62 results about "Linear embedding" patented technology

The invention proposes a rolling bearing fault prediction method based on a continuous deep belief network. The method comprises the following steps: first, extracting the time-domain characteristic quantity of vibration signals of a rolling bearing; then, fusing the extracted time-domain characteristic information using a locally linear embedding method, so as to define a new comprehensive monitoring index for better quantitative evaluation of performance degradation of the bearing; training a continuous restricted Boltzmann machine step by step to construct a continuous deep belief network prediction model; and using a genetic algorithm to optimize the structure of the continuous deep belief network so as to further improve the prediction precision. The prediction method is reliable in result, has good real-time performance, is simple and feasible, and is suitable for rolling bearing fault prediction.

Certain embodiments of the present invention provide a system and method for temporally aligning a plurality of cardiac image sequences. The method includes performing a locally linear embedding algorithm on a first set of cardiac image sequences and on a second set of cardiac image sequences. A graphical representation is created for the first set of cardiac image sequences and the second set of cardiac image sequences. A determination is made whether the first set of cardiac image sequences and the second set of cardiac image sequences were generated from a similar point of view. If a similar point of view is found, the first graphical representation and the second graphical representation are aligned using a minimization function. If a similar point of view is not found, the graphs are aligned with a template and then aligned with each other using the minimization function.

The invention relates to an early failure identification method based on SILLE (Supervised Increment Locally Linear Embedding) dimensionality reduction. The method comprises the following steps of: acquiring characteristic information of early failures of complex rotating machinery from two action domains including a time domain and a frequency domain, establishing time-and-frequency domain characteristic set vectors representing different failure characteristics comprehensively, automatically reducing the high-frequency time-and-frequency domain characteristic set vectors into low-dimension characteristic vectors with higher distinction degrees by virtue of SILLE, and inputting the low-dimension characteristic vectors into a classifier for classification and decision, thus obtaining early failure identification results of test samples. The early failure identification method can be used for giving full play to the superiority of time-and-frequency domain characteristic sets on comprehensive failure characteristic excavation, the superiority of an SILLE technology on information reduction and the superiority of the classifier on mode identification, and guaranteeing the automation, high precision, rapidness and universality of the early failure identification method for the rotating machinery.

The invention provides a method for monitoring the process of a fused magnesium furnace based on an improved supervised kernel locally linear embedding method, and relates to the technical field of fault monitoring and diagnosis. The method includes the steps of mapping sample data X to a high dimensional feature space [phi](X) by using a kernel function; selecting the number of k neighbor points through a MKSLLE (Modified supervised kernel locally linear embedding) algorithm, and adding a regular term when constructing a reconstruction weight matrix; performing dimensionality reduction for an objective function composed of a KPCA-combined global preserving features and local preserving features, and obtaining a mapping matrix from a high dimensional data space to a low dimensional feature space and a coefficient matrix through approximate calculation; and constructing a Hotelling T2 statistic and an SPE statistic and determining control limits thereof. According to the invention, abnormalities and faults can be monitored online in real time in the working process of a fused magnesium furnace, the accuracy of fault monitoring is effectively improved, the occurrence of false alarms and false negatives is reduced, the property loss is reduced, and the personal safety of working staff is guaranteed.

A data classification method. The method includes: providing data mapped in a first space; mapping the data into a second space using locally linear embedding to generate mapped data; applying a synthetic minority over-sampling technique (SMOTE) to the mapped data to generate new data; and mapping the new data into the first space.

The invention provides an En-ULLELDA-based (ensemble-unified locally linear embedding and linear discriminant analysis) method of multi-view model recognition. The method comprises the following steps: firstly, selecting a vehicle ROI (region of interest); secondly, extracting the high-level information that can reflect the model as features through Gabor wavelet transform; then, carrying out the dimension reduction of features on the basis of a novel manifold learning algorithm En-ULLELDA, so as to transform the features to a low-dimensional space; and obtaining the corresponding models from the features of a test sample by a classifier during the recognition. According to test results, the invention can effectively handle the problem of multi-view model recognition on urban roads; and the invention can eliminate the effects of parameter variation on the recognition results by employing the integration technology, so that the invention has higher robustness.

