173results about How to "Reduce reconstruction error" patented technology

The invention provides a universal circuit breaker mechanical fault diagnosis method based on feature fusion of vibration and sound signals. The method includes steps of 1, collecting machine vibration signals and machine sound signals during an engaging and disengaging process of a universal circuit breaker; 2, adopting an improved wavelet packet threshold value denoising algorithm for denoising; 3, adopting a complementary total average empirical mode decomposition algorithm for extracting a plurality of solid mode function components reflecting state information of engagement and disengagement actions of the circuit breaker from the denoising signals; 4, determining the number Z of the solid mode function components; 5, calculating the energy ratio, the sample ratio and the power spectrum entropy as three types of features; 6, adopting a combination core principal component analysis method for performing dimension reduction on a feature sample with unified three types of features of the vibration and the sound signals and obtaining M principle components; 7, establishing a related vector machine based sequence binary tree multiple classifier model.

The invention discloses a hyperspectral imager and an imaging method based on compressive sensing, mainly solving the problem that the existing hyperspectral imager has high sampling rate and high sensor realization difficulty. The imager comprises a battery of lens, a dispersive device, a spatial light modulator, a linear detector and a peripheral circuit. The acquired linear light source is split in the space through the dispersive device to form the plane light source formed by spatial dimension and spectral dimension. The plane light source converges again in the direction of spatial dimension after being modulated by the spatial light modulator to form the linear light source formed by spectral dimension. The linear detector completes sampling and quantizing. The imaging method is characterized by utilizing the obtained hyperspectral compressive observation vector to obtain the hyperspectral images through grouping and reconstitution. The hyperspectral imager improves the average reconstitution accuracy of each spectrum by utilizing the joint sparse characteristic among the hyperspectral spectra, has the advantages of simple structure and low cost and is suitable for compressive sensing and imaging of hyperspectra.

The invention discloses a deep belief network image recognition method based on Bayesian regularization and belongs to the field of artificial intelligence and machine learning. The deep belief network plays a more and more important role in the field of digital detection and image recognition. The invention provides a deep belief network based on Bayesian regularization on the basis of the network sparsity characteristic and changes of connection weights to solve the problem of overfitting in the training process of the deep belief network. By applying Bayesian regularization to the network training process, balance between error decreasing and weight increasing is effectively adjusted. The classification experiment of a digital script database proves effectiveness of the improved algorithm. An experimental result shows that in the deep belief network, the deep belief network image recognition method can effectively overcome the overfitting phenomenon and improve accuracy of digital recognition.

The invention belongs to the technical field of image super-resolution reconstruction, and discloses a multi-scale residual attention network image super-resolution reconstruction method based on attention. The method comprises the steps of selecting a common image data set as a to-be-experimented image set, dividing the to-be-experimented image set into an image training set and an image test set, and performing image preprocessing, designing a multi-scale residual structure unit module, introducing a channel attention mechanism, and building a multi-scale residual attention neural network model based on channel attention, inputting the preprocessed image training set into a multi-scale residual attention neural network model based on channel attention for model training, and inputting the preprocessed image test set into the trained model for testing to obtain a finally reconstructed high-resolution image. According to the method, a basic unit is enabled to focus on extraction of high-frequency information, important feature map information in a channel is better highlighted, important information in an image is better extracted, and reconstruction errors are reduced.

The invention provides an opening-closing fault diagnosis method for an air circuit breaker based on vibration signals, wherein an acceleration sensor is used to collect machine body vibration signals generated during opening-closing courses of the air circuit breaker. The method comprises the steps that firstly, the acceleration sensor is used to collect the machine body vibration signals generated during opening-closing actions of the air circuit breaker and transform the vibration signals into digital signals, so that initial vibration signals are obtained; secondly, an improved wavelet packet threshold de-noising algorithm is used to process the collected vibration signals; thirdly, a complementary ensemble-average empirical mode decomposition algorithm is used to extract intrinsic mode function components from the de-noising vibration signals; fourthly, the quantity Z of the intrinsic mode function components is determined; fifthly, the intrinsic mode function components of the first Z orders are selected and extracted as sample entropies of a characteristic quantity; sixthly, binary tree multi-classifiers based on a relevance vector machine are established; and seventhly, the binary tree multi-classifiers based on the relevance vector machine obtained at the sixth step are used to establish a fault recognition model of the air circuit breaker.

