37 results about "Adaptive kernel" patented technology

An image descreening process first smoothes the image, where smoothing is accomplished by applying a convolution with a low pass filter (LPF) kernel, which is a parameter to the descreening function. Using the smoothed image, a determination is made for each pixel for which pixels around it participate in the modified filter. For a current pixel, a window is considered having the size of the LPF kernel, with the current pixel at the center. A threshold T1 which is given as a parameter, is used to mark the pixels in the current window. Considering a pixel in the window, if for all color components the difference between this pixel value to the center pixel value is less than T1 in absolute value the pixel is marked with a 1. Otherwise, the pixel is marked with a 0. Finally, an adaptive version of the LPF is applied. If the number of pixels marked with a 1 in the window is less than a third of the kernel size, the original pixel value is restored. Additionally, for a color component for which there is a small change in values within the original (non-smoothed) window (i.e. the difference between the maximal value to the minimal value in this component is less than another threshold T2), the value of this color component is restored. If these conditions do not hold, a new value for each component is determined. To be the convolution of the original window, the LPF kernel is masked with the 0 / 1 markings from the second step. That is, the modified convolution uses an adaptive kernel which is identical to the LPF kernel in the locations marked with one, but has zero entries in the locations marked with zero.

The invention relates to a scale adaptive kernel correlation filter fast target tracking method. According to the scale adaptive kernel correlation filter fast target tracking method, separation of afront background is promoted and a scale estimation module is added through introducing a two-dimensional variable bandwidth Gaussian window into a KCF framework, scale variations of a target are estimated according to a matching result of feature points of an image block, and whether the target is sheltered is judged by combining displacement prediction based on the scale variations. Once the target is sheltered, the user considers that no scale conversion occurs at the moment and template update is not performed, further the tracking precision is enhanced in this way. The scale adaptive kernel correlation filter fast target tracking method has the advantages of simple and fast calculation, and can significantly improve the calculation efficiency on the basis of compatible scale adaptation.

The invention discloses an air compressor monitoring diagnosis system and method adopting an adaptive kernel Gaussian hybrid model, and relates to the field of air compressor control technologies. The system comprises a site equipment layer, an equipment control layer and a management and monitoring layer. The site equipment layer is composed of PLCs200, sensors, air compressors, actuators and a water pump, and with the PLCs200 as slave stations, control over the site equipment layer is completed. The equipment control layer comprises an upper computer and a PLC300, with the PLC300 as a master station, the whole air compressor system is controlled through a variable-structure adaptive PID controller based on a support vector machine, and the upper computer monitors the air compressor system. The equipment control layer is in communication with the management and monitoring layer through the industrial Ethernet, and then remote monitoring and data transmission of the upper computer are achieved. The Gaussian hybrid model and the kernel principal component analysis method are integrated in the fault diagnosis method adopted in the upper computer, optimal kernel function parameters are solved through the iterative optimization method, and the purpose of distinguishing different mode data is achieved. The air compressor monitoring diagnosis system and method have higher diagnosis precision and higher practical value.

The invention discloses an adaptive kernel function and instantaneous frequency estimation-based fault arc detection method of a photovoltaic system. The adaptive kernel function and instantaneous frequency estimation-based fault arc detection method comprises the steps of acquiring a signal xt by a time window with TNCT length, performing non-linear frequency modulation wavelet transform to obtain a iteration time-frequency diagram of the xt, and constructing characteristic quantity, used for judging moment when frequency spectrum energy is increased, according to a frequency component in theselected iteration time-frequency diagram; and obtaining a corresponding matrix distribution form of the xt in the time-frequency domain by adaptive optimal kernel time-frequency distribution after an abrupt changing point where the energy is increased is found out, performing square sum on the matrix according to time dimension to obtain a line vector, selecting a plurality of frequency bands, performing frequency-dimensional integral operation on each frequency band to obtain a plurality of characteristic values input into a trained naive bayesian model, and judging a state of the photovoltaic system in a current time frame. By a plurality of effective time-frequency characteristics, a fault arc in the photovoltaic system is accurately identified, meanwhile, no misoperation under various types arc working conditions also can be ensured, and the photovoltaic system safely and stably run.

