract more pure target features from samples, improve the accuracy of classification, and realize the function of target attitude angle detection.

The invention discloses a myoelectric signal characteristic extraction method. Firstly, the myoelectric signal is changed into Fourier coefficients by Fourier transformation, and the Fourier coefficients of a preset frequency band are selected and normalized. A stack automatic encode is constructed, and network parameter of each layer are set for training and updating parameters; A stacked restricted Boltzmann machine is constructed, and the parameters of each layer network are set for training and updating. At that completion of the pre-trainning of the automatic encoder and the limit Boltzmann machine, The parameters of the two parts of pre-training are connected together, that is, the output of the automatic encoder is the input of the limited Boltzmann machine, which forms a through-network to realize the feature extraction of EMG signal, and solves the problems that the time-domain feature changes greatly and the frequency-domain feature extraction is inadequate when the existingEMG signal extraction methods are used.

The invention discloses a target tracking method and system based on an attention mechanism and global reasoning, and belongs to the technical field of computer vision, and the method comprises the steps of carrying out the target tracking through a target tracking model based on a twin network, wherein the target tracking model comprises a template branch and a search branch, the template branch and the search branch each comprise a backbone network, a parallel attention mechanism and a global reasoning module; obtaining an initial frame picture and a current frame picture, taking the initial frame picture and the current frame picture as input of the template branch and input of the search branch respectively, and obtaining a first score graph and a second score graph; performing weighted summation on the first score graph and the second score graph to obtain a regression graph; and determining the position of a target according to the regression graph. Compared with an existing tracking algorithm, the method has a better tracking effect.

The invention provides a method for constructing a pistillate line recombinant inbred line group for balsam pear. The technical scheme comprises following steps: group construction base requirement, sowing time, seedling growing, soil preparing and fertilizing, field planting, fertilizing and watering management, frame building and pruning, artificial pollination and insect and disease controlling, seed collecting and seed storing. By means of single plant self-cross, plants from F2 to F6 are all preserved as possible; the characteristic of single plant is kept completed. By means of the method, a large enough recombinant inbred line group from F2 to F6 is obtained; the advantage of the recombinant inbred line group (RI) is that it can be used for long time and can be tested repeatedly. Therefore, the recombinant inbred line group can be used for constructing molecule mark linkage and mapping of quantitative trait loci (QTL). By means of the method, single plant in the pistillate linerecombinant inbred line group for balsam pear is kept as possible; the method is easy to operate and the group has stronger group integrity.

The invention provides an animation sketch coloring method based on a dynamic memory module and a generative adversarial network, and belongs to the technical field of image processing. An image coloring task is completed by using a dynamic memory module for the first time, firstly, an animation sketch is input into a network, and a picture feature graph is obtained; information is dynamically read from the image tag by using a dynamic memory module, the image features and the image tag information are combined, and a generator and a discriminator are trained to generate a color image according with tag description. According to the invention, the dynamic memory module and the conditional generative adversarial network are utilized to achieve the task of generating the color picture from the animation sketch and the corresponding picture label, and the picture quality is improved by dynamically combining the label information and the image information; according to the invention, evaluation is carried out by using a real image data set, and the improvement on the most advanced baseline is observed.

The embodiment of the invention is suitable for the technical field of image processing, and provides a feature extraction network training method and device, terminal equipment and a medium, and the method comprises the steps: determining a background interval of each pixel point in a video frame sequence; according to the background interval, foreground pixel points in a target video frame are determined, and the target video frame is any video frame in the video frame sequence; acquiring a foreground image in each target video frame according to the foreground pixel points; determining a positive sample and a negative sample of a foreground image of the target video frame to construct a positive sample pair and a negative sample pair; and adopting the positive sample pair and the negative sample pair to carry out self-supervised training on a preset feature extraction network. The feature extraction network obtained by training through the method is high in matching degree in downstream tasks of target detection and segmentation.

The invention discloses a metro user path identification method, which is used for matching and identifying metro user paths and comprises the following steps: constructing a metro station-base station fingerprint database through road test data and user MR data; establishing a subway user identification model by using the personal track of the current user, the historical matching result set and the subway station-base station fingerprint database; identifying and judging whether the current user is a subway user or not by using the subway user identification model, and if the current user is the subway user, performing the next step; and a hidden Markov model is utilized to match subway user paths. According to the subway user identification method and device, subway user identification is achieved by extracting the user track features and utilizing the random forest identification model, the whole features are more complete based on the current track features and in combination with the historical data features and the fingerprint database features, and the algorithm precision of the random forest identification model is improved.

The invention provides a legal text type case retrieval method and system based on a pre-training language model. The method comprises the steps: arranging information of a legal text type case to be retrieved into data information comprising a main sentence and a retrieved sentence according to the text data of an original legal main sentence and the text data of a retrieval pool, and taking the data information as the input data of model training; performing word segmentation processing and invalid part-of-speech screening on a main sentence and a retrieved sentence in the input data, and obtaining final data with key information based on an artificially constructed crime name table positioning function; carrying out position vector calculation on the data with the key information, and determining the position relation between the data; and utilizing the trained pre-training language model to retrieve a legal text type case related to the query main sentence case. Effective text features are reserved to the maximum extent, the text length is reduced, meanwhile, it is guaranteed that text semantic information is not damaged, and the proportion of key features is enhanced. In terms of data, the precision and the performance of the model are essentially improved.

The invention relates to a lumbar vertebra image classification and recognition system and device based on a multi-modal image and a medium. The system comprises a data acquisition module, a key information extraction module, a multi-modal image detection module and a classification and recognition module. The data acquisition module is used for acquiring a case text of a patient and multi-modal image data in the same period, and respectively sending the case text and the multi-modal image data to the key information extraction module and the multi-modal image detection module; the key information extraction module is used for carrying out key information extraction on the received case text and sending the extracted case key information to the classification identification module; the multi-modal image detection module is used for performing registration and preliminary positioning detection on the acquired multi-modal image data, and sending a preliminary positioning detection result to the classification identification module; and the classification and identification module is used for carrying out accurate segmentation and identification on a focus area in the lumbar vertebra image according to the case key information and the preliminary positioning detection result. The method can be widely applied to the field of image recognition.