94 results about "Normalization algorithm" patented technology

The present invention relates to systems and methods for the optimized delivery of one or more advertisements from a group of advertisements. The method of the present invention comprises receiving a request for one or more advertisements and retrieving advertisements responsive to the request received. Click through data associated with the one or more advertisements is retrieved and is normalized using a normalization algorithm. The normalized click through data is used to assign weights to the one or more advertisements. Tracking codes are generated for the one or more advertisements to track user interactions with a given advertisement. Advertisements are distributed according to the one or more advertisements' associated weights with associated tracking codes in one or more formats to one or more locations through which advertising may be conducted.

The invention relates to an SOH (state-of-health) online estimation method of a battery pack. The SOH online estimation method comprises the following steps of: measuring temperature T, voltage V and current I of an electric automobile during actual operation to acquire the following function: SOC (state-of-charge)=f(T, V, I) by utilizing a normalization algorithm, establishing a database with one-to-one correspondence between the temperature T, the voltage V and the current I and the SOC, measuring the SOC of a monomer battery with lowest voltage in an online manner by utilizing the database, and further measuring the SOH of the battery pack. The SOH online estimation method disclosed by the invention has the beneficial effects that: 1) as the measured SOC is that of the monomer battery with the lowest voltage, which is a key factor for restricting the discharge capacity of the battery pack, a user can know the SOH of the batteries in a more visual manner, the remaining capacity of the batteries can be judged in a more accurate manner by combining with the SOC value, and the continue voyage course of the electric automobile can be estimated in a more accurate manner; and 2) if the problem is caused by attenuation of the battery capacity, the problem can be obviously seen through the SOH, and the efficiency of after-sale service is improved.

The present invention relates to systems and methods for the optimized delivery of one or more advertisements from a group of advertisements. The method of the present invention comprises receiving a request for one or more advertisements and retrieving advertisements responsive to the request received. Click through data associated with the one or more advertisements is retrieved and is normalized using a normalization algorithm. The normalized click through data is used to assign weights to the one or more advertisements. Tracking codes are generated for the one or more advertisements to track user interactions with a given advertisement. Advertisements are distributed according to the one or more advertisements' associated weights with associated tracking codes in one or more formats to one or more locations through which advertising may be conducted.

A method and system for determining whether arcing is present in an electrical circuit. The methods includes sensing a change in an alternating current in the circuit and developing a corresponding input signal, analyzing the input signal to determine the presence of broadband noise in a predetermined range of frequencies, and producing a corresponding output signal. The method further includes determining a type of load connected to the electrical circuit, based at least in part upon the input signal and the output signal, incrementing one or more of a plurality of counters in a predetermined fashion in accordance with the input signal and the output signal and determining whether an arcing fault is present based at least in part on the states of one or more of a plurality of counters. The method also includes decrementing one or more of the plurality of counters based upon a secondary analysis.

A method and system for determining whether arcing is present in an electrical circuit. The methods includes sensing a change in an alternating current in the circuit and developing a corresponding input signal, analyzing the input signal to determine the presence of broadband noise in a predetermined range of frequencies, and producing a corresponding output signal. The method further includes determining a type of load connected to the electrical circuit, based at least in part upon the input signal and the output signal, incrementing one or more of a plurality of counters in a predetermined fashion in accordance with the input signal and the output signal and determining whether an arcing fault is present based at least in part on the states of one or more of a plurality of counters. The method also includes decrementing one or more of the plurality of counters based upon a secondary analysis.

The invention relates to a standardized processing method which is based on three-dimensional head face surface modeling, the steps comprise a CT or MRI-based DICOM image measurement method, a modeling method which is based on stratified 12-order Fourier series fitting under a cylindrical coordinate with the forward pole point, a normalization algorithm which has continuity of the measurement method and the modeling method and is based on face feature points and a standardized processing method of the distribution of head size scale factors and head height length index-head width length index two-dimensional distributed head face three-dimensional surface. The standardized processing method of the invention overcomes that the general non-contact measurement methods are affected by hair and shaking when in measurement of head, overcomes the shortcomings of the general three-dimensional head modeling methods on parameter control, precision and face feature reflection aspects and proposes the head face three-dimensional surface standardized processing method.