A method of computed-tomography and a computed-tomography apparatus where a interpolation kernel width is adaptively determined as a function of the distance from the x-ray source to the reconstruction pixel. The width of the kernel is the projection of the reconstruction pixel on the detector. The method can be implemented in the channel direction. The method can also be implemented in the segment direction, or in the channel and segment directions at the same time. Backprojection is performed using the adaptive kernel width and may by used with helical and circular scanning, and with cone-beam or fan beam x-ray CT.

A processor for enhancing an input luminance signal including: a circuit for calculating a chroma edge value associated with the input luminance signal; a circuit for measuring a luminance gradient associated with the input luminance signal; a peaking filter for processing the input luminance signal; a gain adjustment circuit having a gain function that is adjustable relative to the calculated chroma edge value, the gain adjustment circuit being operable to adjust the magnitude of the output of the peaking filter wherein: in a first range of luminance gradients, the output is an attenuated version of the input; in a second range of luminance gradients the output is directly proportional to the input; in a third range of luminance gradients, the output is inversely proportional to the input; wherein the enhanced luminance signal is the sum of the input and output of the peaking filter.

The invention discloses a method and device for recognizing a short speech speaker. The method comprises the following steps: after an input training short speech signal is preprocessed, a Mel frequency Cepstral coefficient is extracted as a training feature vector, and an adaptive kernel likelihood fuzzy C-means clustering algorithm is used for performing clustering analysis to establish a speaker speech reference model; after the input test short speech signal is preprocessed, the Mel frequency cepstral coefficient is extracted as a test feature vector, distance between the test feature vector and the speaker speech reference model is calculated, and the identity of the short speech speaker is identified according to the distance. Via the method and device for recognizing the short speech speaker disclosed in the present embodiment, the Mel frequency Cepstrum coefficient is extracted as a feature, and the feature and the adaptive kernel likelihood fuzzy C-means clustering algorithm are used for performing the clustering analysis to establish the speaker speech reference model; after an execution model is matched, the identity of the short speech speaker is recognized, recognitionaccuracy is improved, and practical application requirements are met.

The method is applicable to the technical field of computers. The invention provides a training method, a classification method, a system, equipment and a storage medium of a capsule network. In the method, a capsule network is constructed by a convolution layer, a primary capsule layer, a polymerization capsule layer and a category capsule layer; in the polymerization capsule layer, affine transformation is utilized to obtain an intermediate vector, routing processing is utilized to obtain a secondary vector, and under the condition that the average distance between the secondary vector and each intermediate vector is the shortest, a routing parameter corresponding to the routing processing is dynamically updated based on an iterative routing mechanism of adaptive kernel density estimation. Therefore, an iterative routing mechanism of adaptive kernel density estimation can be utilized to provide an explicit iterative objective function, and the number of iterations is flexibly determined, so that the capsule network training effect and the multi-label text classification effect are better.

The invention relates to a multi-feature fusion and scale adaptive kernel correlation filter tracking method, which obtains initial information of a target. The FHOG feature of a fusion gradient histogram and the CN feature of color attribute after dimension reduction of a target area are obtained. The thirty-one-dimensional FHOG feature, the post-dimension reduction two-dimensional CN feature andone-dimensional gray feature are fushed by linear fusion so that totally thirty-four features are used as the final feature map after fusion. On the basis of the fusion feature map, the translationalposition of the target is predicted by using the kernel correlation filter. After the translation position is predicted, scale filter is added to predict the scale change of the target. After predicting the translational position and scale of the target, two filter templates are updated by linear interpolation method and tracked until the last frame. Through the combination of feature fusion andscale filter, the method has good robustness in the case of distortion, scale change, illumination change, background similar disturbance and so on in the process of target tracking.