The invention discloses a crowd abnormal behavior identification method based on SURF stream and LLE sparse representation. The method is mainly used for solving the problems that crowd characteristics extraction in the complex scene is not accurate, the behavior characteristics dimension in the crowd behavior detection is high, the data volume is large and the local manifold structure of the characteristics is unstable. The method comprises the following steps: (1) inputting a test video sample and a training video sample, creating a SURF stream field and acquiring characteristic point motion vector information; (3) respectively classifying the characteristic point vector information of each frame in the test video sequence and the training video sequence into 216-dimensional characteristics, and enabling the video sequence to form behavior characteristic set; (3) utilizing a locally linear embedding sparse representation formula to classify the characteristic set, and obtaining a sparse representation coefficient; (4) computing a residual error and judging the category of the test video. The crowd abnormal behavior identification method can effectively remain the local manifold structure of the test sample, and improve the judgment capability to the sample.

The invention discloses a face recognition method combining dual-tree complex wavelet transform and discrete wavelet transform. The method comprises the following steps of: carrying out feature extraction on an input face image by using a method combining dual-tree complex wavelet transform and discrete wavelet transform; carrying out dimension reduction on an extracted feature vector X by using a supervised locally linear embedding method; carrying out cosine similarity calculation on the feature vector of the tested face image and the feature vectors of training set images; and setting the input image and a training image with highest similarity into one group so as to obtain a face recognition result. By utilizing the method combining dual-tree complex wavelet transform and discrete wavelet transform, multidirectional and abundant face feature extraction can be realized and dimension reduction can be realized quickly, thereby realizing accurate and efficient face recognition.

The invention relates to a depth motion map-scale invariant feature transform-based gesture recognition method. The method mainly comprises the following three parts: in the motion data acquisition aspect, an original depth image provided by the Kinect somatosensory technology is adopted as the input variable of a gesture recognition system. In the human body gesture feature construction aspect, a depth motion map-scale invariant feature transform-based extraction method is adopted, and data obtained after feature extraction are subjected to dimension-reduction treatment through the supervised locally linear embedding (SLLE) method. In this way, a gesture motion characteristic quantity is represented. In the gesture classifier recognition aspect, a support vector machine based on a discriminant is adopted to realize the sample training and modeling process of the characteristic quantities of a depth image sequence. Meanwhile, an unknown gesture is classified and predicted. The method of the invention can be adapted to different lighting environments, and is stronger in robustness. The method can also efficiently recognize gesture sequences in real time. Therefore, the method can be applied to the real-time gesture recognition field of man-machine interaction.

The invention discloses an image noise reduction system and method based on K-SVD (Singular Value Decomposition) and locally linear embedding, and particularly relates to a signal sparse representation and reconstruction technique based on dictionary learning and an image noise reduction technique based on manifoid learning. A K-SVD method is adopted to be a frame as the signal sparse representation and reconstruction technique based on the dictionary learning, the locally linear embedding used as a constrain condition is added into a target function while the signal sparse representation is solved, so that the relation of the decomposed sparse coefficients is strengthened, the influence of the random noise on the sparse coefficients is overcome, and the image noise reduction effect better than that of the original K-SVD method is obtained.

The invention relates to a method for monitoring faults in a smelting process of a multimode magnesia electrical smelting furnace. According to the method, a historical normal data set of different working modes in the smelting process of the multimode magnesia electrical smelting furnace is obtained; a subspace separation model based on a mass nucleus locally linear embedding method is created; T2 statistic control limit of global public subspace of historical normal data and SPE (squared prediction error) statistic control limit of local special subspace of each of different working modes are calculated; a new data set in a current working mode is acquired in real time; T2 statistic of global public subspace of new data and SPE statistic of corresponding local special subspace in the current working mode are calculated; and if the T2 statistic of the global public subspace of the new data exceeds the T2 statistic control limit of the global public subspace of the historical normal data, or the SPE statistic of the corresponding local special subspace of the new data exceeds the SPE statistic control limit of the local special subspace of the historical normal data in the working mode, the current working mode in the smelting process of the multimode magnesia electrical smelting furnace has possibility of fault occurrence.

The invention provides an aircraft engine transient state fault detecting method which includes first extracting marked features from rotating speed data of an aircraft engine based on a Mahalanobis adaptive Hessian locally linear embedding feature extractor; then utilizing an adaptive BP neural network classifier to analyze and determine whether faults are generated in the transient state according to the result of the feature extraction; finally processing the detected faults through a diagnosis system to generate a maintenance guide. The method solves the problem that the existing aircraft fault detection is only confined to engine data collection in the steady state. The method adopts combination of the Hessian locally linear embedding (Mahalanobis adaptive Hessian locally linear embedding) and adaptive BP (Back Propagation) neural network. The method can effectively improve the fault detection performance of the aircraft engine in the transient state.

