50 results about "Levenberg–Marquardt algorithm" patented technology

Systems and methods for determining characteristic transition values such as elbow values in sigmoid or growth-type curves, such as the cycle threshold (Ct) value in PCR amplification curves. A double sigmoid function with parameters determined by a Levenberg-Marquardt (LM) regression process is used to find an approximation to a curve that fits a PCR dataset. Once the parameters have been determined, the curve can be normalized using one or more of the determined parameters. Normalization is advantageous for determining the Ct value if one chooses the arbitrary fluorescence level (AFL) approach to calculating Ct values for amplification curves. After normalization, the normalized curve is processed by applying a root-finding algorithm to determine the root of the function representing the normalized curve, which root corresponds to the Ct value. The Ct value is then returned and may be displayed or otherwise used for further processing.

The invention discloses a method for monitoring exhaust temperature of an aircraft engine. The method comprises following steps: (1) acquiring and storing data of exhaust temperature of the engine; (2) extracting historical data of the exhaust temperature of the engine in an appointed time interval, and building a multi-dimensional training sample set and a testing sample set which are used for building and testing prediction models and building model input for prediction; (3) building a prediction model on the basis of a neural network in a convolution sum discrete process, utilizing a Levenberg-Marquardt algorithm to train the prediction model and testing and storing the prediction model; and (4) utilizing the prediction model to predict the exhaust temperature of the engine and applying the prediction model to monitoring states of the aircraft engine. The method is favorable for increasing the state monitoring and repair guarantee level of the aircraft engine, and accordingly achieves the purpose of guaranteeing air safety.

Provided is a large-area complex-terrain-region unmanned plane sequence image rapid seamless splicing method which comprises the following steps: to begin with, with air strip arrangement features of unmanned plane image sequence being prior knowledge, carrying out inter-image multiple-overlap SIFT feature point extraction and matching; then, carrying out matching point gross error removing and purifying based on random sample consensus algorithm, and solving transformation parameters of each image in spliced regions in an adjustment manner through an Levenberg-Marquardt algorithm; next, carrying out overlapped region image optimized selection according to the relative position relationship between central projection image point displacement rules and the images, and determining splicing lines; and finally, carrying out image uniform-coloring and fusion at the edge-connection places, and outputting spliced images, thereby realizing mass unmanned plane image seamless splicing. The seamless splicing method helps to improve the extraction efficiency of the SIFT feature points, guarantee the geometric accuracy of the spliced images, and eliminate the tiny color difference at the two sides of the image splicing line, and thus the spliced images with natural color transition and good natural object and landform continuity are obtained.

The invention provides a three-dimensional laser etch decomposing algorithm based on the combination of particle swarm optimization and LM algorithm (Levenberg-Marquardt Algorithm). The method comprises the steps of smoothly denoising; detecting the peak value; decomposing the waveform; fitting. According to the method, the threshold is set and the threshold is detected, so as to determine the quantity of smoothly denoised wave with good signal-to-noise ratio; the particle swarm optimization is performed to acquire the rough strength parameter value and the rough wide parameter value of single waveform to be used as the initial values of the LM algorithm to improve the decomposing precision as well as reducing the error influence of the initial value.

The invention discloses an external parameter calibration method used when a camera and a two-dimensional laser range finder are used in a combined mode and relates to three-dimensional information image processing of objects. According to the external parameter calibration method used when the camera and the two-dimensional laser range finder are used in the combined mode, a foldable calibration plate is used for achieving external parameter calibration when the camera and the two-dimensional laser range finder are used in the combined mode, the foldable calibration plate is a long plane which is formed by splicing small rectangular black planes and small rectangular white planes alternately and can be unfolded and folded, the foldable calibration plate is unfolded when used, and the unfolding angle can be adjusted freely; calibration of internal parameter of the camera is achieved by means of a Matlab camera calibration tool box; a camera and two-dimensional laser range finder system is established; the foldable calibration plate is placed, and calibration is completed under the condition that the number of times of information acquisition of a system is smallest; image information and two-dimensional information are acquired and processed by the camera and the two-dimensional laser range finder respectively; processed camera information and processed two-dimensional laser range finder information are matched in a one-to-one mode for combined calibration; a calibration result is optimized by means of the Levenberg-Marquardt algorithm, and then calibration is completed.

