1183results about How to "Implement classification" patented technology

The invention discloses a zero sample image classification method based on combination of a variational autocoder and an adversarial network. Samples of a known category are input during model training; category mapping of samples of a training set serves as a condition for guidance; the network is subjected to back propagation of optimization parameters through five loss functions of reconstruction loss, generation loss, discrimination loss, divergence loss and classification loss; pseudo-samples of a corresponding unknown category are generated through guidance of category mapping of the unknown category; and a pseudo-sample training classifier is used for testing on the samples of the unknown category. The high-quality samples beneficial to image classification are generated through theguidance of the category mapping, so that the problem of lack of the training samples of the unknown category in a zero sample scene is solved; and zero sample learning is converted into supervised learning in traditional machine learning, so that the classification accuracy of traditional zero sample learning is improved, the classification accuracy is obviously improved in generalized zero sample learning, and an idea for efficiently generating the samples to improve the classification accuracy is provided for the zero sample learning.

The invention relates to an intelligent individuation video advertisement pushing method which comprises the steps that advertisement putting site image information is collected and stored; the advertisement putting site image information is subjected to face detecting; face detecting comprises the steps that various human faces are recognized in the advertisement putting site image information, and face images of various human faces and the human face number in image information in a certain time period are obtained; the advertisement putting site image information is subjected to face tracking; face tracking comprises the steps that the process that a certain human face watches an advertisement in the advertisement putting site image information is tracked, so that the time interval of advertisement watching of the human face in the advertisement putting site image information is obtained; according to face tracking results, human faces of different people are subjected to gender identification and age estimation, the gender and age information of each person is obtained, and according to the gender and age information, the advertisement audience is sorted; and a recommending decision algorithm is used for generating divided-period advertisement recommending lists for various advertisement terminals.

The invention, which belongs to the technical field of remote-sensing image classification, relates to a multi-mode-characteristic-fusion-based remote-sensing image classification method. Characteristics of at least two modes are extracted; the obtained characteristics of the modes are inputted into an RBM model to carry out fusion to obtain a combined expression of characteristics of the modes; and according to the combined expression, type estimation is carried out on each super-pixel area, thereby realizing remote-sensing image classification. According to the invention, the characteristics, including a superficial-layer mode characteristic and a deep-layer mode characteristic, of various modes are combined by the RBM model to obtain the corresponding combined expression, wherein the combined expression not only includes the layer expression of the remote-sensing image deep-layer mode characteristic but also includes external visible similarity of the superficial-layer mode characteristic. Therefore, the distinguishing capability is high and the classification precision of remote-sensing images is improved.

The invention relates to an entity relationship extracting method of the Zang language. The method comprises the following steps: extracting training linguistic data from the Zang-Chinese text linguistic data information; constructing a Zang word vector model; acquiring an entity relationship characteristic vector from the Zang word vector model; using the entity relationship characteristic vector as an input to construct an entity relationship classification model based on a neural network; and applying multiple layers of characteristic extractions to the entity relationship characteristic vector, thereby finally acquiring a Zang language entity relationship classification. The extraction of the Zang language entity relationship is achieved by constructing the Zang word vector model, researching and solving lexical semantic characteristics and sentence characteristic vector expression methods of the Zang language entity relationship, and further constructing the Zang language entity relationship classification model, accordingly increasing the accuracy in the Zang language entity relationship classification, and providing technical supports and services to the researches in the fields of the Zang language knowledge mapping, question-answering system, information extraction, information search, and the like.

The invention provides a method and a system for extracting a Chinese event. The method comprises the following steps of: performing phrasing, word-splitting, entity identification and analysis for syntax and dependence relationship on a text with a to-be-extracted event in turn; marking the words meeting an extracting condition as candidate triggering words, according to internal structures of the words; filtering the triggering words meeting a filtering condition according to the probability, the word class and the internal structures of the words; extracting the triggering words by utilizing the maximum entropy identifying model and obtaining the reliability of each of the triggering words; dividing the triggering words into a consistency processing training set and a consistency processing testing set according to the reliability of each of the triggering words; utilizing a maximum entropy classifier to extract the triggering words from the consistency processing testing set; and utilizing a maximum entropy classifying model to classify the triggering words, thereby obtaining an event set. According to the method and the system provided by the invention, started from the characteristics of Chinese, the internal structures of Chinese words and the semantic consistency of the Chinese words in sections and chapters are comprehensively considered and analyzed, so that the property of extracting the Chinese event is increased.

