212results about How to "Reduce data dimensionality" patented technology

The invention discloses a GPS (Global Position System) spatio-temporal data increment road network real-time updating and trace matching system, which comprises a data acquisition module, a GPS spatio-temporal data module, a data pre-processing module, a map trace matching module, a road network generating module and an increment road network updating module. According to the system, the acquired data recorded by a GPS positioner and provided with time and space attributes is pre-processed, a map is matched, a road is generated, a newly formed road section is compared with road network data recorded in a database, road network information is updated, and thus road network data is updated. According to the GPS spatio-temporal data increment road network real-time updating and trace matching system provided by the invention, the road network information is compared continuously with incremental data information, so that a real-time capability is realized, road position information is more accurate along with the increment of data, and roads which are not contained in the road network information are updated.

The invention relates to a method for extracting and processing gait feature information and identifying identification when people walk, and aims at reducing interference of external factors such as a complicated background and the like, realizing better adaptivity to reality conditions, more exactly extracting effective information which can reflect the walking feature of a motion human body so as to improve accuracy of gait identification. The technical proposal of the invention provides the method for gait information processing and identity identification based on a fusion feature. The method comprises the following steps: inputting a video sequence, segmenting profile information of a body object in a video image by target detection, synchronously extracting gait feature parameters by adopting boundary center distance and Radon conversion, carrying out corresponding post processing on the obtained feature parameters, taking a support vector machine as a classifier for classification identification, and evaluating the identification effect. The method is mainly applied to identity identification based on the gait feature information.

A hyperspectral image sparse unmixing method based on random projection includes the following four main steps: (1) data are read by a computer under the environment of MATLAB R2008b; (2) the hyperspectral image data and the hyperspectral library data are randomly projected by the computer; (3) a target function for sparse unmixing is constructed, and the split Bregman algorithm is used for optimizing the target function and working out an extremum until reaching convergence and stopping conditions; (4) an appropriate threshold value is set to process a abundance fraction matrix, so that a final abundance fraction graph and end members can be obtained. The hyperspectral image sparse unmixing method based on random projection utilizes a hyperspectral database to choose the end members, and overcomes the defect that the end members worked out by the conventional algorithm cannot strictly correspond to the spectra of pure materials in the standard hyperspectral database; and moreover, the hyperspectral image sparse unmixing method based on random projection uses the random projection technology to carry out dimensionality reduction on raw data, thus achieving the effects of saving memories and reducing the calculation load. The hyperspectral image sparse unmixing method based on random projection realizes rapid quantitative analysis on hyperspectral images, and has practical value and a broad application prospect in the field of hyperspectral remote sensing image analysis.

The invention discloses a driving intention identification method based on an improved HMM and SVM double-layer algorithm. The method includes: step 1, classification of driving intentions of drivers: dividing the driving intentions of the drivers into five categories: emergent left lane-changing, normal left lane-changing, lane maintaining, normal right lane-changing, and emergent right lane-changing; step 2: experiment data acquisition and processing; step 3: offline training of the improved HMM and SVM double-layer algorithm; and step 4: identification of the driving intentions of the drivers. According to the method, a human-vehicle-road system is completely considered, information of vehicles, roads and drivers is acquired, the HMM and SVM double-layer learning model is employed, and the accuracy and the timeliness of a driver lane-changing intention identification model are improved.

The invention discloses a fusion reasoning system and a fusion reasoning method for intelligent tags of news programs, relates to the technical field of news program tags. the intelligent news programtag identification system comprises an intelligent identification actuator, a historical tag library, an internal knowledge base, an internal case library and an analysis reasoner; the intelligent identification actuator executes identification tasks of various news program materials, and basic tag extraction is conducted on video images, voice and text information; Wherein the historical tag library stores materials, metadata and tags; The internal knowledge base is used for supplementing an intelligent recognition result and providing more information for subsequent analysis and reasoning;Wherein the internal case library is a case set established based on a historical tag library; According to the news program tag automatic fusion reasoning system and method, the intelligent recognition method is comprehensively utilized, the internal knowledge base and the internal case base are established based on the historical tag base, automatic fusion reasoning of news program tags is completed, and classification is accurate and efficient.