A measuring and calculating method of a profile line part without datum comprises the following steps: firstly, giving an initial translation value (Delta x, Delta y) and a rotation angle Delta theta of a scanning data point, defaulting an initial value to be zero, if the scanning data point has a large deviation with a theory type surface, manually giving the initial value; then, carrying out translation and rotation to the scanning data point and successively calculating a minimum distance di between the scanning data point and the theory type surface and rejecting a thick error point; secondly, using an improved Levenberg-Marquardt algorithm to perform an iterative solution to a least-squares fitting value of the theory type face to the scanning data point; finally, evaluating the scanning data and outputting an evaluation result. By using the measuring and calculating method of the profile line part without the datum disclosed in the invention, a concrete number can be given during evaluating the profile line, precision of the evaluation result is high and repeatability is good. A program writing mode is used to automatically realize calculation during calculating so as to improve batch test efficiency.

The invention discloses a neural network-based acoustic glass defect detection method which comprises the following steps: acquiring a knocking signal of a glass sample through a pickup in an actual production environment; preprocessing the knocking signal; performing feature extraction on the pure knocking signal; setting initial parameters of a BP neutral network: taking an extracted feature as the input of the neutral network, setting the number of nodes on an input layer of the BP neutral network to be 7, setting the number of nodes on a hidden layer of the BP neutral network to be 15, setting the number of nodes on an output layer of the BP neutral network to be 2, and setting an output result to be (0, 1) which indicates that the glass sample has a defect, and to be (1, 0) which indicates that the glass sample is defect-free; training the BP neutral network, setting the learning rate of the BP neutral network to be 0.1, setting a target square error value to be 0.1, training the BP neutral network through a LeVenberg-Marquardt algorithm, and stopping training if an error of the neutral network is less than the set target square error value. A signal feature extracted through the method is high in distinction degree, and can complete a glass defect detection task more accurately and efficiently.

The invention discloses a method for extracting the depth of a shallow buried pipe in a reclamation land, and the method comprises the steps: arranging a ground penetrating radar measuring line in a direction perpendicular to the distribution direction of the buried pipe, employing a high-frequency ground penetrating radar antenna for detection, and obtaining an original radar image of a research area; building a method based on the variance statistics, and extracting a target region of the buried pipe on a radar image; calculating the similarity level and correlation coefficient of adjacent echo signals, extracting feature points (shown in the description) of a reflection characteristic curve of a pipe target; carrying out the curve fitting of the feature points through employing a Levenberg-Marquardt algorithm, and obtaining the top point of a reflection characteristic curve; obtaining the electromagnetic wave propagation velocity based on the above through combining the diameter of the buried pipe and employing a least square algorithm, and finally calculating the depth of the buried pipe more accurately. The method employs the lossless detection technology based on the ground penetrating radar, and is a new method for the acceptance check of the land reclamation and consolidation projects of our country.

The invention discloses a two-dimensional plane iteration type camera calibration method. The method is characterized in that, on the basis of a two-dimensional visual measurement-based camera calibration method, the internal parameter of a camera is obtained. After that, a calibration plane is rotated by a certain angle around the origin OW of a world coordinate system, and then the internal parameter is solved again until the difference between parameters obtained every two adjacent times is smaller than a certain threshold value. Finally, an optimized objective function is built, and the calibration result is optimized through a levenberg-marquardt algorithm. Compared with a traditional two-dimensional visual measurement-based calibration method, the distortion of a lens in the shootingprocess is considered, and the calibration precision of the camera is improved.

The invention discloses a confocal axial response curve peak position extracting algorithm and belongs to the technical field of confocal scanning optical measurement. The method comprises the following steps of at first, normalizing data, and selecting effective data from measured data; taking a sinc4(a(x-b)) function as an objective function, and calculating the initial value of target parameter according to measured parameters; taking the sinc4(a(x-b)) function as the objective function, and using Levenberg-Marquardt algorithm to perform fitting to obtain the target parameter; according to the target parameter, calculating peak position. According to the confocal axial response curve peak position extracting algorithm, a confocal axial response curve peak position is obtained by taking sinc4 as the fitting objective function, so that data utilization rate and measurement uncertainty can be improved, meanwhile, iteration times is reduced, and the operation time of the extracting algorithm is reduced.