The invention brings forward a traffic label identification method of vehicle-mounted laser scanning point cloud data. The method comprises the following steps: 1, point cloud real-time preprocessing; 2, point cloud structure feature obtaining; 3, multi-scale Markov random field point cloud clustering; and 4, traffic label identification. The method has the following advantages: "surface-shaped", "linear-shaped" and "scattered" features of point clouds are reinforced, differences between points are enhanced, and while under-segmentation is avoided, reasonable segmentation of the scale of a traffic label component can be rapidly realized; classification and identification of a traffic label can be conveniently realized from point cloud data of deficiency of a part of the traffic label, caused by ground object shielding or ground object self-occlusion; the method can effectively satisfy the requirements for rapid extracting, monitoring and identifying city components at present, can be conveniently promoted to the industry of manned or unmanned navigation and barrier avoiding based on computer vision, helps a driver to perform navigation and decision-making under complex road conditions, and effectively reduces the traffic accident probability.

The invention provides a road zebra crossing automatic extraction method based on vehicle-mounted laser scanning point cloud, and relates to public traffic road zebra crossings. According to the method, global positioning system data for recording vehicle positions and tracks in real time is used for extracting a plurality of cross sections from the vehicle-mounted laser scanning point cloud data, and the road and non-road classification is realized through detecting the elevation mutation of road shoulders of the roads in the scanning line data; then, the three-dimension road data is converted into an intensity characteristic image with space distribution characteristics, the laser scanning point normal distribution characteristics are utilized for dynamically cutting the road zebra crossings, the GPS (global positioning system) track data is used again for calculating the linear morphology closed operation direction and size, and the extraction of the road zebra crossings is realized. Through the cross section subdivision on the vehicle-mounted moving scanning data, and the three-dimension road surface data detection is converted into the detection of the elevation mutation of the road shoulders of the roads in the two-dimension profile for realizing the road and non-road classification. Compared with a method of directly processing mass three-dimension data, the method has the advantages that the calculation quantity is small, and the efficiency is high.

The invention relates to the technical field of Internet application, in particular to a transport-layer-characteristic-based traffic classification method for the Internet application. The method comprises the following steps of: acquiring a data packet which serves as training data, and constructing a classifier by utilizing transport layer characteristics of a data stream extracted from the data packet; and classifying the data stream extracted from a subsequently acquired data packet by utilizing the constructed classifier. The application type of the stream can be identified only by extracting the transport layer characteristics of the stream without knowing load contents such as a port number, a characteristic field and the like of the data packet. The privacy of a user is uninvolved, and a dynamic-port-number-based or encrypted stream can be identified by the technology. Moreover, the method is highly accurate and integral, novel application can be identified and the user can be prompted to check suspected virus attack streams.

The invention provides a multi-feature-fusion-convolutional-neural-network-based plankton image classification method. The method comprises: lots of clear plankton images are collected and a large-scale multi-type plankton image data set is constructed; a global feature and a local feature are extracted by using an image conversion and edge extraction algorithm; an original feature image, a global feature image, and a local feature image are inputted into a depth-learning multi-feature-fusion convolutional neural network to carry out training, thereby obtaining a multi-feature-fusion convolutional neural network model; and then the plankton images are inputted into the multi-feature-fusion convolutional neural network model and classification is carried out based on a finally outputted probability score. According to the invention, the angle of biological morphology, the computer vision method, and the depth learning technology are combined; and thus the classification accuracy for plankton images, especially large-scale multi-type plankton images is high.