The invention discloses an fMRI dynamic brain function sub-network construction and parallel connection SVM weighted recognition method which comprises the steps that (1) data are preprocessed; (2) the time series of each brain area is extracted; (3)interested brain areas are selected; (4) dynamic brain function sub-networks of all the brain areas are constructed; (5) all the sub-network classifiers are trained; (6) values are assigned to all sub-classifiers to form parallel connection SVM classifiers; (7) unknown samples are classified. Compared with a traditional static function network, information on the time dimension is added on the constructed dynamic brain function networks; prior knowledge is combined for constructing dynamic sub-networks on different interested brain areas, and the feature dimensions are reduced while useful information is reserved; SVM classifiers of all the sub-networks are trained, the parallel connection SVM classifiers is formed by determining the weight of the sub-classifiers through the recognition rate, the brain areas are integrally weighed and classified, and the classifiers have better robustness.

The invention proposes a spatio-temporal data stream video behavior recognition method based on deep learning. Compared with a traditional deep learning method which utilizes a signal data stream, namely a single video stream, the spatio-temporal data stream video behavior recognition method based on deep learning utilizes spatio-temporal data streams, that is a spatial stream and a temporal stream, the temporal stream recognizes category of a target in a video from a static video frame, the spatial stream recognizes motion of the target from motion components in the video, and classification results of the temporal stream and the spatial stream are integrated to obtain a final behavior category. The spatio-temporal data stream video behavior recognition method based on deep learning recognizes the target and the motion separately, can reduce calculation burden of a neural network, and effectively increases accuracy.

The invention discloses a fault diagnosis method and a fault diagnosis device of a power transformer. The method comprises the following steps: establishing a state characteristic data table based on an in-oil dissolved gas sample with a definite fault type; carrying out normalized treatment on the state characteristic data table and establishing a normalized fault table; calculating based on the normalized fault table to obtain various fault type clustering centers; based on the clustering centers, establishing a state standard spectrum matrix; calculating through an improved main component analysis method to obtain a characteristic value, a characteristic vector and a main component contribution rate; setting a threshold value and correspondingly selecting a main component; and calculating an Euler distance between a sample to be detected and the main component of a state characteristic sample main component and taking a state characteristic sample corresponding to a minimum distance value as a diagnosis result. The fault diagnosis method and device of the power transformer have the following advantages that a state standard spectrum is calculated by utilizing fuzzy clustering, and subject data removal and sample quantity restriction are avoided; meanwhile, the dimension of the data can be reduced and main characteristics for representing fault types are refined; and the accuracy of latent fault diagnosis in the power transformer is effectively improved.

The invention relates to an indoor positioning method based on WiFi, which comprises the following steps of: in an offline stage, acquiring fingerprint vectors of N reference point positions in an indoor positioning area, and storing the fingerprint information of the N reference points into a fingerprint database DB; roughly positioning at an online stage, namely determining a target floor; utilizing the K-means algorithm to carry out clustering analysis on the sub-fingerprint libraries DBjk of the corresponding floors, and further dividing positioning sub-regions; in the real-time positioning stage, firstly, carrying out AP selection, and then using a KNN classification algorithm for determining a sub-region where a target is located; and finally, finding out K nearest neighbors, and estimating the position (x, y) of the target in a weighted average mode. According to the method, for a large-range indoor positioning scene, the intensity information of all APs is reserved; for indoorfloor positioning, an SVM classifier is used, an encoder is added to a classifier model, the data dimension is reduced through introduction of the encoder, redundant information and noise interferenceare effectively reduced, and the classification precision is improved.

The invention discloses a network intrusion detection method. The network intrusion detection method includes: searching network data to construct a test network data set; performing feature extraction on the test network data set by utilizing a kernel principal component analysis method; constructing a training data set, putting the training data set into a support vector machine classifier for training; obtaining feature datasets, obtaining an optimal feature subset from the feature data set by using a genetic algorithm; utilizing a firefly swarm optimization algorithm to obtain the overalllocal optimal feature subset and the optimal support vector machine parameters from the optimal feature subset, processing the training data set according to the overall local optimal feature subset,and inputting the training data set into a support vector machine classifier for classification modeling to obtain a network intrusion detection model. According to the method, the simplicity and convenience of the algorithm are improved, abnormal data can be more effectively found from samples, the detection accuracy of network intrusion is effectively improved, the missing report rate and the false report rate are reduced, and the overall performance of network intrusion detection is improved.

