298 results about "Local linear" patented technology

The invention discloses a method for clustering gene expression data based on manifold learning, and the method provided by the invention comprises the following steps: acquiring a gene expression data matrix A through an acquisition system, and preprocessing the gene expression data matrix A by using a local linear smoothing algorithm; introducing the preprocessed data matrix A, and constructing a weighted neighborhood figure G in a three-dimensional space; taking the shortest path between two points as the approximate geodesic distance between two points; calculating a two-dimensional embedded coordinate by using an MDS (minimum discernible signal), and mapping the three-dimensional data matrix A to a two-dimensional visual space; and carrying out clustering on the two-dimensional visual space subjected to mapping by using a k-mean clustering algorithm so as to obtain the clustering result. The clustering method has the characteristics of low calculating cost, capability of eliminating high-order redundancies, suitability for pattern classification tasks, and the like; and by using the method disclosed by the invention, the current states of cells, the effectiveness of medicaments to malignant cells, and the like can be discriminated effectively according to the clustering result. The invention also provides a system for clustering gene expression data based on manifold learning.

The invention relates to the field of intelligent traffic, in particular to a dynamic traffic flow prediction method based on space-time correlation. The method comprises the steps that a space-time matrix is established after traffic flow data are preprocessed, an adjacent local linear reconstitution method is used for training the space-time matrix, a set of adjacent and weight values used for prediction are found out, prediction is conducted after non-negative correction, and at last the space-time matrix is updated through a prediction value. The dynamic traffic flow prediction method based on space-time correlation has the advantages that the adaptability is high, the method is suitable for any microwave detection road section; the feasibility is high, the data can be trained and predicted as long as a historical traffic flow database is given; the calculation speed is high, the complexity is low, and the calculation time is in a second level; the prediction precision is high, the randomness and volatility of dynamic data are removed, and the accuracy and reliability of a prediction result are improved; the prediction efficiency is high, multi-step traffic flow prediction of multiple five-minute time periods can be achieved, and high-efficiency short-time and long-time traffic flow prediction can be achieved.

Methods and systems efficiently apply known distortion, such as of a camera and lens, to source image data to produce data of an output image with the distortion. In an embodiment, an output image field is segmented into regions so that on each segment the distortion function is approximately linear, and segmentation data is stored in a quadtree. The distortion function is applied to the segmented image field to produce a segmented rendered distortion image (SRDI) and a corresponding look-up table. To distort a source image, a location in the output image field is selected, and the uniquely colored segment at the same location in the SRDI is found. The look-up table provides the local linear inverse of the distortion function, which is applied to determine from where in the source image to take image texture data for the distorted output image.

An embodiment of the invention provides a visual analysis based mountain railway side slope rockfall detecting method. The method mainly comprises collecting an image of a mountain railway, performing local linear detection and Hough exchange on the image and obtaining an image of a railway area; performing differential treatment and binaryzation segmentation process on the image of the railway image and obtaining a foreground target image; constructing a classifier based on a well-trained deep network and inputting the foreground target image into the classifier and judging whether the foreground target image belongs to the target image containing rockfall or not according to an output result of the classifier. According to the embodiment of the invention, the railway area is identified at first, then the image in the railway area is segmented, the foreground target is detected, and finally the target is classified through deep learning and interference targets are removed. The scheme provided by the embodiment of the invention has advantages of wide detection range, low cost and the like of a video analysis method. At the same time, the rockfall image detection accuracy is improved.

Aspects of the present invention include systems and methods for generating a novel view interpolation. In embodiments, feature correspondences and geometrical contexts are used to find additional correspondences based on the assumption of the local linear transformation. The accuracy and the number of correspondence matches may be improved by iterative refinement. Having obtained a set of correspondences, a novel view image can be generated.

The invention discloses a human face identification method based on manifold learning, and belongs to the technical field of image processing. The method solves the problem of excessive resource consumption of the traditional method for directly processing high-dimension images. The method is combined with two kinds of methods including the nearest characteristic sub space classifier method and the local linear embedding method for realizing the dimension reducing processing on human face images, then, the nearest classifier is adopted for identifying the data subjected to dimension reduction, firstly, the human face image high-dimension data is firstly built, and the human face image samples are stretched into one-dimension vectors in lines; then, the built human face image high-dimension data is subjected to dimension reduction processing, and the low-dimension expression of all obtained human face images is obtained; and finally, the data is embedded into the space at the low dimension. Through the training on the images, the images to be tested are collected in real time, the human face identification is carried out, the method is more reasonable than a local linear embedding method based on Euclidean distance, the identification accuracy is higher, the method has lower operation complexity than a method of directly adopting high-dimension data for identification, and the method is simpler and more convenient.