The invention discloses an unmanned aerial vehicle visual target tracking method based on scale self-adaptive kernel correlation filtering, which comprises the following steps of selecting a trackingtarget, calculating to obtain the color and gradient initial probability density of a first frame of the tracking target, and training a classifier and detecting the central position of the target byusing the kernel correlation filtering algorithm for the first frame of data; establishing a one-dimensional kernel correlation filter from the second frame to detect the change of the target scale, and calculating kernel correlation filtering by using a convolution theorem; constructing a similarity function by utilizing the current target feature and the initial feature, if the similarity is smaller than a set threshold value, considering that the target identification is inaccurate or the target is lost, entering global search, otherwise, representing that the target is identified and tracked, and obtaining target position information; and sending the position information of the tracking target to an unmanned aerial vehicle flight control system in real time to control the position of the unmanned aerial vehicle. According to the method, the problem of fixed tracking scale of a kernel correlation filtering algorithm is optimized, and the tracking precision of target characteristicsis effectively improved.

A method of computed-tomography and a computed-tomography apparatus where a interpolation kernel width is adaptively determined as a function of the distance from the x-ray source to the reconstruction pixel. The width of the kernel is the projection of the reconstruction pixel on the detector. The method can be implemented in the channel direction. The method can also be implemented in the segment direction, or in the channel and segment directions at the same time. Backprojection is performed using the adaptive kernel width and may by used with helical and circular scanning, and with cone-beam or fan beam x-ray CT.

The invention relates to a chemical process fault detection method based on adaptive kernel principal component analysis. The method comprises the following steps: S1. constructing an offline model according to historical data to obtain an Euclidean distance between kernel principal component load vectors in the historical data and a kernel principal component T of the historical data; S2. performing online monitoring based on online data, obtaining DVKPC, NDVKPC and AKPC of the online data in sequence in combination with the Euclidean distance between the kernel principal component load vectors in the historical data, and then calculating a T2 statistical magnitude and a control limit of the online data based on AKPC; and S3. judging whether the T2 statistical magnitude exceeds the control limit, if so, indicating a fault, or otherwise, returning to step S2. Compared with the prior art, by adaptively selecting the kernel principal component of an online data sample, the mutation information can be effectively focused to prevent the mutation features from being scattered or flooded, thereby reducing the fault detection false alarm rate and improving the fault detection performance.

The invention discloses a crowd counting method for a real scene. The method is characterized by adopting a geometric Gaussian adaptive kernel function to generate a real crowd density map as a real value to guide training; inputting the crowd image into the convolution operation of the local receptive field, fusing the generated adaptive feature maps of different scales, inputting the fused feature maps into a deep neural network model, and obtaining the number of people in the scene through regression solution and an integral method according to the obtained estimated density map. Compared with the prior art, the method is simple and convenient, strong in real-time performance and high in accuracy of people counting, can quickly obtain a high-quality crowd density map by adopting a lightweight network structure, does not lose too much statistical precision, occupies a small memory space of model parameters, and is particularly suitable for various scenes with high real-time performance requirements, such as attendance systems, crowd monitoring systems and the like.

The invention discloses an image processing method, and the method comprises the steps: firstly inputting an image L polluted by noises; secondly carrying out the interpolation of the image L, and obtaining the initial estimation of the image L and the first-order derivative of the image; thirdly constructing the covariance matrix of all pixels in the image through employing the first-order derivative calculated at step B; fourthly constructing an adaptive kernel function based on the covariance matrix; fifthly carrying out the adaptive sharpening of the adaptive kernel function; finally achieving the interpolation through employing a weight matrix after adaptive sharpening, and obtaining a final image. The method achieves a purpose that an image reconstruction image is not liable to be affected, and the reconstruction result is not liable to be polluted by noises. Moreover, the method effectively solves problems of image edge blur and noises.