The invention relates to a non local joint sparse representation based hyperspectral image super-resolution reconstruction method. The method comprises: firstly, performing dictionary training on a low-spatial-resolution hyperspectral image with an online dictionary training method to obtain a corresponding spectral dictionary; secondly, performing joint sparse representation on similar pixel vectors by virtue of a full-color image of a same scene, and reconstructing a high-resolution image; and finally, processing the reconstructed high-resolution image by utilizing an iterative reverse projection technology to obtain a high-resolution hyperspectral image with smaller reconstruction error and higher visual quality. According to the method, the similar pixel vectors are subjected to non local joint sparse representation by utilizing the non local self-similarity property of the image, so that the visual quality of the reconstructed image is improved and structural features of edges, textures and the like of the image are reconstructed more effectively in an empty region while image spectral information is kept complete; and multiple wavebands of the hyperspectral image are subjected to sparse representation and reconstruction at the same time, so that the hyperspectral image with relatively high definition and identification degree can be reconstructed.

A video coding method according to the present invention is for coding a signal to be coded which represents a video, and includes: generating a prediction signal predictive of the signal to be coded, based on a coded signal coded prior to the coding of the signal to be coded (S16); quantizing a prediction error obtained by subtracting the prediction signal from the signal to be coded to generate quantized coefficients (S12); inversely quantizing the quantized coefficients to generate a quantized prediction error signal (S13); generating first filter information, based on statistical properties of only the prediction signal among the prediction signal and the quantized prediction error signal, and generating second filter information, based on statistical properties of only the quantized prediction error signal among the prediction signal and the quantized prediction error signal (S14); and performing entropy coding on the quantized coefficients generated in the quantizing and the first and second filter information generated in the generating of filter information so as to generate a coded signal (S15).

The invention relates to a sensing network clustering type space time compression method based on network coding and compression sensing. Targeted at problems of performance defects of reconstruction errors and computing complexities which are not low enough in existing research schemes during exploration of correlation of time and space of sensing data of a wireless sensor network, the invention brings forward a clustering type space time compression method with reference to network coding and a compression sensing theory; time space correlation of sensing data is deeply excavated; through design of appropriate network coding coefficients and observation matrix elements, network coding and the compression sensing theory are fused and unified in a real number domain; data reconstruction is ensured to be feasible and a high success rate is ensured; through construction of sensor node (cluster head node) independent codes and combination with a node combination decoding idea, reconstruction of compressed data of the method is enabled to have lower reconstruction errors. Meanwhile, exploration is carried out on the correlation between the time and the space step by step to guarantee low complexity of the reconstruction process.

The invention discloses a digital holography reconstruction method based on an iterated denoising shrinkage-thresholding algorithm. The method comprises the following steps of 1, reading an original image, and acquiring diffraction light complex amplitude O(x, y) by using object light information of the original image, wherein the amplitude of the original image is O<0>(Xi, Eta); 2, intervening the diffraction light complex amplitude O(x, y) obtained through fresnel diffraction and reference light R(x, y) added with o and Pi second phase shift quantity to separately achieve light wave complex amplitude U<1>(x, y) / U<2>(x, y) on a holographic plate, laminating the two holographic images I<(1)> and I<(2)> to form a phase shift holographic image I-bar (x, y), constructing a sensing matrix A, and allowing y=Pi I-bar=Pi RO=AO, wherein Pi is a known measurement matrix, R is a spare basis matrix, and y expresses obtained observation data; and 4, figuring out and acquiring a reconstructed image of the original object by using the iterated denoising shrinkage-thresholding algorithm (IDNST). On the basis of TwIST algorithm, dual shrinkage of a threshold value and a regularization parameter are introduced to the IDNST algorithm, the signal-to-noise ratio can be improved, moreover, the convergence speed is increased, the reconstruction precision is improved, and the reproduction quality reaches a more excellent level.