The pedestrian re-identification method based on deep multi-view feature distance learning is specifically implemented according to the following steps: step 1, extracting region feature vectors; Step2, region division: according to all the feature vectors of the image obtained in the step 1, carrying out normalization through a normalization algorithm l2 norm; Representing a vector set of the image in a summation mode, and performing l2 norm normalization processing on image representation; dividing One image into N regions, and obtaining depth region aggregation characteristics; Step 3, LOMO feature extraction: respectively extracting traditional LOMO features from pedestrian images in the reference set and the test set; 4, carrying out multi-view feature distance learning, and obtaining two distances from the two aspects of depth region aggregation features and LOMO features through XQDA training of the two features; Step 5, a weighted fusion strategy: carrying out parameter weighted fusion on the two distances obtained in the step 4 to obtain a final distance, and obtaining a matched grade according to the final distance; The robustness of pedestrian re-identification can be obviously improved; And the pedestrian re-identification performance is improved.

The invention discloses a frequency locked loop method based on double self-tuning second-order generalized integrators. A self-tuning filter is cascaded with the second-order generalized integrator to obtain a third-order structure of the self-tuning second-order generalized integrator, so that the self-tuning second-order generalized integrator has good filtering property in the case of serious distortion of a power grid; a differencing node and a first-order high-pass filter is imported in the structure of the self-tuning second-order generalized integrator, so that the self-tuning second-order generalized integrator has the ability of eliminating DC component disturbance; and the adaptive control of nonlinear closed loop of frequency is converted into a linear design effect by using an amplitude normalization algorithm, and the programming implementation of DSP is facilitated. The frequency locked loop method disclosed by the invention is not only suitable for normal working conditions of the power grid, and is still suitable for the faulty power grid, the structure is simple, and the method is easy to implement, thereby having a wide application prospect in new energy grid connection and other fields.

The invention provides an engine noise control method based on an FXLMS algorithm aiming at the problem of in-vehicle noise caused by an engine air inlet system, belonging to the field of noise control. The method comprises the following steps: S1, establishing a main control system model of engine intake noise by using an FXLMS algorithm, and constructing a reference signal x (k) of the main control system model by using the rotating speed of the engine; S2, establishing an off-line identification structure, identifying a secondary channel transfer function H2(z) in an active control system model, and providing an identification result to the active control system model; and S3, controlling the engine noise by using the identified active noise control system model. In addition, an improved variable step length algorithm is proposed. The algorithm adds the parameter gamma to the step length of the normalized algorithm and replaces beta in the sine variable step length to adjust the amplitude range of the step length. The algorithm not only has the advantages of fast convergence and small steady-state error of the sine variable step length algorithm, but also has the characteristicsthat the normalized algorithm is suitable for a time-varying reference signal, and the parameter is easy to select.

The invention discloses a method for detecting position information of a patrol robot of a transformer substation through a mixed observation device. The method combines an initial position detection closed-loop control system, the mixed observation device, a model reference normalization algorithm, a fuzzy controller and a salient pole effect compensating controller to detect rotor position and velocity information of the patrol robot of the transformer substation accurately and effectively. The method can detect position and velocity information of the patrol robot of the transformer substation accurately and effectively.

The invention relates to a defect identification method for an electrowetting display screen. A convolution neural network with a batch normalization algorithm is added in the defect identification method for the electrowetting display screen. The convolution neural network comprises a four convolution layers, three pooling layers, three batch normalization layers, two full-connection layers, twodropout layers and an output layer. Batch normalization layers are added after the convolution layers, so that each convolution layer has the same data distribution. The generalization ability of thenetwork is improved, network convergence is accelerated, and therefore the model training speed and the defect identification precision are improved.