The invention discloses a reliability simulation method related to failure of multiple components of a typical mechanism of an aero-engine. The method comprises the following steps: firstly, determining an initial state of a Markov chain based on an active learning kriging model, and then identifying all most probable failure domains in a multi-failure performance function according to Markov chain Monte Carlo simulation; then constructing an adaptive kernel density estimation important sampling function through samples in the failure domain, taking important samples in the adaptive reduced sample pool as candidate samples, and adding new sample points through a learning function to update the proxy model until the proxy model approaches a real performance function; and finally, calculating a failure probability by utilizing the finally updated proxy model. The Markov chain initial state determination method and the adaptive sample pool reduction strategy provided by the invention are combined with the active learning Kriging proxy model, so that the modeling precision is improved, the number of candidate samples is reduced, and the calculation precision and the calculation efficiency are relatively high.

The invention relates to a quantitative classification calculation method for quantitatively describing the demands of different positions in an urban park-and-ride facility attraction demand range and belongs to the urban traffic planning field. The method comprises the following steps that: a virtual analysis network of a park-and-ride facility attraction demand range area is constructed according to demand point samples and geographic coordinate positions which are obtained through investigation; the distribution density of sampled demand points is calculated through using a kernel function in probability theory; and the total demand density of each position in the facility attraction demand range is calculated and classified through using an adaptive kernel density analysis theory and a numerical value superposition method. The quantitative classification calculation method can be utilized to quantitatively analyze demand intensity distribution of each point at different directions and different distances in the park-and-ride facility attraction demand range, and can classify the demands in the facility attraction demand range through the total demand density, and has a favorable role of reference in the determination and analysis on distribution characteristics of the park-and-ride facility attraction demand range, and provides a new method for park-and-ride facility demand forecast.

The invention provides a tracking method based on dual-model adaptive kernel correlation filtering, which comprises the following steps: initializing the position of a pre-estimated target, calculating a Gaussian tag, and establishing a main feature model and an auxiliary feature model; extracting HOG features to serve as features of a main feature model, extracting deep convolution features to serve as features of an auxiliary feature model, and setting initialization parameters; calculating a response layer of the pre-estimated target by utilizing the main characteristic model, and obtainingan optimal position and an optimal scale of the pre-estimated target by the response layer through a Newton iteration method; if the maximum confidence response value max of the response layer corresponding to the optimal scale is greater than an empirical threshold u, determining a pre-estimated target position, and updating the main feature model; if max is smaller than or equal to an empiricalthreshold u, stopping updating the main feature model, expanding a search area, extracting CNN features of a target pre-selected area, performing dimensionality reduction on deep CNN features by using a PCA technology, estimating a new target position by using the dimensionality-reduced CNN features, and updating an auxiliary feature model until thevideo sequence ends.

The invention discloses a pattern recognition method and device. The pattern recognition method includes the steps of carrying out feature extraction on input training data; after clustering analysisis conducted through an adaptive kernel possibility fuzzy C-means clustering algorithm, creating a reference model and saving the reference model into a reference model database; conducting same feature extraction on input testing data to obtain a testing feature vector, and executing pattern matching on the testing feature vector and each reference model in the reference model database to obtaina pattern recognition result. The pattern recognition device includes a training unit and a recognition unit. According to the scheme of the embodiment, the reference model is created after the clustering analysis is carried out through the adaptive kernel possibility fuzzy C-means clustering algorithm (AKPFCM), adaptive adjustment of the parameter of a kernel function can be achieved according tothe distribution condition of the feature data, better clustering of the feature data is achieved, and the clustering performance is effectively improved.