The invention discloses an indoor positioning method based on a CSI (Channel state information) signal. The method has the main processes that firstly, a low-path wave filter is used for performing filtering processing on the collected CSI signal to remove most background noise; secondly, the filtered signal is subjected to PCA (principal component analysis) processing to reach the dimension reduction de-noising effect; thirdly, the processed signal is subjected to people walking signal feature (including signal energy intensity, energy variance, wave peak and wave trough number and phase) extraction; finally, a machine learning algorithm is used for training to obtain an indoor positioning model based on CSI. The indoor positioning method based on the CSI provided by the invention solves the problems that in the multi-path environment, the CSI signal are susceptible to interference; the noise is too great; the precision is low when the method is used for indoor positioning.

The invention relates to a sleep apnea syndrome detection method based on pulse and blood oxygen signals. The method includes: digging characteristic parameters related to SAS (statistical analysis system) from the pulse and blood oxygen signals by means of probability principal component analysis, setting up an SAS detection characteristic matrix, adopting a pouch decision tree for mapping the relation between the characteristic matrix and SAS disease degrees, and setting up an SAS detection model. By means of probability principal component analysis, input data dimensionality is decreased while quality of the characteristic matrix is improved; owing to the pouch decision tree, SAS detection accuracy is improved; by utilization of the sleep apnea syndrome detection method based on the pulse and blood oxygen signals, the SAS disease degrees of testees can be outputted only by inputting the pulse and blood oxygen signals of the testees.

The invention relates to a convolutional neural network adversarial transfer learning method based on Waserstein distance and application thereof and the method comprises the steps: employing a to-be-migrated convolutional neural network to obtain a source domain feature set and a source domain fault judgment set of a source domain mark sample set and a target feature set of a target domain sampleset; and with maximization of a Wasserstein distance between the source domain feature set and the target feature set and minimization of the sum of the Wasserstein distance and a judgment loss valueof the source domain fault judgment set as a target, realizing adversarial migration learning of the convolutional neural network based on a convergence criterion. According to the invention, the Wasserstein distance is introduced into the transfer learning of the convolutional neural network. The maximum Wasserstein distance is used as a target; the distinguishing sensitivity of the features extracted from the two sample sets is improved; and the minimum sum of the Wasserstein distance and the loss value of the source domain fault judgment set is taken as a target, so that the judgment precision of the convolutional neural network is improved, the requirements on sample data and a network structure are low while the fault diagnosis capability is ensured, and the invention can be suitablefor migration among multiple working conditions and is high in practical applicability.

The invention discloses a phase encoding characteristic and multi-metric learning based vague facial image verification method. The phase encoding characteristic and multi-metric learning based vague facial image verification method comprises (1) a training phase, namely, partitioning sampling images and extracting multi-scale primary characteristics of every image block, performing fisher kernel dictionary learning through the above characteristics to generate into partitioning fisher kernel coding characteristics, performing multi-metric matrix learning on the above coding characteristics to generate a plurality of metric matrixes and obtain the metric distance after training samples are performed on multi-metric matrix projection, calculating the average metric distance and variance of positive samples and negative samples to a set and confirming a final classification threshold through a probability calculation formula of Gaussian distribution and (2) a verification phase, namely, partitioning input facial images and extracting multi-scale primary characteristics, generating partitioning fisher kernel coding characteristics, obtaining the final metric distance through the multi-metric matrix and comparing the distance and the threshold to obtain a facial image verification result. The phase encoding characteristic and multi-metric learning based vague facial image verification method has the advantages that the identification rate is high and the universality is strong.