The invention discloses a method for detecting real-time automotive load. A load detector is used to collect real-time signals inducted by a load sensor and corresponding to the nominal load, the real-time signals transmitted to a detecting center via a on-board terminal and wireless network; a computer processing system of the detecting center metalationobtains the load value of the vehicle according to a fitting curve between the load value and the induction signal by means of a local linear fitting. The method has the advantages of low cost, high performance cost ratio, stability and reliability, which is particularly suitable to be used in application occasions where accuracy of the load of vehicles is not required strictly (ten percent allowable error) while real-time control is required for the load of vehicles, such as the management and dispatching system of a freight fleet.

The invention discloses a method for detecting an unknown malicious code in the technical field of information safety, which can detect the malicious code in a file in advance under the situation that a malicious code library is not updated. The method comprises the following steps: extracting the feature vector of a file in a training set by utilizing a Byte n-grams method; carrying out the dimension reduction to the extracted feature vector of the file in the training set by adopting a local linear embedding algorithm; taking the feature vector after being subjected to dimension reduction as input, training a kernel cover classifier by utilizing a kernel cover learning algorithm; extracting the feature vector of the file in a test set by utilizing the Byte n-grams method again; carrying out the dimension reduction to the extracted feature vector of the file in the test set by adopting the local linear embedding algorithm; inputting a result after being subjected to dimension reduction into the kernel cover classifier for classification; and calculating the classification result and determining whether the file in the test set contains the malicious code. With the adoption of the method, the detection speed of the file is improved, and the advanced accuracy detection of the malicious code is realized.

The invention discloses a fusion-feature-based video vehicle re-identification method and system. The method comprises: a target vehicle and a matching range are determined; according to the target vehicle and the matching range, a color feature similarity degree between a target vehicle image and a to-be-matched vehicle image is calculated by using a color-histogram-based similarity calculation method; on the basis of the target vehicle and the matching range, a similarity degree of coding direction gradient histogram features of the target vehicle image and the to-be-matched vehicle image iscalculated by using a direction gradient histogram feature similarity calculation method based on local linear constraint coding and weighted spatial pyramid; and weighting fusion is carried out on the calculated color feature similarity degree and the coding direction gradient histogram feature similarity degree. According to the invention, on the basis of the weighting fusion of the calculatedcolor feature similarity degree and the coding direction gradient histogram feature similarity degree, the similarity result of vehicle re-identification is obtained; and the accuracy, robustness, anduniversality are improved. The method and system are applied to the image processing field widely.

An image processing system and demosaicing method are provided to calculate estimated missing color sensor values in an image using a linear prediction from the raw color sensor value at the current pixel location. The raw image is divided into regions of sensor values, and the linear relations between color planes for each region are determined by a regression method that calculates the degree to which different color planes co-vary within each region The missing color sensor values per region are calculated as a scaled and shifted version of the raw color sensor values using linear regression coefficients determined from the local linear regression process.

The invention relates to a method for establishing a three-dimensional geometrical structure of a dog left ventricle conduction system based on a local linear embedding method, in particular to a method for mapping a plane image into a curved surface image. The method comprises the following steps to complete a ventricle conduction system: firstly, mapping dog left ventricle three-dimensional geometrical configuration to a plane by an LLE algorithm, and recording as a two-dimensional linear graph of the ventricle conduction system; then, extracting the conduction system in a two-dimensional graph of a real ventricle conduction system, and integrating the conduction system into a new graph; and finally, mapping the integrated new graph to the left ventricle three-dimensional geometrical configuration by using the LLE algorithm again so as to obtain the three-dimensional geometrical structure of the left ventricle conduction system. The invention retains the local linear characteristic of a space structure, has the characteristics of high efficiency and speed and uses real data to reflect the real dog left ventricle conduction system.

The invention relates to a mechanical arm motion control method and system based on visual real-time teaching and self-adaptive DMPS. The method comprises the steps that a teaching object is set, theteaching object is controlled to do demonstration motion, Kinect is used for obtaining a depth map and cooperated with a PnP algorithm to carry out three-dimensional pose positioning and tracking on the teaching object, a space mapping system is built to map the pose of the teaching object to the tail end of a mechanical arm, control information of all joints of the mechanical arm is calculated according to inverse kinematics and sent in real time so as to indirectly control the mechanical arm to move, finally, the demonstration motion information is recorded on line, and the self DMPS algorithm is applied to carry out local linear optimization and learning on the information. Constraint of hardware structures of a mechanical arm and dependency to a complex sensor in a traditional teachingmode are omitted, the teaching hardware cost and teaching difficulty are lowered, safety of the teaching process is improved due to the non-contact characteristic, meanwhile, applicability is high, and the self-adaptive DMPS method enables the whole system to have good anti-interference performance.