The invention discloses an interactive network modeling method of complex electromechanical system in process industry based on adaptive symbol transfer entropy, the symbolic common parameters of timeseries are obtained on the basis of multivariate space reconstruction, the probability density and distribution of the original time series are estimated by using the adaptive kernel density estimation method, and divide the sequence into equal probability, by obtaining the best number of symbols and dividing intervals, coarse-grained symbolic representation of the original sequence is implemented, in order to improve the accuracy of the measurement of interaction information between variables, the symbolic sequence of monitoring variables is analyzed by transfer entropy, and the net information transfer quantity is calculated, so as to obtain the basic parameters needed for system interaction network modeling, and establish the network model reflecting the interaction mechanism of the actual system bottom layer. The network model will provide a basis for system state assessment, fault propagation analysis and diagnosis decision-making, so as to improve the scientific and intelligentdecision-making level of complex electromechanical systems in process industry under complex operating conditions.

The invention provides a fatigue crack assessment method based on adaptive kernel density estimation auxiliary particle filtering. The method comprises the following steps of: determining a key assessment part, constructing a state equation and an observation equation of the fatigue crack random expansion model; and constructing a nonlinear state space model to describe an evaluation process of fatigue crack propagation, and performing simulation calculation of a crack propagation state on the nonlinear state space model of fatigue crack propagation of the rotorcraft structure based on a self-adaptive kernel density estimation auxiliary particle filtering method. The invention corrects a re-sampling step by using a kernel density estimation method under the condition that a state model isuncertain, so that the resampled particles maintain diversity and improve the accuracy of the filtering estimation result. An adaptive kernel bandwidth selection scheme is proposed by using particle weights to obtain better kernel density estimation results and applied to the particle filter framework. The estimation accuracy of fatigue crack propagation is improved, and the safety of the rotorcraft structure is improved.

Image processing system for generating a multidimensional adaptive oriented filter to be applied to the point intensities of a d-dimensional image, comprising analyzing means with means (5, fi) to estimate at each image point a probability measure (Fi) of the presence of a type of feature of interest and a weighting control model (10) issuing a weighting control vector (11, VC) constructed from said probability measure, for the user to control synthesized adaptive kernels at each image point; and synthesizing means for generating the filter kernels at each image point adapted to the type of the features of interest, whose filtering strength is controlled by the weighting control vector. The system may comprise a selection unit (40) for the user to select synthesizing means for generating “pre-mixing filtering means” comprising combining means (30, XH) dependent on the type of the image features having inputs for the weighting control vector (11, VC) and the image data [I(x)] and having an aspect for weighted adaptive kernels (35, H) adapted to the type of the image features to produce the filtered image signal [H(x)], and / or “post-mixing filtering means” comprising both isotropic and anisotropic filtering means [15, gi)] applied independently of the type of the image features, whose outputs (Gi) are combined at each image point and adapted using the weighting control vector (11, VC) to produce the filtered image signal [G(x)].

The invention discloses a chemical process fault diagnosis method for adaptive kernel principal component analysis. A suitable kernel function is selected; a kernel matrix and a corresponding covariance matrix are calculated; the covariance matrix is calculated to obtain a feature value; different thresholds are set according to the condition that whether a feature value is zero; the feature valueexceeding a threshold is selected as a kernel principal component model; a calculation statistical magnitude and a control limit are compared to determine fault occurrence. In addition, a sliding window with the proper length is added based on the kernel principal component analysis method based on a threshold and a kernel principal component model is extracted adaptively based on a threshold method; data in each sliding window are processed in a targeted manner and the factor of the fault occurrence is diagnosed, so that the fault diagnosis of the chemical process is processed adaptively.

The invention relates to a ground temperature observation data analysis method based on an adaptive kernel density estimation algorithm, and belongs to the field of ground temperature observation dataanalysis. According to the method, a self-adaptive algorithm is introduced on the basis of a traditional fixed window width kernel density estimation algorithm, namely, an adaptive coefficient is introduced into a window width parameter, the influence caused by the sparse degree of the sample observation value can be effectively reflected, then the adaptive algorithm is improved, the window widthparameter is replaced with the optimal window width, the result obtained under the ground air temperature observation data meets the requirement that all mean square errors are minimum, and the improved adaptive method is completely suitable for the ground air temperature observation data.