The invention discloses a method for identifying SNP in an individual in Sanger sequencing oriented to PCR products of diploid. According to the method, firstly, fluorescent data of four bases of adenine A, guanine G, cytosine C and thymine T contained in a chromatogram map are independently separated, filtering and noise reduction processing is carried out on the separated fluorescent data by adopting a small wave multiscale analysis method; the waveform characteristics of the fluorescent data of the four bases are further analyzed, a first peak and a second peak of the waveform are detected, and the peak distance, the height specific value and the fluctuation degree specific value of the waveform characteristics are selected as the elements for judging SNP loca, a BP nerve net with the structure of 3-10-1 is selected as a classifier for the detection of the SNP loca, and training is carried out on the BP nerve net by adopting a Levenberg Marquardt algorithm; output is mapped as SNP evaluation scores from 0 to 100 by adopting piecewise linear transformation, the classification of the SNP loca is defined from a 1 level to a 5 level according to the evaluation scores, and the SNP confidence coefficient of the loca is judged according to the classification. The method for identifying the SNP in the individual in the Sanger sequencing oriented to the PCR products of the diploid can effectively detect the SNP loca in the individuals in sequencing files.

A power capsule real-time positioning method based on a permanent magnet includes the steps of firstly, collecting magnetic field intensity signals of the in-vivo permanent magnet in a power capsule through an in-vitro three-dimensional magnetic sensor array; secondly, conducting wavelet transform on data measured through the three-dimensional magnetic sensor array so as to filter out noise produced by breathing, electromagnetic interference and the like; thirdly, eliminating errors from the magnetic field intensity signals filtered in the second step according to a magnetic field mathematic model through an Levenberg-Marquardt algorithm and obtaining the position of the capsule in a gastrointestinal tract, wherein the three-dimensional magnetic sensor array comprises four three-dimensional magnetic sensors located on the same plane, and each three-dimensional magnetic sensor collects three magnetic field intensities in perpendicular directions. The power capsule is positioned in the gastrointestinal tract by detecting the magnetic induction intensity of the in-vivo permanent magnet through the in-vitro magnetic sensors. Because a human body has magnetic permeation capacity and little influences a static magnetic field, a positioning system can reach higher positioning accuracy; in addition, the system has the advantages of being small in size and simple in structure.

The invention discloses a laser tracker positioning method based on a laser tracker multi-position measuring system. The laser tracker positioning method comprises the following implementing processesthat a laser tracker multi-position measuring system is established; a laser tracker positioning is self-calibrated based on a Levenberg-Marquardt algorithm; and a parameter [mu]i is selected; lasertracer position coordinates are optimized; and the laser tracer position coordinates are optimized through the Levenberg-Marquardt algorithm and a singular value decomposition and transformation method of a covariance matrix. According to the laser tracker positioning method based on the laser tracker multi-position measuring system, the precision of the laser tracer position coordinates can be improved, multi-shaft machine tool coordinate errors are more accurately corrected, and thus the corrected results are more accurate.

The invention discloses a method for adjusting the accuracy of an area scanning three-dimensional measuring system in real time. The method includes the steps that first, whether the accuracy of the area scanning three-dimensional measuring system meets the requirement or not can be determined by judging whether the internal and external parameters of a camera meet the current working condition requirement, if the internal and external parameters of the camera meet the current working condition requirement, measuring continues, otherwise, a Levenberg-Marquardt algorithm is used for optimizing the internal and external parameters of the camera, so that the average value of objective functions is made to be minimum, and at the moment, the internal and external parameters of the camera are considered to be optimal; afterwards, whether the average value of the objective functions is smaller than an error threshold value or not is judged, if yes, the optimized internal and external parameters of the camera are used for measuring continuously, and if not, a user is prompted to conduct calibration again. By the adoption of the method, self testing of the accuracy and automatic optimizing of the parameters of the camera can be conducted in real time in an on-line mode, and under the condition that repeated calibration is avoided, the average value of reprojection errors of the camera can be kept at the pixel of about 0.0028 for more than twenty days.