The present invention discloses an image classification and processing method based on different illuminances. The realization method mainly comprises the steps: collecting video images with different illuminances as training sample images; extracting the features of the sample images as experiment data; performing learning training of the feature data, and forming a classifier; realizing the classification of unknown images with different illuminances; outputting unknown image classification results; automatically determining whether the illuminance processing needs to be performed or not according to the unknown image classification results; performing processing of the unknown images requiring illuminance processing; and outputting the final result images. The classification of images with different illuminances is realized, the problem is solved that the threshold determination error is big, the classification precision is greatly improved, and the work efficiency of the monitoring video image processing technology is improved.

The invention discloses a system for automatic classification analysis for websites based on website contents. The system comprises a capture module, a website text content analysis module, a word segmentation module, a feature training extracting module and a website classification module. The feature training extracting module selects a plurality of features words with maximum weights by calculating importance degree, distinction degree and feature keyword weight of every candidate feature word and sorting the candidate feature words according to the feature keyword weights, wherein the feature keyword weights are used as weightings after the normalization of the selected feature words and a website classification vector template is created according to the given sets of the selected feature words and the feature keyword weights. The website classification module is used for generating a feature spatial vector according to the given set of the selected feature words and the weightings which are obtained by the feature training extracting module and identifying the classification of a website by calculating the similarity between the feature spatial vector and the feature spatial vector of the website. The system is capable of effectively solving the problem of network information in a mess and allowing users to searching information for positioning conveniently and accurately.

The invention provides an electrocardiograph detection method based on a multi-scale deep learning neural network. The method includes the steps that a server receives to-be-analyzed electrocardiograph data sent by a client side, wherein the to-be-analyzed electrocardiograph data is collected by a collecting unit of the client side or collected by a collecting device of an intelligent client side and then transmitted to the client side; the server conducts arrhythmia recognition treatment on the to-be-analyzed electrocardiograph data through a pre-established arrhythmia detection model to obtain a recognition result, wherein the arrhythmia detection model is established in advance through a multi-scale deep convolution artificial neural network algorithm training electrocardiograph data sample; the server sends the recognition result to the client side so that the recognition result can be displayed by the client side. By means of the method, the electrocardiograph data can be accurately analyzed, the method is convenient and rapid to implement, and practicability is improved.

The embodiment of the invention provides a garbage classification method, device and system. The problem that the garbage classification effect needs to be improved in the prior art is solved. The garbage classification method includes the steps that identity information of a user is verified and the user passing the verification is registered as a legal user; a garbage bag acquisition request sent by the user and the identity information of the user sending the garbage bag acquisition request are received; whether the user sending the garbage bag acquisition request is a legal user or not is judged, if yes, a garbage bag issuing permission control instruction is sent to a garbage bag issuing machine, and the garbage bag issuing machine issues garbage bags with class labels to the legal user; and control instruction execution feedback sent by the garbage bag issuing machine is received and a unique identification code is correlated with the identity information of the user sending the garbage bag acquisition request and then stored. By the utilization of the garbage classification method, device and system, user garbage classification convenience can be remarkably improved, implementation is convenient, and application and popularization are easy.

The utility model discloses a integrated method based on multi-level set evolution and high resolution remote sensing image partition, classification and change inspection, which is characterized in that (1) image preprocessing (radiation, registration and filtering); (2) the multi-level set evolutional partition and classification model, after registration, the GIS data determines the initial profile of each level set function and performs the partition and classification to the first phase image; (3) the model described in the (2) is still adopted, and the initial profile of each level set function is optimized, increment type partition and classification is adopted for the second to T phase; (4) the objective after partition is used as unit, the ith and (i+1)th two adjacent phase image classification results are compared to determine the change area; (5) return back to (3) until the partition, classification and change inspection of all T phase image are finished. The utility model has the advantages that: compared with the traditional pixel-oriented K value method, the classification and inspection precision are improved, The utility model is applicable for the change inspection of sequence remote sensing image and has wide application in hazard monitoring and land resource investigation.