The invention discloses an air compressor monitoring diagnosis system and method adopting an adaptive kernel Gaussian hybrid model, and relates to the field of air compressor control technologies. The system comprises a site equipment layer, an equipment control layer and a management and monitoring layer. The site equipment layer is composed of PLCs200, sensors, air compressors, actuators and a water pump, and with the PLCs200 as slave stations, control over the site equipment layer is completed. The equipment control layer comprises an upper computer and a PLC300, with the PLC300 as a master station, the whole air compressor system is controlled through a variable-structure adaptive PID controller based on a support vector machine, and the upper computer monitors the air compressor system. The equipment control layer is in communication with the management and monitoring layer through the industrial Ethernet, and then remote monitoring and data transmission of the upper computer are achieved. The Gaussian hybrid model and the kernel principal component analysis method are integrated in the fault diagnosis method adopted in the upper computer, optimal kernel function parameters are solved through the iterative optimization method, and the purpose of distinguishing different mode data is achieved. The air compressor monitoring diagnosis system and method have higher diagnosis precision and higher practical value.

The invention provides a fatigue crack evaluation method based on adaptive kernel density estimation assisted particle filter, the method includes: determining the key evaluation components, constructing the state equation and observation equation of the fatigue crack random growth model, constructing the nonlinear state space model description In the evaluation process of fatigue crack growth, the simulation calculation of the crack growth state is performed on the nonlinear state space model of the fatigue crack growth of the rotorcraft structure based on the adaptive kernel density estimation assisted particle filter method. Aiming at the condition that the state model is uncertain, the present invention uses the method of kernel density estimation to correct the resampling step, so that the resampled particles maintain diversity, improves the accuracy of the filter estimation result, and uses the particle weight to propose an adaptive Based on the kernel bandwidth selection scheme, better kernel density estimation results are obtained and applied to the particle filter framework, which improves the estimation accuracy of fatigue crack growth, thereby improving the safety of the rotorcraft structure during use.

The invention discloses a remote sensing image pixel-level classification method based on an adaptive convolution kernel convolutional neural network, which is used to solve the technical problem of poor adaptability of the existing remote sensing image pixel-level classification method. The technical solution is to first calculate the density and distance value of the data points, then adaptively select the clustering center as the convolution kernel, and finally add the learned convolution kernel to CNN to train the softmax layer of the network, and perform remote sensing on the trained network Image pixel-level classification. The present invention uses the improved clustering algorithm MCFSFDP based on fast searching and finding density peaks, clusters to obtain adaptive convolution kernels, and substitutes them into the CNN structure based on pre-trained convolution kernels. Compared with the CNN structure based on the K-means clustering artificially setting the clustering category pre-learned convolution kernel, the adaptively learned convolution kernel can effectively represent the characteristics of the digital shorthand data information and improve the pixel-level classification effect of the remote sensing image .

The invention relates to a human body tracking method based on a self-adaptive kernel function and mean value shifting. The human body tracking method includes two stages, the first stage is a learning stage, a set of training samples of human body walking is firstly read, human body prospect shapes are mapped to be coordinates in a low-dimensional space through a dimensionality reduction algorithm, a low-dimensional human body shape space is obtained, the human body prospect shapes are then recovered through an interpolation reconstruction algorithm, and parameters, capable of mapping from a low dimension to a high dimension, of the interpolation reconstruction algorithm can be obtained. The second stage is a tracking stage, a human body optimum kernel shape in a video frame is searched for in the low-dimensional human body shape space, and the human body in the video frame is tracked by using a mean value shifting algorithm. Compared with the prior art, the human body tracking method improves the shape of the kernel function in a traditional mean value shifting algorithm, so that the shape of the kernel function is not fixed and changes in a self-adaptive mode according to changes of shapes of the tracked human body, histogram modeling and matching of the kernel function are further performed in the high dimension space, and therefore the performance of a human body tracking technology is improved.