The invention discloses a speech emotion recognition method based on semi-supervised feature selection. According to the method, a specific classifier is trained for each speaker, so that the negative influence of speaker difference on speech emotion recognition is reduced. The training method comprises the steps of extracting the features of a label sample and a no-label sample of a certain speaker, obtaining the statistic result of all the features by means of multiple statistic functions, and executing the normalization algorithm; selecting a feature which can highlight the speech emotion of the speaker to be tested by means of the semi-supervised feature selection algorithm, wherein the semi-supervised feature selection algorithm can consider the manifold structure of data, the classification structure of data and information provided through the no-label data of the speaker to be tested at the same time; finally, training the classifier for recognition of speech emotion of the speaker to be tested by means of a support vector machine. By the adoption of the method, high recognition accuracy can be realized when the sample number for the speaker normalization algorithm is small.

A method of processing seismic data so as to provide an image of a sub-surface region, comprises providing plurality of migrated shot gathers that contain information about the region, summing portions of the migrated shot gathers to provide a pilot stack, partitioning the plurality of gathers into a plurality of groups and summing the gathers in each group to provide a substack, wherein each group includes at least two migrated shots and wherein a substack is generated from each group, applying an amplitude normalization algorithm to the pilot stack so as to generate an amplitude-normalized pilot stack, calculating a weight function by comparing each substack to the normalized pilot stack, weighting each substack using the weight function so as to generate a plurality of weighted substacks, summing overlapping portions of the weighted substacks so as to generate a output stack, and using the output stack to generate an image.

The present invention discloses an interaction platform based method for rapidly detecting a text in a complex background. The method comprises the following steps: S1: inputting an image for preprocessing; S2: rapidly positioning of a text candidate area: performing contour detection on the image processed in the step S1, and performing frame selection on each closed area by using a rectangular frame, then rapidly positioning all angle points in each rectangular frame by using an SIFT algorithm, and performing preliminary screening by using the number of the angle points as a preliminary screening condition; S3: firstly adjusting the image to a uniform size by using a normalization algorithm, then projecting contents in an original image to an exact center of the normalized image by a forward mapping method, and correcting an angle appropriately; and S4: text / background screening: extracting a feature firstly, and using a candidate area that is screened by a trained classifier as a detected text area. According to the method provided by the present invention, text detection in a complex background is realized, and a major problem that is solved is how to rapidly position a horizontal text and a text with a rotation angle in the image in the case of maintaining relatively high accuracy.

The embodiment of the invention provides a method for controlling the attitude angle of a three-degree-of-freedom parallel mechanism based on kinematics normalization. The parallel mechanism comprisesthree sets of actuators, three sets of support rods and a movable platform. The control method comprises the steps of establishing a body coordinate system OB-XbYbZb at the center of a circle where ahinge point on the movable platform is located, and establishing a fixed reference coordinate system Op-XpYpZp on the earth, wherein when the parallel mechanism is located at a working initial position, the body coordinate system coincides with the reference coordinate system; describing the attitude of the movable platform by applying the generalized coordinates of the body coordinate system relative to the reference coordinate system, wherein the generalized coordinates comprise three euler angles, and the three euler angles are composed of a transverse rocking angle, a longitudinal rockingangle and a yaw angle; acquiring a rotation matrix of the body coordinate system relative to the reference coordinate system; according to the length of the actuators or the length of the support rods, and applying the kinematics normalization algorithm to obtain the attitude angle of the movable platform based on the relation between the rotation matrix and the kinematics.

The invention discloses a picture processing method and device to improve the definition of character pictures and enable a user to recognize the content in the character pictures conveniently. The method comprises the steps that classifying training is carried out through pictures in a preset training set via a classification algorithm to obtain a judging model used for distinguishing the character pictures and the non-character pictures, wherein the preset training set comprises the character pictures and the non-character pictures; whether a picture to be processed is a character picture or not is judged through the judging model; if yes, image enhancing is carried out on the picture to be processed through an image normalization algorithm. By means of the picture processing method and device, the definition degree of the characters in the character pictures can be increased, and the user can recognize the character content in the character pictures easily; compared with the method that enhancement processing is carried out on all pictures to be processed, the picture processing method and device have the advantages that by means of the technical scheme, enhancement processing is carried out only on the character pictures, the image processing burden is relieved, and resources in the device are saved.