The invention provides a Duhem-based piezoelectric actuator neural network parameter identification method. The method comprises the following steps: 1, a discrete parameter model is deduced by a Duhem model differential equation; 2, according to an interconnection relationship between the discrete parameter model and parameters, a neural network is built; 3, a static test principle is used to acquire an initial input and output data set for training of the neural network; 4, according to a Levenberg-Marquardt algorithm, the initial input and output data set and a preset training target, the neural network is trained, and according to a training result, a weight-adjustable value of the discrete parameter model is calculated; and 5, according to the discrete parameter model and the weight-adjustable value, values of the parameters are calculated.

The invention relates to an image rebuilding algorithm of DOT / XCT (diffuse optical tomography / X-ray computer tomography) dual-mode imaging based on a boundary element, and belongs to the technical field of molecular imaging. The method comprises the following steps of on the basis of providing an organization boundary by XCT , using a boundary element algorithm to solve the positive problem of DOT, namely solving the distribution of light flux at the boundary under the condition of known distribution of optical parameters of different organizations; and in the inverse problem, adopting a Levenberg-Marquardt algorithm to rebuild the absorption coefficients and scattering coefficients of different organizations, namely solving the absorption coefficients and scattering coefficients of the different organizations from the detected light flux data through the process of optimizing a non-linear minimum square model. Compared with the traditional rebuilding method based on finite elements, the boundary is utilized to rebuild in the DOT / XCT dual modes, and the rebuilding of the optical parameters of the organizations is more accurate and stable.

The invention discloses a large-scale precision turntable calibration method based on a multi-station measurement system of a laser tracker. The method comprises the following steps of building a laser tracker multi-station measurement system, self-calibrating the laser tracker stations based on the Levenberg-Marquardt algorithm, selecting the parameter mui, optimizing coordinates of the laser tracker stations, fitting the center of circle of the rotating shaft of the turntable, and calibrating the positioning accuracy of the large-scale precision turntable. The method utilizes the Levenberg-Marquardt algorithm and the singular value decomposition transformation method of the covariance matrix to optimize the coordinates of the laser tracker stations, and calibrates the positioning accuracy of the turntable by establishing a geometric relationship model between the optimized coordinates of the laser tracker stations and the rotation angle of the turntable. The proposed method for calibrating the positioning accuracy of the turntable based on the multi-station measurement system of the laser tracker is suitable for the case where the turntable and the three-axis machine tool are notlinked, and is particularly suitable for large-scale high-precision turntables. At the same time, the method can provide a theoretical basis for the calibration of multi-axis machine tools.

The invention discloses an elastic motion estimating method based on improved Levenberg-Marquardt optimization, and belongs to the field of video coding compression. Aiming at the defect that a Hessian matrix of a reference frame and an inverse matrix thereof need to be recalculated in the process of solving an elastic motion model with the traditional Gauss-Newton algorithm, the Hessian matrix of the reference frame is replaced by that of the current frame with reference to the thought of a Levenberg-Marquardt algorithm, and a hybrid Hessian matrix of the traditional Levenberg-Marquardt optimization algorithm is replaced by a Gauss-Newton Hessian matrix, so that the Hessian matrix and the inverse matrix thereof only need to be calculated once during each motion estimation; meanwhile, an early termination strategy is introduced, so that the purpose of reducing the elastic motion estimation calculation complexity is fulfilled.

The invention discloses a photovoltaic generating capacity prediction method based on a fuzzy EBF (Elliptical Basis Function) network. The photovoltaic generating capacity prediction method comprises the following steps of: selecting the influence factor of a photovoltaic generating capacity, collecting the historical data of the influence factor of the photovoltaic generating capacity and photovoltaic generating capacity historical data corresponding to the historical data of the influence factor of the photovoltaic generating capacity, and determining a sample set; generating the historical data of the influence factor of the photovoltaic generating capacity in a training sample set into an input vector, taking the photovoltaic generating capacity historical data corresponding to the historical data of the influence factor of the photovoltaic generating capacity as an output vector, carrying out normalization processing, and determining a training sample; utilizing the training sample to train the fuzzy EBF network through a Levenberg-Marquardt algorithm, collecting the influence factor data of the photovoltaic generating capacity of a day to be predicted, generating a prediction input vector, carrying out the normalization processing, inputting the prediction input vector into the trained fuzzy EBF network to obtain a photovoltaic generating capacity prediction output vector, and carrying out reverse normalization processing on the prediction output vector to obtain a prediction photovoltaic generating capacity vector of the day to be predicted. By use of the photovoltaic generating capacity prediction method, the prediction accuracy of the photovoltaic generating capacity is improved.