The invention discloses a method for reconstructing a three-dimensional curve face through point cloud. The method comprises the steps of inputting point cloud data P possibly carrying noise and abnormal values and the number m of peaks of the curve face needed to be reconstructed; initializing a dictionary matrix V and a connection matrix B; updating the dictionary matrix V and the connection matrix B in an iteration mode till convergence; outputting a reconstructed triangular net to complete reconstruction of the three-dimensional curve face. By means of the method, the triangular net is directly reconstructed through the point cloud, the noise input into the point cloud can be removed well, non-uniform sampling is processed in a robust mode, and the characteristics in the point cloud can be recovered well. Compared with a conceal method, the triangular net is directly reconstructed through the point cloud so as to avoid accumulated errors caused by multi-step optimization in the conceal method. Compared with a combination method, the reconstruction errors serve as a target function, so that the reconstruction errors of the curve face reconstructed through the method are usually smaller than those reconstructed by the existing combination method.

The invention discloses a video synthesis method based on lip language synchronization and miracle adaptation effect enhancement. According to the method, the portrait and the audio stream to be synthesized are directly and integrally coded; a cyclic decoder network retaining original face information is used to decode converted abstract features into an image sequence, and then five discriminatornetworks are used to carry out adversarial training on the synthesized image sequence according to a real image sequence, so that a total reconstruction error is minimized. Compared with an existingvideo synthesis method, the method has the advantages that the continuity of face change between the front frame and the rear frame is guaranteed, the definition of face pictures in the frames is improved, meanwhile, under the action of the lip language synchronous discriminator and the miracle adaptation discriminator, the synthesized video appears more natural, and the authenticity of the visualeffect is greatly enhanced. The method has high practical value in the aspect of improving the user experience of virtual live broadcast and man-machine interaction.

The invention relates to a Fisher discriminant dictionary learning-based warehouse goods identification method. The method comprises the following steps of: firstly dividing warehouse goods images acquired under different conditions into two parts: a training sample set and a test sample set; respectively preprocessing the two sample sets, rearranging pixel values and carrying out PCA dimensionality reduction; learning the training sample set through a Fisher criterion method to obtain a discriminant dictionary, and representing a test sample by using linear weighting of the discriminant dictionary; solving an L2 norm minimization problem by adoption of a least square method, so as to obtain a sparse representation matrix of the test sample under the discriminant dictionary; and finally realizing warehouse goods identification via ei formed by various types of reconstruction errors and sparse encoding coefficients. According to the method provided by the invention, the problems that the traditional identification method is greatly influenced by selected features, the identification process is relatively complicated and plenty of classification information is lost in the construction processes of common dictionaries are solved; and the correct and rapid identification of different goods can be realized, so that foundation is laid for the realization of intelligent warehouses.

The invention provides a low-resolution human face recognition method based on sparse maintaining canonical correlation analysis. The invention combines with the sparsity and canonical correlation analysis idea, and proposes the low-resolution human face recognition method based on sparse maintaining canonical correlation analysis. The method meets the requirements of maximum correlation of the extracted features through employing the canonical correlation analysis idea, achieves the fusion of the high and low resolution human face feature discrimination information, employs the sparsity idea to maintain the structural information, and improves the robustness of high and low resolution human face recognition. The method achieves the effective fusion of the high and low resolution human face feature discrimination information, improves the feature representation and discrimination capability, and meets the requirements of correlation and structural information maintaining.

The invention provides an adaptive threshold value iterative reconstruction method for distributed compressed sensing, and mainly solves the problems that signal reconstruction time is long, reconstruction errors are large and the like in the prior art. The method includes the steps: (1) calculating an adaptive step and an adaptive threshold value h; (2) calculating an iterative value according to an iterative formula; (3) comparing the iterative value with the calculated adaptive threshold value h to obtain an iterative result; (4) updating a support set and modifying the iterative result; (5) stopping iteration and acquiring estimation signals when meeting an iteration stopping condition, otherwise, continuing iteration. The adaptive threshold value iterative reconstruction method has the advantages of step and threshold value adaptivity, shorter reconstruction time, small reconstruction error and the like.