The invention discloses a bearing variable working condition fault diagnosis method based on Hessian locally linear embedding (HLLE). The method may improve the stability of a bearing fault characteristic and a fault diagnosis capability under a variable working condition. The method comprises: acquiring an inherent manifold characteristic of a manifold topological structure in a bearing original vibration signal by using a HLLE method; performing fast Fourier transform (FFT) on the inherent manifold characteristic to obtain a spectrogram, extracting, from the spectrogram, the amplitude at the bearing fault character frequency and the amplitudes at special frequency such as the second harmonic frequency, the third harmonic frequency or the like in order to form a bearing fault characteristic vector; and on the basis of the acquired fault characteristic, classifying the bearing fault states by using an information geometry-based support vector machine (IG-SVM) so as to achieve a variable working condition fault diagnosis capability. The invention provides a bearing with a fault characteristic extracting scheme capable of effectively resisting to working condition interference by using a fault characteristic extracting method based on the HLLE-FFT. The method guarantees the accuracy of bearing fault diagnosis and has good practical engineering application value.

The invention discloses a handwriting digital recognition method and a handwriting digital recognition device. The method comprises the following steps of acquiring a target figure and an image data set in a target image which are to be recognized, wherein the image data set comprises at least one piece of image data; performing vector conversion on each piece of image data in the image data set to obtain vector data corresponding to each piece of image data, wherein the vector data forms a vector data set; performing dimension reduction on the vector data set by using a local sparse linear embedding dimension reduction method based on an orthogonal matching tracing algorithm to obtain a dimension reduction vector data set; and according to the dimension reduction vector data set, identifying a digital image corresponding to the target figure in the target image. Not all of computational solutions obtained by the orthogonal matching tracing algorithm in the method are zero. Compared with the prior art, the method and the device have the advantages that local sparsity can be obviously embodied, stability is realized, and the accuracy of the digital image which is finally recognized is relatively high.

The invention provides a method for establishing a facial feature database in a video image. On the basis of a locally linear embedding (LLE) algorithm, a face manifold unit in the video image is learned, high-dimensional image space of the image is reduced to low-dimensional image space, and a relationship between an internal low-dimensional structure and high-dimensional observation data is searched; and clustering is carried out on the low-dimensional image space by using a K-means clustering algorithm, so that similar face image data are close, a face of each clustering center is used as a representative face image, a redundant example is removed, and a representative face feature base is established finally. In the LLE algorithm, the embedding dimension of the low-dimensional space is determined according to a Sammon coefficient. According to the invention, facial feature base establishment is realized in the video image; and some representative images are selected from many training images and are used for follow-up algorithm learning. Therefore, while the face identification rate is kept well, the training time and the occupied internal storage are reduced substantially, so that the processing speed increases.

The invention discloses a dynamic function connection local linear embedded feature extraction and brain state classification method and system. The dynamic function connection local linear embedded feature extraction method comprises the following steps: collecting functional magnetic resonance imaging data in a resting state; after preprocessing, extracting an average time sequence signal of each brain region through a brain template; calculating and constructing a dynamic function connection matrix by using a sliding time window, and taking the dynamic function connection matrix as to-be-processed high-dimensional brain dynamic description original data; and carrying out manifold learning on the dynamic function connection matrix by using a local linear embedding algorithm to obtain a low-dimensional manifold subspace model, and extracting a feature part in the low-dimensional manifold subspace model to obtain a dynamic function connection local linear embedding feature. For the dynamic function connection local linear embedded feature extraction method, the feature extraction method is rapid in calculation and ideal in data processing effect, can construct significant crowd feature description, does not depend on the absolute value of the amplitude of an imaging signal, is migratable between different MRI machines, is excellent in classification and discrimination performance, and can conveniently utilize a machine learning model to realize brain state classification.

Certain embodiments of the present invention provide a system and method for temporally aligning a plurality of cardiac image sequences. The method includes performing a locally linear embedding algorithm on a first set of cardiac image sequences and on a second set of cardiac image sequences. A graphical representation is created for the first set of cardiac image sequences and the second set of cardiac image sequences. A determination is made whether the first set of cardiac image sequences and the second set of cardiac image sequences were generated from a similar point of view. If a similar point of view is found, the first graphical representation and the second graphical representation are aligned using a minimization function. If a similar point of view is not found, the graphs are aligned with a template and then aligned with each other using the minimization function.