The invention discloses a fisheye camera calibration system, method and device, electronic equipment and a storage medium. The system comprises a polyhedral target, a fisheye camera and electronic equipment; the polyhedral target comprises an inner surface and a plurality of marking layers arranged on the inner surface, and the inner surface is composed of a plurality of hexagonal planes and pentagonal planes; the fisheye camera is used for shooting the polyhedral target and collecting a target image, and the target image comprises the polyhedral target and images of a plurality of mark pointsarranged on the inner surface of the polyhedral target; the electronic equipment is used for obtaining initial values of k1, k2, k3, k4, k5, u0, v0, mu, mv, Tj and Rj; the initial values of k1, k2, k3, k4, k5, u0, v0, mu, mv, Tj, Rj are optimized by using the Levenberg-Marquardt algorithm to determine imaging model parameters of the fisheye camera.. Through the polyhedral target, a plurality of target images can be acquired without moving the target or a camera, and rapid and high-precision calibration of the fisheye camera can be realized only by acquiring one target image.

The invention provides a method for determinating the permeability of a fluid in a porous medium. The method comprises the following steps: establishing a control model for fluid flow in the porous medium by using a steady-state Stokes equation, and dividing the porous medium into a plurality of staggered grids; enabling a fluid to flow in the porous medium in a first direction, establishing a least squares objective function by selecting displacement pressure differences at different times and cumulative production as dynamic data on the basis of a radial-flow core displacement experiment, characterizing an oil-water relative permeability curve by using a cubic uniform spline model, and in combination with a radial-flow numerical simulator, continuously adjusting a control parameter vector of a phase permeability characterizing model by a Levenberg-Marquardt algorithm so as to minimize a quadratic sum of errors of observed values and predicted values of the dynamic data. According tothe scheme, the oil-water relative permeability curve can be accurately calculated on the basis of processing of radial-flow core displacement experimental data, and an effective tool is provided forstudying the percolation mechanism of oil, water, asphalt and a resin fluid in the porous medium.

The invention provides a neural network prediction and control method for the water turbidity in the medicine automatic cleaning process of traditional Chinese medicine decoction pieces. For the problem that the water pressure in the medicine automatic cleaning process of the traditional Chinese medicine decoction pieces cannot be adjusted according to the actual amount of the cleaning medicine and the amount of sand, a Levenberg-Marquardt algorithm is used to train a neural network prediction model of turbidity of the overflowing water and the water pressure of the cleaning pump in the medicine automatic cleaning process of the traditional Chinese medicine decoction pieces. A cumulative function of a water turbidity tracking deviation and the water pressure change amount of the pump overa period of time in the future is minimized by a differential evolution algorithm. The water pressure of the pump in the medicine automatic cleaning process of the traditional Chinese medicine decoction pieces is calculated in real-time. The optimization and automatic control of a water turbidity tracking set value in the in the medicine automatic cleaning process of the traditional Chinese medicine decoction pieces is realized. The invention provides the neural network prediction and control method for the water turbidity in the medicine automatic cleaning process of the traditional Chinese medicine decoction pieces which is intuitive in understanding, simple in design, and prone to implement.

The invention relates to an analysis method of indentation data. The core of the method is based on the numerical optimation of a Levenberg-Marquardt algorithm, an optimized target function is a difference between the indentation data obtained by finite element modelling and experiment indentation data, and an optimized parameter is a mechanical parameter of a tested material to be obtained. A plurality of loading and unloading schemes and constitutive models of materials can be defined in the finite element model, therefore, the analysis method of the data is very strong in universality.