The invention relates to a wind turbine generator set bearing fault feature extraction method based on vibration data. The method includes the steps of: 1. using a JADE algorithm to perform blind source separation on an observation signal, so as to obtain a source signal; 2. calculating kurtosis and negentropy of the source signal; 3. calculating a singular value of a source signal envelope matrix; and 4. utilizing a local linear embedding method to extract fault features. The wind turbine generator set bearing fault feature extraction method based on vibration data combines the blind source separation with the local linear embedding method, and is particularly suitable for rotary mechanical equipment such as a bearing; and the method can effectively eliminate noise mixed in a bearing vibration signal process, and separate a fault source signal, thereby providing more accurate information for fault feature extraction.

The invention relates to an identification method of a speaker. The method comprises: a speaker identification model is generated, background voices and target speaker voices are used as training data to obtain a first Gauss mixing-universal background model, a total changing space, a second Gauss mixing-universal background model, and a local linear discrimination analysis model; and a total changing factor of a to-be-identified voice and a posterior probability of the total changing factor are calculated based on the first Gauss mixing-universal background model, the total changing space, the second Gauss mixing-universal background model, and the local linear discrimination analysis model, the local linear discrimination analysis model is inputted to carry out conversion to obtain a low-dimension vector, and the vector is inputted into a rear-end identifier and an identification result is outputted. According to the invention, the discriminating property of speakers is enhanced; and the identification performance of the speaker is improved. Meanwhile, dimensionality reduction of the total changing factor is realized; the identification speed is enhanced; and the practicability is high.

The invention provides a server fan speed adjusting method. The method includes that on the basis of a commonly-used linear or temperature-section-based sectional local linear temperature-duty ratio curve, an offset amount related to temperature increase and decrease speed is added or reduced to correct the curve, so that a nonlinear temperature-duty ratio curve that changes of PWM (pulse width modulation) duty ratio can adapt to temperature change rules better is acquired; speed of a server fan is adjusted according to a relationship, between chip temperature and the PWM duty ratio, determined by the nonlinear temperature-duty ration curve. The server fan speed adjusting method is conducive to better solving the problems of chip cooling and fan denoising.

The invention discloses a penicillin fermentation process fault monitoring method based on MLLE-OCSVM, and relates to the technical field of the fault monitoring of the data drive. The method comprises two phases of off-line modeling and on-line monitoring. The off-line modeling comprises the following steps: firstly processing the three-dimensional data of the fermentation process; using a local linear embedding method (MLLE) in a manifold learning algorithm to execute the feature extraction to an original high dimensional data sample later; and finally using a one-class support vector machine (OCSVM) to execute the modeling construction monitoring statistics, and using a kernel density estimation method to determine the control limit. The on-line monitoring comprises the following steps: processing the newly-collected data according to the model, calculating the statistics and comparing with the control limit, and judging whether the fermentation process is run normally. The method does not need to assume that the fermentation process variable complies with the specific distribution of gauss or non-gauss, and the accuracy rate of the fault monitoring is higher.

A system and a method are disclosed for clustering images of objects seen from different viewpoints. That is, given an unlabelled set of images of n objects, an unsupervised algorithm groups the images into N disjoint subsets such that each subset only contains images of a single object. The clustering method makes use of a broad geometric framework that exploits the interplay between the geometry of appearance manifolds and the symmetry of the 2D affine group.

The invention discloses a prevention and control method of online risk assessment based on wind power fluctuation. The prevention and control method comprises the steps of: predicting wind power output ranges at a future moment t<h+1> and corresponding probabilities by using a Markov chain model based on wind power fluctuation; for each predictive wind power output range, estimating a power grid operating state by using a local linear processing method; acquiring the severity of the corresponding wind power output range by using an obtained result and a severity function capable of reflecting the stability margin of a static voltage; acquiring an operating risk indicator in a computation period by using an acquired result; judging whether the risk indicator exceeds a set threshold value; if so, entering into a prevention and control model; on the premise of not discarding wind, adopting prevention and control to guarantee the power grid operating state change caused by random wind power fluctuation in a stability domain and to reduce an operating risk caused by short-term wind power fluctuation, controlling the risk indicator within the set threshold value; if not, carrying out risk assessment at a next moment.

The invention provides a nuclear power device fault diagnosis method based on local linear embedding and a K-nearest neighbor classifier. The method comprises steps of (1) acquiring operation data of a nuclear power device in steady-state operation and typical accident states as training data; (2) using the mean-variance standardization method, carrying out dimensionless standardization processing on the training data to obtain high-dimension sample data; (3) using the local linear embedding algorithm, extracting low-dimension manifold structures of the high-dimension sample data so as to obtain low-dimension characteristic vectors; (4) inputting the low-dimension characteristic vectors into a K-nearest neighbor classifier to carry out classification training; (5), acquiring real-time operation data of the nuclear power device, and repeating the steps of (2) and (3); and (6) using the trained K-nearest neighbor classifier to make decisions for classification of the characteristic vectors. According to the invention, by taking advantages of the nonlinear manifold learning method in the aspects of characteristic dimension reduction extraction, the provided method is suitable for fault diagnosis of nonlinear data high-dimension systems, and has quite high fault diagnosis accuracy.