The embodiment of the invention provides a method for processing a head-related impulse response data set. The method comprises the following steps: a processor groups the head-related impulse response (HRIR) data set according to an angle of elevation and / or a horizontal angle; the processor selects the HRIR sample point with a minimum reconstruction error value in the grouped HRIR data set as a high-dimensional HRIR sample point through locally linear embedding (LLE); the processor reduces the dimension of the HRIR sample point to a low-dimensional space to obtain low-dimension HRIR sample points; the processor adopts sectional straight line fitting to select characteristic HRIR sample points in the low-dimension HRIR sample points; and the processor combines the characteristic HRIR sample points into a three-dimensional matrix according to an angle-of-elevation sequence and / or a horizontal angle sequence. The embodiment of the invention solves the problem of excessive errors of a characteristic HRIR selection method.

The invention discloses a natural gas pipeline leakage detection method, and relates to the technical field of pipeline leakage detection. According to the natural gas pipeline leakage detection method, the problem of high false alarm rate of a pipeline leakage detection system is solved. The method comprises the following steps of acquiring a sound wave signal by using an acoustic sensor; optimizing a variational mode decomposition algorithm by utilizing a gull algorithm; performing noise reduction pretreatment on the sound wave signal by using an optimized variational mode decomposition threshold denoising method to obtain a denoised sound wave signal; respectively extracting time-frequency characteristics of the sound wave signal to construct a high-dimensional feature vector matrix, reducing the dimension of the high-dimensional feature vector matrix by using a local linear embedding algorithm, and extracting a sensitive feature vector beneficial to classification; and searching anoptimal penalty factor and a kernel function by utilizing the gull optimization algorithm to optimize the performance of a least square support vector machine, taking the sensitive characteristic vector as a training sample of the least square support vector machine, and taking the collected sound wave signal as a test sample to detect whether a natural gas pipeline leaks.

The invention discloses a method based on multiple correspondence and K. The method comprises the following steps: (1) according to the obtained traffic accident data set, selecting and classifying the variables that affect the occurrence of traffic accidents; (2) Through the statistics of the number of categories of each variable and the corresponding accident number in the database, the variablecategories of the merged abnormal values are screened to obtain the accident data table; (3) processing the obtained accident data table to obtain a binary index matrix; (4) Multiple correspondence analysis is carried out by taking accident type as the variable representing accident characteristics, and the coordinates of multiple correspondence analysis of each variable type are obtained; 5) uselocal linear embedding algorithm to reduce that dimension of the variable category coordinate obtained from the multi-correspondence analysis of the accident data, and obtaining the LLE reduced dimension coordinate; (6) Use of K-Means clustering algorithm is used to cluster the variables, and the results are analyzed according to the clustering results. According to the clustering result, the invention comprehensively probes into the causes of traffic accidents from multiple dimensions, and not only analyzes two-dimensional correspondence analysis diagrams.

The invention discloses an image super-resolution reconstruction method based on representational learning and neighbor constraint embedding, and problems of inaccurate feature extraction and fixed size of neighbors are solved. The method includes main steps: pre-processing a group of training sample images, and constructing a low-resolution image block dictionary and a high-resolution image block dictionary; inputting a low-resolution test image, and performing partitioning and extracting features of the low-resolution test image; calculating the Euclidean distance between the features, and searching K neighbors of image blocks; and constructing an adaptive constraint function, obtaining k adaptive neighbors via neighbor constraint, obtaining a final high-resolution image by employing a locally linear embedding method, and accomplishing the image super-resolution reconstruction. According to the method, the characteristic of deep sparse self-coding network learning is employed, neighbor selection is accurate, the size of the neighbors is selected in an adaptive manner, the reconstruction image quality is effectively enhanced, detailed information is improved, and the method is applicable to super-resolution reconstruction of various natural images.

Disclosed are a face recognition system and method based on a locally distributed linear embedding algorithm. The system comprises an image collecting module, an image preprocessing module, a feature analysis module, a dimensionality mapping module, a recognition database and a recognition module. The method comprises the steps that (1) face images are obtained; (2) the obtained face images are stored; (3) preprocessing is conducted on the obtained face images; (4) feature analysis and extraction are conducted on the preprocessed face images, and face features in a high-dimension space are mapped into a low-dimension space; (5) the unknown face images are recognized according to the spatial relationship between the corresponding face images in the low-dimension space and images in the recognition database. According to the face recognition system and method based on the locally distributed linear embedding algorithm, the features of the obtained face images are analyzed and a dimensionality conversion program is operated according to the obtained face images so that during recognition, points in a similar neighborhood can be found out in the high-dimension space automatically according to the face images, after the face images are mapped into the low-dimension space, the relative order relation cannot be changed, and therefore the face recognition accuracy can be effectively improved.