The invention relates to a passive distribution SAR imaging process method based on double-stage multi-resolution reconstruction and belongs to the technical field of SAR imaging signal processing methods. The method comprises the following steps of, firstly, performing filtering and modulation on target echo signals received by an airborne receiver, separating out the target echo signals from different radiation sources and modulating the target echo signals to a base band; secondly, processing various channel signals separated out in the first step through the Fourier reconstruction algorithm to obtain the coarse-resolution images of a target; thirdly, for the coarse-resolution images obtained in the second step, obtaining the high-definition image of the target in every coarse-definition pixel unit among the images through a sparse reconstruction method; lastly, obtaining the high-definition image of a whole scene through data screening and image splicing. The passive distribution SAR imaging process method based on the double-stage multi-resolution reconstruction can solves the grating lobe problem due to sparse sampling and meanwhile has higher computation efficiency compared with traditional sparse reconstruction methods.

The invention discloses an electronic nose feature selection optimization method on the basis of multiple Fisher kernel discriminant analysis. The electronic nose feature selection optimization method comprises the following steps: firstly, acquiring a sample feature matrix; initializing parameters and establishing a fundamental kernel function according to the parameters; then calculating a composite kernel matrix on the basis of a fundamental kernel matrix, calculating a projection of the composite kernel matrix in a high-position feature space, then feeding the projection into a classifier to carry out mode identification to determine a kernel function with the highest identification rate; finally, on the basis of the kernel function, calculating a projection of a new sample matrix in the feature space, using the projection as an electronic nose signal and using the electronic nose signal as an input of the classifier to carry out mode identification. The electronic nose feature selection optimization method has the obvious effects of solving the problem of poor data discrimination after high-dimension projection is implemented by a single kernel function method, solving the problem of redundancies between sensors, optimizing a sensor array, reducing data dimensions and improving the identification rate of the electronic nose signal so as to provide beneficial guide for a doctor to select a suitable treatment method.

The invention discloses a sparse spectrum signature extraction method for a voice lie detection system. The sparse spectrum signature extraction method for a voice lie detection system includes the following steps: 1) extracting a Mel frequency spectral coefficient and a wavelet packet frequency band cepstrum coefficient of a voice signal, and integrating the Mel frequency spectral coefficient with the wavelet packet frequency band cepstrum coefficient to form a cepstrum characteristic; 2) training the cepstrum characteristic by means of a K-singular value decomposition algorithm to obtain a mixed overcomplete representation dictionary; and 3) on the mixed overcomplete representation dictionary obtained in the step 2, utilizing an orthogonal matching pursuit algorithm to perform sparse coding on the cepstrum characteristic to obtain the sparse spectrum signature. The sparse spectrum signature extraction method for a voice lie detection system can make up the deficiency of providing medium and high frequency band information by a traditional Mel frequency spectral coefficient, and can also solve the redundancy problem of a non-linear fusion parameter set to reduce the calculating complexity of a classification model.

The present invention discloses a fake user detection method based on hierarchical clustering. The method comprises the following steps: backing up a website user data table, mapping a user unique identification field into a string composition mode, and pre-classifying massive data according to different composition modes; calculating the string similarity between the elements in each classification; and setting a suitable threshold to carry out hierarchical clustering on each category, and then discovering fake accounts in group hidden in the massive registration data. Experiments show that the method proposed by the present invention is effective and the method has less dependence on data dimension and data characteristics than the prior art; and the method proposed by the invention can be applied to the fake user detection under the current big data environment.

The invention relates to an HU invariant moment and support vector machine-based garment style identification method. The method comprises the steps of preprocessing a garment image to obtain an outer contour of a garment; extracting an HU invariant moment characteristic of the outer contour of the garment; and performing support vector machine (SVM)-based garment style identification. The preprocessing of the garment image refers to a process that the garment image is subjected to segmentation processing, a 8-adjacent connection region with a maximum area is found as a garment region, and internal pore filling is performed on the garment region; the obtaining of the outer contour of the garment refers to a process that the preprocessed garment image is subjected to external edge detection to obtain a contour image of the garment; the extraction of the HU invariant moment characteristic of the outer contour of the garment refers to a process that a 7-order HU invariant moment eigenvector of a contour shape characteristic of the garment is extracted; and the SVM-based garment style identification refers to garment style multi-classification identification performed by adopting an SVM multi-classifier. The method can achieve the identification accuracy of 83%, has a relatively good effect of identifying garment styles with similar contours, has the characteristics of quickness and accuracy, and can be suitable for identification of garment styles in garment images.