The invention discloses an echinococcosis serum Raman spectrum diagnostic apparatus based on an optimal back-propagation neural network. A laser Raman spectrometer scans each sample twice and is usedfor acquiring serum samples of healthy people and echinococcosis patients, acquiring spectrum data corresponding to the samples, averaging the spectrum data to serve as Raman spectrum data of the samples and transmitting the data to a computer. The computer receives the Raman spectrum data, Raman fluorescent backgrounds in the Raman spectrum data are eliminated by an adaptive iteration penalized least-square method, and noise of the Raman spectrum data without the fluorescent backgrounds is removed by a self-normalization algorithm. The computer takes 3 front main components as new variable input space for constructing a sample matrix after dimension reduction of the Raman spectrum data without the fluorescent backgrounds and after noise removal by a partial least-square method, classification is performed by the back-propagation neural network, 74 samples are selected among 123 samples and used for a training set, and 49 samples are used for a testing set.

The invention provides a sound wave signal feature extraction method for micro leakage of a gas pipeline based on random resonance. The method comprises the following steps that a discrete random resonance system model of a nonlinear filter is established, output signals of the random resonance system model are converted, windowing is carried out on the converted sound wave signals, the signals are converted to a Melfilter domain, discrete cosine transform is carried out to extract feature parameters, and denoising is carried out by utilizing a cepstrum mean-value normalization algorithm. According to the method, the random resonance system model is introduced into sound wave signal feature extraction in the Mel frequency cepstrum coefficient algorithm, and compared with an MFCC feature extraction algorithm, the signal to noise ratio is higher, identification is accelerated, and the method of the invention is more helpful for dynamic feature extraction.

The invention discloses a transformer substation acoustic signal feature extraction method based on a dynamic normalization algorithm, and belongs to the field of monitoring. The method comprises thesteps that: field acoustic characteristic signals of acoustic signals sent by primary equipment in different operation states are collected, and fault diagnosis is performed according to the acousticsignals collected by the primary equipment in various operation states, and is characterized by: obtaining a feature vector by utilizing a Mel frequency cepstrum coefficient based on discrete cosine transform; adopting a dynamic normalization algorithm to perform fault classification by comparing a reference template vector; and finally, based on the calculation result of the similarity between the test template and the reference template, analyzing actual operation states of primary equipment of various transformer substations, and classifying possible faults. Therefore, the resources occupied is little, the requirement for hardware calculation performances is low, and therefore, the transformer substation acoustic signal feature extraction method is especially suitable for dynamically and quickly matching fault characteristic signals in a transformer substation site, and is suitable for system operation on a single-chip microcomputer system. The method can be widely applied to the field of operation monitoring and state monitoring of unattended substations.

The invention discloses a camera self-calibration method based on an absolute conic image, belonging to the technical field of image processing and camera calibration. The method comprises the following steps: extracting feature points and feature lines, calculating images of vanishing points, vanishing lines and imaginary circles; according to the image of the vanishing point and the imaginary circle, the image of the absolute quadratic curve is calculated and the intrinsic parameter matrix of the camera is obtained. The rotation matrix is calculated by using the orthogonal property of the rotation matrix, and the translation vector is calculated by using the position of the camera center in the world coordinate system. The Euler angle of the camera is calculated by comparing the erasingline calculated by perspective projection model with the erasing line obtained by fitting the erasing points. Normalization algorithm is used to solve the basic matrix, determine the constraint relationship between different images, and complete camera self-calibration. The invention has the advantages of designing a new target, which is simple, convenient and practical to calibrate, has wide practicability, improves the accuracy of data processing, and reduces the influence of coordinate transformation on data.