The invention discloses a network intrusion detection method based on a conditional variation auto-encoder, belongs to the technical field of network space intrusion detection, and the method comprises the steps: firstly training a conditional variation auto-encoder which takes a logarithmic hyperbolic cosine function as a loss function for the expansion of a data set, and then training a classifier based on the convolutional neural network through the data set after data expansion to serve as an intrusion detector of the whole model. According to the method, the classification detection performance on a network space intrusion data set is improved by utilizing the generation capacity of conditional variation self-encoding and the excellent feature extraction capacity of the convolutionalnetwork for data features, and meanwhile, the problem of performance reduction caused by imbalance of data set samples is also solved.

The invention discloses a short-term wind power prediction method based on complete ensemble empirical mode decomposition (CEEMD) and a random forest. The method comprises the following steps of 1) using a CEEMD technology to decompose an original wind power sequence into a series of intrinsic mode functions (IMFs) with different characteristics; 2) using an approximate entropy to calculate each intrinsic modal function complexity, merging modal functions with similar approximate entropy values into new components which are a random component, a detail component and a trend component; 3) carrying out zero equalization processing on different component data; 4) using a partial autocorrelation function (PACF) to determine an input variable set for the different components; and 5) constructing a random forest (RF) prediction model for each new component, superposing each component prediction result to acquire a final short-term wind power prediction value, and through an example, verifying validity of the method of the invention. By using the method of the invention, short-term wind power prediction precision is effectively increased and a short-term wind power prediction problem of an electric power system can be well solved.

The invention discloses an SAR (synthetic aperture radar) target identification method based on nuclear sparse representation, mainly solving the problem of low error tolerance in the prior art. The method comprises the following realization steps: (1) respectively mapping a training sample matrix and a test sample to a nuclear space, randomly reducing the dimension of the mapped sample to the required dimension, and normalizing the dimension; (2) solving a reconstructed coefficient vector between the normalized test sample and the training sample matrix; and (3) solving the energy of the reconstructed coefficient of the test sample in each class, and substituting the energy into a class judging formula to obtain a final identification result. Compared with the prior art, the SAR target identification method is characterized by improving the error tolerance of the algorithm, so that the SAR target identification method has higher identification precision and high arithmetic speed in the SAR target identification application; and meanwhile, an application range is popularized to a low-dimensional sample, thus having better universality.

The invention provides a high spectral abnormity detecting method based on dynamic weight deep self-coding and relates to a high spectral abnormity detecting method. The invention aims to settle a problem of low detecting precision caused by partial model pollution by an abnormal model in an existing high spectral abnormity detecting method. The method comprises the steps of 1, obtaining an optimized DBN model; 2, obtaining a coding image and a reconstruction error image; 3, obtaining a local coding image, and performing step 5; 4, obtaining a local reconstruction error set, and performing step 6; 5, obtaining a local distance factor, and performing step 7; 6, obtaining all dynamic weights of the local distance, and performing step 7; and 7, obtaining an abnormity detecting operator value,setting a threshold, and when the abnormity detecting operator value is larger than or equal with the threshold, determining the detected pixel as an abnormal target, and otherwise, determining the tested pixel as a background pixel; taking a next pixel in a detected image as the detected pixel, and performing the steps 3-7 until all pixels in the detected image are determined. The high spectralabnormity detecting method is used for a high spectral abnormity detecting period.

A method for designing QMF set includes designing original filter according to preset initial cut-off frequency, calculating system function of analysis filter according to system function of original filter, calculating system function of each analysis filter and composition error under each cut-off frequency, obtaining system function of each analysis filter and each integrated filter on confirmed QMF set according to cut-off frequency corresponding to minimum composition error.

The invention discloses a design method for designing a two-path orthogonal graph filter group, models a design problem of the two-path orthogonal graph filter group into a band constraint optimization problem, takes the reconstruction error of a filter as a target function and takes stopband attenuation as a constraint condition. Therefore, an iterative method is adopted to solve the problem. In single-step iteration, through a Taylor formula and functional approximation, a highly nonlinear non-convex target function is converted into a convex quadratic function, and a non-convex optimization problem is approximate to the subproblem of convex optimization. By use of the method, the two-path orthogonal graph filter group with better integral performance can be obtained.