The invention discloses a band steel surface defect feature extraction and classification method, and belongs to the fields of mode recognition and image processing. The band steel surface defect feature extraction and classification method comprises the steps: extracting a reference sampling size chart of a band steel surface defect sample database; obtaining a reference sampling image, and constructing a gradient size and direction co-occurrence matrix; by aiming at a defect inner area of the reference sampling image, constructing a grayscale size and direction co-occurrence matrix; generating a feature vector sample training library; trimming a training sample set and extracting a multiplying factor by a method of combining K-nearest neighbour with R-nearest neighbour; improving a classifier by using a multiplying factor of the trimmed sample; obtaining a multi-class classifier model; according to the reference sampling size chart, converting the defect test sample into a reference sampling image, then extracting a 25-dimensional feature quantity, inputting the 25-dimensional feature quantity into the multi-class classifier model, and finishing the defect automatic recognition. According to the band steel surface defect feature extraction and classification method, the scale and rotation are not changed, the influence by other adverse factors is restrained, and recognition efficiency and accuracy are improved.

The invention relates to the image processing and image classification technology, and provides an image classification technical scheme with high efficiency and accuracy; the zero sample image classification method based on category attribute migration learning comprises the following steps: building a multi-class classifier model based on attributes to determine whether a tested sample has certain attribute, wherein the model is built according to an indirect attribute prediction IAP; learning the classifications from which the attribute of the tested sample is immigrated, thus learning attributes with finer grain; carrying out transfer learning so as to realize the zero sample image classification. The IAP training phase refers to the multi-class classifier; in a test phase, the method can test the probability of the tested sample belonging to various label classifications, thus obtaining the prediction meaning characteristic of the tested sample; finally the method can use the prediction meaning characteristic to determine the unknown classification to which the tested sample belongs. The method is mainly applied to image processing.

The invention discloses a small-sample polarized SAR ground feature classification method based on a deep convolutional twin network, and mainly solves a problem that a conventional method is low in classification precision because the number of polarized SAR data mark samples is smaller. The method of the invention comprises the steps: 1), inputting a to-be-classified polarized SAR image and a real ground object mark of the to-be-classified polarized SAR image, and carrying out the Lee filtering; 2), extracting an input feature vector from the filtered to-be-classified polarized SAR data, andcarrying out the dividing of a training sample set and a test sample set; 3), carrying out the combination of each two samples in the training sample set, and obtaining a sample pair training set; 4), building the deep convolutional twin network, and carrying out the training of the deep convolutional twin network through the training sample set and the sample pair training set; 5), carrying outthe classification of the samples in the test set through the trained deep convolutional twin network, and obtaining the classes of ground features. According to the invention, the method expands thetraining set under the twin configuration, achieves the extraction of the difference features, enables the classification precision of a model to be higher, and can be used for the target classification, detection and recognition of a polarized SAR image.

The application discloses a user data analysis system based on big data in communication field, comprising a data gathering assembly used for acquiring data of mobile users from other network elements of a mobile network core network, and transmitting the acquired user data to a data processing and analysis assembly; the data processing and analysis assembly used for analyzing the user data of the corresponding type from the data gathering assembly according to a preset user data analysis model, and transmitting an analysis result to a data storage assembly for storage; a data service assembly used for extracting the required analysis result from the data storage assembly according to a data analysis requirement of a value-added service platform, and pushing the analysis result to the value-added service platform and an APP client; and the data storage assembly used for storing the acquired user data and the analysis result. The application also discloses a user data analysis method based on big data in communication field.

The invention discloses a classification method and a device of human body posture. The classification method of human body posture comprises the steps of: determining a human body region in an existing frame image; extracting the human body shape so as to obtain the external outline in the human body, then obtaining the information of the position of human head by human head detection; and then determining that the human body posture is in vertical standing-up status when the head part is right in the middle of the external outline of the human body; and further calculating the aspect ratio of the external outline of the human body when the head part is located at one side of the external outline of the human body, and then determining the types of human body posture according to the aspect ratio of the external outline of the human body, thereby achieving automatic identification and classification to the human body posture, without human intervention, thus making up the detects thatthe prior art does not provide a method for classifying the human body posture.