The present invention discloses a multi-attribute decision tree power grid stability margin assessment method based on linear discrimination analysis. a key variable discovery model is established based on the offline simulation data and the real-time monitoring data of a power grid to perform effective screening of historical sample data to reduce the data dimensions, a combination relation model among key variables is established to discover the association relation among the variables, extract combination features capable of reflecting important degree contrast of each variable, establish the association relation between the power grid operation state and the transient stability margin, determine the main reasons of system stability level changing, form a concise and accurate knowledge rule base and regulate the decision reference so as to rapidly assess the current stability level according to the system operation state, provide quantification information support for operators' auxiliary decisions and improve the standardization, the rapidity and the adaptive capability of the power grid stability assessment, and therefore the multi-attribute decision tree power grid stability margin assessment method based on the linear discrimination analysis has wide application prospects.

The invention belongs to the field of pattern recognition and provides an intrusion detection method based on an SVM. The intrusion detection method based on the SVM comprises the following steps of preprocessing data, carrying out feature weighting on the data, carrying out feature extraction on the data and carrying out intrusion detection through the improved SVM. The intrusion detection method based on the SVM reduces the influences of secondary factors on the classification performance, lowers dimensionality, reduces data redundancy information, reduces the operand and improves the classification performance and the intrusion detection accuracy of the SVM.

The invention discloses a multi-factor short-term load prediction method based on PCA-DBILSTM (Principal Component Analysis-DBILSTM). Firstly, normalization and One hot encoding are carried out on original input data to obtain a multi-dimensional matrix, principal component extraction is carried out on the multi-dimensional matrix by utilizing a PCA method, and then prediction is carried out by utilizing a DBILSTM network prediction model. Compared with a traditional power load prediction method, the method has the advantages that the average absolute percentage error (MAPE) and the root-mean-square error (RMSE) are both reduced, and the result shows that the method has high prediction precision.

The invention discloses a quality grading and perceptual hash characteristic combination-based unmanned aerial vehicle image retrieval method, and belongs to the technical field of image processing. The method comprises the steps of firstly, performing automatic quality grading and quality label allocation on an unmanned aerial vehicle training image set, extracting perceptual hash codes, and establishing database applications corresponding to images and texts; secondly, setting quality label and perceptual hash code-based data attributes for to-be-retrieved images, like the sub-steps in the first step; thirdly, performing Hamming distance matching on the obtained data attributes of the to-be-retrieved images and images in a database; and finally, allocating a certain weight to a similar image set obtained by matching in combination with similarity and quality labels to establish a weight function, performing resorting according to a weight progressive increase sequence, and outputting an image result. According to the method, image characteristics can be quickly extracted and image quality grading can be effectively finished, so that quick and accurate retrieval of the unmanned aerial vehicle images is realized.

The invention discloses a method and device for removing the sedimentary background under high-dimensional seismic data input. The method comprises carrying out Wheeler conversion on high-dimensionaloriginal seismic data of a reservoir to be measured to obtain high-dimensional Wheeler domain seismic data; carrying out cluster analysis processing on the high-dimensional Wheeler domain seismic datato obtain a cluster result; in the cluster result, respectively calculating the average value of seismic trace data of a single cluster on the same sampling point, taking the average value as the seismic data of the corresponding cluster on the same sampling point, and forming cluster seismic trace data; carrying out linear superposition processing on the cluster seismic trace data to obtain a linear superposition vector, and determining sedimentary background data according to the linear superposition vector; and carrying out Body operation on the high-dimensional Wheeler domain seismic dataand the sedimentary background data to obtain lithology seismic data of the reservoir to be measured. By means of all the embodiments, the sedimentary background can be removed accurately and efficiently, and the accuracy of the fine seismic sedimentology research is improved.