The invention discloses a dangerous working area accident automatic detection alarm method based on deep learning, and the method comprises the steps: obtaining original video data, carrying out the preprocessing, and converting a video into an input training set acceptable for a deep learning network; learning a feature mode in a training video through a convolutional space-time automatic encoderdecoder, and performing training optimization by using the training set to obtain a workshop accident detection model; and acquiring a real-time monitoring video to be detected, detecting the reconstruction error of each frame of monitoring video image by adopting the workshop accident detection model, and if the local minimum reconstruction error of a plurality of continuous real-time monitoringimages is greater than a threshold value, sending corresponding alarm information and corresponding monitoring position information to a workshop administrator terminal. On the basis of analysis of alarge number of videos, video special learning of a normal scene is carried out, a fully trained detection model is obtained, abnormal accidents in a workshop can be rapidly and accurately detected,and accident detection can be carried out in any workshop scene.

The invention discloses a method of acquiring a three-dimensional foot shape by using a foot shape video and sensor data acquired by a smart phone. The smart phone provided with an IMU sensor is coiled on a naked foot to shoot the foot shape video and corresponding IMU data at 360degrees, and by establishing a factor diagram, a video camera attitude, a three-dimensional point cloud coordinate, the IMU data, a video contour, and other various acquired data are used as factors, and then by defining constraint error equations among the factors and solving the equations, the parameters of the video camera are acquired, and are introduced in a reference model for foot shape reconstruction. The point cloud distribution, the normal vector, the surface curvature, and other information of the reference model are fully used to guide deformation of a reconstructed foot shape, and then under the condition of the reconstructed foot shape being not deviated from a reconstruction point cloud, the reconstructed foot shape is close to a real foot shape as much as possible, and interferences of noise points are reduced, and therefore a process of acquiring a corresponding foot shape three-dimensional grid model conveniently is realized. Acquisition costs of foot shape data are reduced, and accuracy and robustness of parameter calculation are improved.

The invention discloses a human movement detection method based on movement dictionary learning. The human movement detection method includes that at the training stage, using a local property representing method to extract human movement properties from different video clips, and learning a human movement dictionary with strong distinguishing ability through training; considering reconstructing errors and new errors when modeling the movement dictionary so as to model better; at the testing stage, enabling a space-time sliding window to traverse the sparse codes of sliding windows of the whole video, and judging whether the space-time sliding window comprises a human movement according to the response values of the sparse codes to different dictionary items. The human movement detection method based on the movement dictionary learning can obtain the human movement dictionary through training without a negative sample, and the training process is easy and quick to finish.

The invention provides a fault sensor information reconstruction method based on measured value association degree. The method comprises the following steps: S1, aimed at a fault sensor, calculating association degree of response of a measuring point where a sensor is on in normal operation and other measuring points; S2, through comparing association degree values, determining an association model, establishing required response variables; S3, and using a partial least squares to establish a reconstruction model of reconstruction variables and the response variables, and using measured data of structural health monitoring, performing fault sensor response information reconstruction on the fault sensor. The method has very good effect on fault sensor information reconstruction, obviously reduces reconstruction errors, and ensures variation trend of the reconstruction values to keep consistent with variation trends of practical values. The method has obvious integrated structure reliability.

The invention discloses an Internet-of-Things time series data anomaly detection method and system. The method comprises the steps of obtaining to-be-tested Internet-of-Things time series data; dividing to-be-tested Internet-of-Things time series data to obtain a to-be-tested time series data segment set; and inputting the to-be-tested time sequence data segment set into the trained semi-supervised self-encoding model to obtain a detection result, wherein the trained semi-supervised self-encoding model is obtained by training the semi-supervised self-encoding model based on the LSTM and the attention mechanism by taking the to-be-trained Internet-of-Things time series data of the unmarked Internet-of-Things time series data and the marked Internet-of-Things time series data as input, taking the corresponding class label as output and taking the minimum loss function as a target. According to the invention, the accuracy of time series data anomaly detection can be improved, and the costis reduced.