The invention discloses a multimedia interactive teaching method and a system thereof, and a TV. The teaching method comprises the steps of: pre-storing image characteristic information and audio / video files of a plurality of learning cards; picking up the image data of the learning card held by a user; comparing the picked up image data with the pre-stored image characteristic information by a normalization algorithm to obtain a corresponding relative coefficient; and playing the audio / video file of the learning card corresponding to the maximal relative coefficient if the relative coefficient is larger than a preset threshold. The method of the invention has the advantages of picking up the image data of the learning card held by the user, comparing the image data with the pre-stored image characteristic information and playing the audio / video file of the learning card corresponding to the maximal relative coefficient, so that the user learns according to the learning card thereof instead of negatively accepting the learning contents, which improves the interactive effect of the user; besides, the system has a simple structure, and the system is not necessarily supported by a computer and a piece of peripheral equipment thereof, so that the cost is low.

The invention provides a recognition method in allusion to facial images with different dimensions. The method comprises the following steps: 1, carrying out discrete cosin transformation (DCT) on facial images and obtaining the same DCT coefficient, forming a sample set by a to-be-recognized unknown facial image and a known facial image set for matching, carrying out DCT on each two-dimensional facial grayscale image in the sample set, and selecting 56*46 components from a low-frequency part of a two-dimension DCT coefficient matrix of the transformation result to retain; 2, carrying out principal component analysis (PCA) on the transformed sample set to ensure that the dimensionality of a facial image characteristic value matrix is reduced to 20; and 3, calculating the characteristic matching value between every two the facial images by utilizing a correlation coefficient normalization algorithm, and selecting the facial image with the maximum characteristic matching value as a known facial image matched with the unknown facial image. The method provided by the invention is capable of solving the influences, caused by the image dimension differences of the facial images with different dimensions in the recognition process, on the recognition result, thereby improving the recognition correctness.

The invention discloses a user credit score calculation method and system, and relates to the technical field of data analysis, and the method comprises the steps: obtaining the index data of each dimension from user behavior data through employing a classification algorithm model, calculating the addition / deduction score and standard deviation of the index data through employing a work stealing algorithm, and employing Z-Score standardization algorithm, removing the index data greater than a standard deviation set multiple; normalization algorithm, mapping the index data into index data between 0 and 1 index data, the above steps are repeated to obtain the credit score of the index data of each dimension, and the sum of the credit scores of the index data of each dimension is calculated to obtain the credit score of the user, so that the authenticity of the credit score is improved, and the risk that the user violates the contract is reduced.

The invention discloses a method and system for determining a partial discharge source position. The method comprises the following steps: outputting the different shift of pulse signals at the end of a cable line for many times through a high-voltage lead, calibrating each shift of detection waveform signal, acquiring the verification waveform signal of each shift of detection waveform signal; performing partial discharge source detection on the cable line by using a way of applying voltage step by step, positioning the partial discharge source by using a time-domain pulse method, and acquiring a partial discharge source detection waveform corresponding a signal when the partial discharge source is located at the terminal of the cable line; performing equivalent correction on the discharge quantity calculated by the verification waveform to the same discharge quantity level of the terminal partial discharge source detection waveform by using a waveform normalization algorithm; and comparing the corrected verification waveform with the terminal partial discharge source detection waveform to determine whether the partial discharge source is at the head or the end of the detection cable line.

The invention relates to an equipment fault diagnosis method based on deep learning. Aiming at the characteristics of equipment data, a parameter regularization and Dropout method is added to an original basic convolutional neural network model to improve the generalization ability of the model, and an adaptive batch normalization algorithm is introduced on the basis of adding a BN algorithm afteran activation function, so that the adaptive recognition ability of the model is further improved. The adopted deep learning algorithm directly and automatically learns and extracts features layer bylayer; the device state change contained in the data change can be analyzed without further signal processing or depending on experience knowledge of experts, the fault type of the monitored device can be objectively and accurately reflected, and a basis is provided for guiding subsequent device fault management and maintenance work.