The invention relates to a feature selection method and application based on feature identification degree and independence. The method comprises following steps: calculating the importance degree of each feature by measuring inter-class distinguished ability with feature identification degree and measuring correlational relationship between features with feature independence and sequencing in a descending order; and selecting top k features with the importance higher than those of others to form a feature subset with high class-discrimination performance. Differently-expressed gene subsets selected in application of oncogene expression profile data obtain fine time and class discrimination performance. The feature selection method and application based on feature identification degree and independence have following beneficial effects: easy calculations can be made; time complexity is reduced; selection efficiency runs high; and a good reference is provided for clinical diagnoses and judgments of tumors and other diseases.

The invention discloses an electronic nose parameter synchronous optimization algorithm based on an improved quantum particle swarm optimization algorithm. The method comprises performing wavelet transformation on obtained original electronic nose data; then performing weighting treatment of wavelet coefficients; through the improved quantum particle swarm optimization algorithm based on a novel local attractor computing manner, finding out a weighting coefficient corresponding to the highest electronic nose identifying rate, and classifier parameters to obtain a characteristic matrix of electronic nose signals; inputting the characteristic matrix into a classifier for mode identification. The electronic nose parameter synchronous optimization algorithm based on the improved quantum particle swarm optimization algorithm has the advantages of enhancing early-stage ergodicity and later-stage local optimizing capacity of particles, improving the capacity of quantum particle swarms in searching for global optimal values, and especially for wound infection detection, improving the identification rate of an electronic nose, thereby selecting appropriate treatment methods for doctors and providing beneficial guidance for promoting quick recovery of wounds.

The invention provides a short-term wind power forecasting method and system, comprising: collecting forecasting meteorological data; Inputting the forecasted meteorological data into a short-term wind power forecasting model created in advance to obtain a wind power forecasting value; The training data set of the short-term wind power prediction model is obtained by taking historical meteorological data related to the total output power of the wind turbine as modeling data and extracting the modeling data by PCA principal component. The technical proposal provided by the invention improves the operation speed and the modeling accuracy.

The invention discloses a risk driving scene rapid extraction and grading method for intelligent network connection automobile testing. The risk driving scene rapid extraction and grading method comprises the steps that (1) carrying out intelligent network connection automobile driving time sequence data normalization processing; (2) carrying out dimension reduction processing on the intelligent networked automobile driving time sequence data; (3) performing discretization and symbolization processing on the intelligent networked automobile driving time series data; and (4) carrying out rapidextraction and hierarchical processing on the intelligent networked automobile risk driving scene. According to the risk driving scene rapid extraction and grading method for the intelligent networkedautomobile test, various risk driving scenes in the intelligent networked automobile test working condition can be rapidly extracted and graded, and a theoretical method support can be provided for comprehensive construction and refining optimization of a driving scene library in the subsequent intelligent networked automobile test stage.

The invention discloses an electric power daily load prediction method. The method comprises the following steps: 1, ; and establishing a load category set, establishing a key influence factor set, calculating a classification rule between the load category and the key influence factor characteristic value to establish a classification model, analyzing the category to which the key influence factor data of the day to be predicted belongs, training a load prediction model, and calculating a power load prediction result by using the load prediction model. According to the scheme, the problem that a load prediction model is inaccurate due to divergence of load data can be effectively avoided, the accuracy of a prediction result is improved. Meanwhile, the data dimension in the prediction process can be reduced, the calculation efficiency is improved, the power generation requirement and the power selling requirement can be effectively met, economic benefits are guaranteed, and developmentof power market economy is promoted.

The present invention relates to a depth belief network-based neighborhood weighted averaging hyperspectral image classification method. The objective of the invention is to the problem of the increase of operation time of pre-training and fine-tuning of an existing classification classification compared with pure spectral data. The depth belief network-based neighborhood weighted averaging hyperspectral image classification method includes the following steps that: step 1, spatial information extraction is performed on original hyperspectral data, so that the spatial information of the original hyperspectral data can be obtained; step 2, neighborhood weighting is performed on the spatial information of the original hyperspectral data, so that processed hyperspectral data can be obtained, namely, a matrix M is obtained; and step 3, with the matrix M adopted as a training set, a depth belief network is adopted to carry out classification training. The method of the present invention is applied to the image classification field.