The invention discloses a CT sparse projection image reconstruction method and a CT sparse projection image reconstruction device at a limited sampling angle. The method comprises the steps of obtaining a pseudo-inverse matrix of a projection equation according to projection data, generating a random solution set according to the pseudo-inverse matrix, reserving or replacing each solution in the current random solution set correspondingly, selecting an optimal solution from the current random solution set when the number of iterations reaches a preset maximum value, and adopting the selected optimal solution as a to-be-obtained reconstruction result. The device comprises a memory for storing at least one program and a processor for loading at least one program to execute the method of theinvention. According to the method, the pseudo-inverse of a discretized projection reconstruction equation is used as an initial solution of the algorithm, so that the quality of the initial solutionis guaranteed. A set of solutions are generated through random walk, and then the iterative optimization is carried out respectively. The diversity of optimization paths is guaranteed. Troubles causedby the defects of an initial solution and an iteration path of a traditional reconstruction method can be overcome. The method and the device are applied to the technical field of image processing.

The invention discloses a self-adaptive regularized smoothed l<0> norm method. A regularized SL0 algorithm is improved; in a steepest ascent method in an inner loop, a signal residual item estimated value iterative for the first time and a sparse signal estimated deviation value before and after the iteration are used as the selection basis of current regularization parameters; therefore, the signal sparse degree and the weight value of an error tolerance item in an outer loop every time can be adjusted self-adaptively; the balance of the two is kept in an optimization process; therefore, the reconstruction error of sparse signals can be effectively reduced; the anti-noise interference capability of the algorithm is improved; large-scale matrix inversion operation projected in operation of a feasible solution set in an iterative process can be avoided by introducing a SVD method; and the reconstruction speed to the sparse signals in the method disclosed by the invention is effectively increased.

The invention provides a pipeline magnetic flux leakage testing on-line data compression method mainly used for compressing mass data in pipeline magnetic flux leakage on-line testing, and belongs to the field of signal processing. The method comprises the following steps that (1) detected magnetic flux leakage data are divided into data segments having the same number of bytes; (2) whether each data segment contains pipeline defect information is judged by means of average absolute deviation statistical magnitude, and only the data segments containing the defect information are stored; (3) if multiple data segments contain a large amount of redundant signal data, main content of the data segments is analyzed, and only the first little main content is stored; (4) integral promotion wavelet decomposition is conducted on each detecting signal of each data segment after two-stage compression, threshold processing is conducted on wavelet coefficients produced after decomposition, then adaptive coding is conducted on the processed wavelet coefficients, and finally only bit stream data after corresponding coding are stored. Therefore, the method can achieve high-efficiency compression of pipeline magnetic flux leakage testing on-line data.

The invention discloses an error-controllable CAGE sequence representation algorithm for a dynamic grid. The algorithm comprises four parts of real matrix control grid generation, Poisson equation based weight simplification, sparse matrix control grid generation and control grid optimization. An input three-dimensional shape sequence and a control grid of one frame are given. According to the algorithm, a control grid sequence is obtained through the real matrix control grid generation; then a sparse coordinate matrix with locality is obtained through the Poisson equation based weight simplification; the sparse matrix control grid generation is performed; a reconstruction error is detected; and if a maximum error value is greater than a tolerance threshold input by a user, the control grid optimization is performed and the above three steps are performed again until a value specified by the user is reached. According to the algorithm, the problem in control grid sequence representation of the error-controllable dynamic grid is solved; and the algorithm can be applied to compression representation, acceleration editing and shape migration of a dynamic grid sequence.

The invention discloses a data dimension reduction method for an improved neighborhood preserving embedding algorithm. The method comprises the steps of firstly, constructing an adjacency graph, and calculating out a near point of each sample point by using a geodesic line, thereby forming an adjacency matrix; secondly, calculating a reconstruction weight value, and representing each sampling point with the near point; and finally, calculating a projection matrix, and performing calculation by utilizing a reconstruction weight value matrix to obtain a transform projection matrix. According tothe method, a Euclidean distance is replaced with a geodesic line distance; local structure information of the NPE algorithm is better kept; and the manifold structure processing capability of the algorithm is improved.