The present invention provides a belly image reconstruction device. The device is reconstructed in the free breathing, a Golden-Angle Radial 3D rapid scanning sequence is employed to obtain a liver image. The locally linear embedding (LLE) in the manifold learning method is employed to perform the dimension reduction process of the K space center data, and a respiratory curve is obtained; the data packet is performed according to the respiratory curve, and each group of data is subjected to reconstruction through adoption of a multi-scale low-order recovery theory; and finally, each group of reconstructed data is subjected to registering so as to eliminate the motion artifact and obtain a final image. The belly image reconstruction device does not reject any data with no need for an operator's manual participation, and is high in degree of automation and low in application cost, etc.

Systems and methods for geometric calibration and image reconstruction in computed tomography (CT) scanning using iterative reconstruction algorithms are provided. An iterative reconstruction algorithm can be used to reconstruct an improved image, and then the improved image can be used to adjust inaccurate parameters by using a Locally Linear Embedding (LLE) method. Adjusted parameters can then be used to reconstruct new images, which can then be used to further adjust the parameters. The steps of this iterative process can be repeated until a quality threshold is met.

The invention provides a traffic big data dimensionality reduction method based on manifold learning. The traffic big data dimensionality reduction method comprises the following steps: 1, acquiring spatial-temporal trajectory data of various urban traffic to obtain traffic big data; 2, classifying the traffic big data, uniformly configuring attributes of the traffic big data and loading data into a database; and 3, performing data dimensionality reduction processing on the traffic big data by an LLE (Locally linear embedding) manifold learning method. Through adoption of the traffic big data dimensionality reduction method, the data dimension reduction is performed on high-dimensionality spatial data by a manifold learning algorithm, so that an excessively-large computation amount is avoided; the essential characteristics of the data can be reflected better; and the mining performance of traffic data is enhanced.

A fault detection method of a satellite attitude control system for supervised locally linear embedding relates to telemetry data mining methods for supervised locally linear embedding, solves the problem that the local linear embedding (LLE) algorithm of the existing batch processing mode is difficult to update the database in real time and ensure the accuracy of high-dimensional feature extraction, and comprises the following steps of obtaining high-dimensional original telemetry satellite data, and performing characteristics analysis and preprocessing on the obtained original telemetry satellite data; reducing the dimension of the pre-processed original satellite telemetry data by using the SLLE algorithm to obtain low-dimensional embedded feature information of satellite control systemtelemetry data, and using SPE statistics to complete fault detection. The invention uses the SLLE algorithm to extract high-dimensional data features, combines statistical SPE and T2 to design a fault detection scheme, and finally verifies the effectiveness of the fault detection scheme of the satellite attitude control system through the simulation of satellite telemetry data. The method effectively improves the detection ability of satellite abnormal states and has certain practical application value in engineering.

A data classification method. The method includes: providing data mapped in a first space; mapping the data into a second space using locally linear embedding to generate mapped data; applying a synthetic minority over-sampling technique (SMOTE) to the mapped data to generate new data; and mapping the new data into the first space.

The invention provides a speech recognition method, which comprises the steps of feature extraction, feature optimization and recognition carried out by using a classifier. The step of feature extraction comprises that time-frequency analysis is carried out on speech through adopting multi-scale continuous wavelet transform, Gaussian mixture modeling is carried out on a wavelet coefficient along the direction of a scale axis to acquire a feature parameter GCWT, and then recognition is carried out on the voice; the step of feature optimization comprises that dimension reduction processing is carried out on the feature parameter GCWT by adopting a dynamic weighted locally linear embedding (DWLLE) method. The feature parameter GCWT provided by the invention is better than a traditional feature parameter MFCC, and a dimension reduction effect of the DWLLE method is better than that of LLE (Locally Linear Embedding).

A method and an apparatus for processing an image generates, using a linear embedding operator, a second set of feature points from a first set of feature points of the image, so that a pair- wise nearest neighbor (NN) topological relationship among the first set of feature points is preserved in the second set of feature points. The linear embedding operator is determined by an affinity model comprising a first affinity parameter, a second affinity parameter, and an affinity matrix, wherein a sparsity of the affinity matrix is controlled by the first affinity parameter and the second affinity parameter.