The invention provides a method for automatically identifying and counting white blood cells. The method comprises the following steps of: preprocessing, color space conversion, coarse segmentation, checking of white blood cells, fine segmentation, characteristic parameter extraction, identification and classification, result output and the like. The method can be used for automatically identifying and classifying the white blood cells in blood by extracting a group of cell characteristic parameters and utilizing a support vector machine, and has the advantages of high classification efficiency, good effect, high accuracy, high stability and high robustness.

The invention discloses an audio frequency rapid classification method based on content, comprising the following steps: pretreatment and character extraction are carried out on audio frequency data, and a series of speech feature vectors are obtained; the speech feature vectors are converted into speech content sequences which are represented by a basic acoustic model unit AU and the speech content sequences comprising classification key words are recognized according to a created word list; the confidence score of every speech content sequence comprising the classification key words is calculated; whether the confidence score is within a set threshold distribution range is judged, and if yes, recognizing results of the speech content sequences comprising the classification key words are reserved; the recognizing results are counted, and the audio frequency data which accord with set conditions are classified according to the set of a classification task on the classification key words. The method fully utilizes content information and can realize complex classifications conveniently and process a plurality of audio data files in a parallel way. The efficiency of the file classification processing is high.

The invention discloses a power quality disturbance classification method based on a sparse automatic coded depth neural network. The method comprises the following steps of preprocessing a power quality disturbance original data set, inputting a preprocessed data sample into a stacked sparse automatic encoder to carry out automatic feature learning to obtain a feature expression, a step of transmitting an extracted feature expression to a softmax classifier for fine adjustment and further adjusting network parameters to obtain a trained classification model, and a step of inputting the a testsample into the classification model for power quality disturbance classification. According to the method, nine power quality disturbance signals including two kinds of composite disturbances can beaccurately identified, and the method has good robustness.

The invention discloses a remote sensing image ground object classification method based on superpixel coding and convolution neural network, using adaptive superpixel coding and double channel convolution neural network. The remote sensing image ground object classification method based on superpixel coding and convolution neural network includes the steps: utilizing a superpixel algorithm to perform image pre-segmentation; using a cluster method to merge neighboring and similar superpixel blocks, setting the size of the taken blocks, constructing three double channel convolution neural networks with different input size; inputting samples with different taken block size into the corresponding network; using the convolution neural networks to extract the data characteristics of two sensors respectively; merging the extracted characteristics for classification; and according to the size of the merged pixel block, determining the size of the taken blocks of the samples, and realizing adaptive selection of the utilized neighborhood information. The remote sensing image ground object classification method based on superpixel coding and convolution neural network can realize adaptive selection of the utilized neighborhood information to enable the neighborhood information to realize positive feedback effect and preferably utilize the neighborhood information to send the samples to different networks according to the neighborhood information so as to enable the samples with similar distribution to enter the same network, thus effectively improving the classification accuracy.

The invention discloses an UAV onboard multi-target detection tracking and indication system and method, and belongs to the technical field of target detection tracking and indication. The invention discloses an UAV onboard multi-target detection tracking and indication system comprising a multi-target detection and tracking system and a multi-target laser indication system. The multi-target detection and tracking system comprises an infrared camera, a visible light image sensor and a high-speed parallel image processing and tracking feedback control circuit. The multi-target indication systemincludes an integrated laser, a fast reflector, a fast reflector control module, and a laser control module. In order to improve laser indication accuracy, the UAV onboard multi-target detection tracking and indication system also includes a laser pointing control system. The invention also discloses an UAV onboard multi-target detection tracking and indication method based on the UAV onboard multi-target detection tracking and indication system. The UAV onboard multi-target detection tracking and indication system and method realize multi-target all-weather detection tracking and high-precision stable laser indication under the condition of a UAV onboard platform.