The invention discloses a dynamic industrial process fault diagnosis method based on a GRU depth neural network. The method divides original data into a plurality of sequence units as the input of theGRU, a GRU network is established through a batch normalization algorithm, extract dynamic characteristics can be effectively extracted from the sequence units, by adopting a softmax regression method, faults are classified according to the dynamic characteristics extracted by the GRU, and the probability interpretation of the classification is provided, so that the diagnosis result is further accurate, the problem of dimension disaster is avoided, and the accuracy of the fault diagnosis of the dynamic industrial process is improved.

The invention discloses a method and a device for detecting an object with anomaly in motion direction, which can obtain spatial derivatives and time derivatives, obtain characteristic patterns in the motion direction by fusing the two, further carry out normalization by taking the area of a motion region as comparison basis, merge the characteristic patterns in the motion direction in all the directions, and obtain a final saliency map in the motion direction. The device has simple structure and can significantly reduce the calculation quantity; the normalization algorithm relying on area comparison has better detection effect than the current normalization algorithm relying on extreme values, and is more in line with the definition of the anomaly in the motion direction; and the method and the device can lead the detection success rate for natural scenes and non-natural scenes to be higher than that of the prior art, and have the capability of sustainedly tracking the object with the anomaly in the motion direction. Obviously, the method and the device for detecting the object with the anomaly in the motion direction can reduce the calculation quantity, improve the reliability of the algorithm and improve the detection effect.

The invention discloses a method for evaluating an environmental technology based on a benchmarking method. Specifically, a three-dimensional index system is established; index benchmark values are extracted based on a technical database and an engineering case library constructed by literature research; technical single index evaluation is completed through a normalization algorithm and a hierarchical assignment method. Determining the weight of each index in combination with an entropy weight method and a subjective weighting method, calculating a comprehensive evaluation result of a singletechnology in each dimension through a weighting method, expressing the evaluation result in two forms of a single-dimensional radar map and a three-dimensional point position map, and objectively andcomprehensively evaluating the environmental technology; the method provided by the invention is a systematic and objective technology comprehensive evaluation method, and the method not only can realize dynamic evaluation of the technology, but also has strong operability.

The invention relates to a self-adaptive detection method for synchronism of a lamplight / motor and sound. The self-adaptive detection method comprises the following steps: detecting a phonetic acoustic signal by an ADC (analog to digital converter) controlled by AGC (automatic gain control); outputting a narrow dynamic range phonetic signal with AGC information; restoring the narrow dynamic range phonetic signal with the AGC information into a wide dynamic range phonetic signal by an energy normalization algorithm; carrying out phonetic real-time energy calculation and threshold calculation by the restored phonetic signal; and through comparison of the real-time energy and the threshold, calculating a motor starting and stopping state or a lamplight turning on and off state corresponding to the current phonetic frame. Through the introduction of the AGC information, high dynamic range obtained on the ADC with low sampling depth is detected by the motor opening and closing state or the lamplight turning on and off state information; and with the adoption of the ADC with 12 bit low sampling depth, the motor opening and closing state or lamplight turn on and off state when users speak can be detected at the distance of 5-300cm, so that the overall cost of the system is greatly reduced.

The invention relates to a non-reference screen content image quality evaluation method based on multiple scales. The method comprises the following steps: S1, converting a distorted image from an RGBcolor space to an LMN color space, amplifying an L component by using a bicubic algorithm, and extracting edge features of the distorted image by using an imaginary part of a Gabor filter; s2, amplifying the grey-scale map of the distorted image by using a bicubic algorithm, and extracting the structural features of the distorted image by using a Scharr filter and a local binary pattern; s3, extracting brightness features of the distorted image by using a local normalization algorithm; s4, taking the obtained three features as training data, and training an image quality evaluation model by utilizing random forest regression; and S5, according to the steps S1-S3, obtaining edge features, structure features and brightness features of the to-be-detected image, and predicting the quality score of the to-be-detected image by using the trained image quality evaluation model. According to the invention, the reference-screen-free content image quality evaluation performance can be significantly improved.