The embodiment of the invention discloses a method and a device for classifying audio files. The method comprises the following steps of pre-classifying music, and obtaining a spectrogram of each type of music; for a target audio file to be classified, obtaining a spectrogram of the target audio file; according to the similarity of the spectrogram of the target audio file and the spectrogram of each type of music, determining the type of the target audio file. The method disclosed by the embodiment is used for classifying the audio files according to the spectrograms.

The invention discloses a composite fault diagnosis method and device based on a deep decoupling convolutional neural network. The method comprises the following steps: 1, data set collection and calibration: respectively collecting vibration acceleration signals under single and composite fault experiment conditions of a rotary machine to obtain a plurality of samples, respectively forming a training set and a test set, and respectively setting category labels; 2, building a one-dimensional deep convolutional neural network model; 3, constructing a decoupling classifier with a multi-label output characteristic under a keras framework; 4, training a network model, and finally obtaining an optimal deep decoupling convolutional neural network model; and 5, performing intelligent diagnosis and classification on the composite fault, and outputting to obtain a real-time diagnosis result of the composite fault. According to the method, on the premise that only a single fault signal is used for training the deep decoupling network model, each single fault feature in the composite fault signal is extracted, and decoupling and classification of the composite faults are achieved through a decoupling classifier.

The invention discloses an optical coherent tomographic image retinopathy intelligent testing system and a testing method. A current acquired retina image is mainly determined by an ophthalmology doctor by means of naked eye observation, and large-scale popularization is not facilitated. According to the system, a deep learning concept is used as a technical core; a migration learning strategy isutilized; a convolutional neural network algorithm in a deep learning model is used for establishing a classifier, thereby realizing classification of retinopathy; furthermore an image segmenting algorithm is used for realizing focus extraction and retina layering, thereby obtaining specific information of a pathology position in a picture and quantification information of shape parameters, and generating a related diagnosis report for further diagnosis by the doctor. The optical coherent tomographic image retinopathy intelligent testing system and the testing method have advantages of fillingin gaps in pathology intelligent identification and accurate positioning in a current optical coherent chromatography imaging system, effectively reducing working intensity of the doctor, and furtherpromoting clinical application and technical development of the optical coherent chromatography imaging system in ophthalmology disease diagnosis.

The invention discloses a vision inspection device for appearance defects of rollers, and relates to the technical field of vision inspection devices. The device comprises a first conveying belt, a second conveying belt, a first image device, a first image processing system and a first inferior-quality product removing device and further comprises a rotating disc, a rotating disc driving device, roller sucking devices, a second image capturing device and a second image processing system. The first conveying belt conveys rollers to be inspected, the second conveying belt conveys qualified rollers, and the first image capturing device, the first image processing system and the first inferior-quality product removing device inspect the appearance defects of the upper sides of the rollers and remove unqualified inferior-quality products; and the rotating disc, the rotating disc driving device, the roller sucking devices, the second image capturing device and the second image processing system are used for inspecting the appearance defects of the lower sides of the rollers and remove the unqualified inferior-quality products. Vision inspection can be comprehensively carried out on the rollers in a no-dead-corner manner, and meanwhile, the unqualified inferior-quality products are removed; and automation and intelligence of the whole process are achieved, and precision and efficiency of roller inferior-quality product inspection can be improved.

The invention relates to a differential evolution random forecast classifier-based photovoltaic array fault diagnosis method. The method comprises the steps of firstly, collecting photovoltaic array voltages under various working conditions and currents of photovoltaic strings, and performing identification on various working conditions by different identifiers; secondly, determining a quantity range of decision trees in a random forest model by adopting an out-of-bag data-based classification misjudgment rate mean value; thirdly, performing global optimization on the quantity range of the decision trees by utilizing a differential evolution algorithm to obtain an optimal decision tree quantity value; fourthly, substituting the calculated optimal decision tree quantity value into a randomforecast classifier, and training samples to obtain a random forecast fault diagnosis training model; and finally, performing fault detection and classification on a photovoltaic array by utilizing the training model. According to the method, the model training speed can be greatly increased while the optimal model classification accuracy is ensured, so that the fault detection and classificationof the photovoltaic power generation array are realized more quickly and accurately.