318 results about "Sample vector" patented technology

The invention discloses a behavior identification method based on a 3D convolution neural network, and relates to the fields of machine learning, feature matching, mode identification and video image processing. The behavior identification method is divided into two phases including the off-line training phase and the on-line identification phase. In the off-line training phase, sample videos of various behaviors are input, different outputs are obtained through calculation, each output corresponds to one type of behaviors, parameters in the calculation process are modified according to the error between an output vector and a label vector so that all output data errors can be reduced, and labels are added to the outputs according to behavior names of the sample videos corresponding to the outputs after the errors meet requirements. In the on-line identification phase, videos needing behavior identification are input, calculation is conducted on the videos through the same method as the training phase to obtain outputs, the outputs and a sample vector for adding the labels are matched, and the name of the sample label most matched with the sample vector is viewed as a behavior name of the corresponding input video. The behavior identification method has the advantages of being low in complexity, small in calculation amount, high in real-time performance and high in accuracy.

A reduced rank adaptive digital filtering method is described for a received signal consisting of a sequence of N×1 received vectors. Each received vector is formed from a group of N successive samples. D+1 basis vectors are generated where D is less than N and the dimension of a desired reduced rank subspace. Each successive basis vector is generated by multiplying an immediate preceding basis vector by the covariance matrix for the sequence of received sample vectors and the first basis vector is formed from a given or estimated steering vector. D filter coefficients are generated from correlations between pairs of basis vectors. The adaptive digital filter of the present invention achieves near optimal rank performance with substantially fewer training symbols than heretofore possible.

The invention discloses a method for generating and beautifying handwritten characters with a computer. The method is realized by learning a certain amount of hand written character samples. The method comprises the following steps: decomposing and parameterizing the strokes of five to twenty different handwritten character samples for the same character, and converting each character into the form of sample vector by the image processing method; linear interpolation is conducted among characters to obtain new handwritten characters; evaluating each new character by an automatic computer evaluation method of character writing form aesthetic degree; and adjusting interpolation coefficients repeatedly through a nonlinear programming optimization method so as to achieve the effect of beautifying the character. The system of the invention can automatically generate handwritten characters with better aesthetic degree, and improve the aesthetic degree of handwritten characters of a user on the basis of reserving the basic characteristic of the handwriting of the user.

The invention relates to the technical field of text processing, in particular to an N-Gram participle model-based reverse neural network junk mail filter device. Customized word characteristic items are added to mail particles by using N-Gram technology, and judgment and filter of junk mails are implemented by combining a reverse neural network. The device is implemented by the following steps of: firstly, processing the mails by using a Markov chain and an N-Gram technique, extracting mail sample characteristics, and obtaining a sample mail word-document space by weight calculation and characteristic selection; secondly, matching a mail sample by using the customized word characteristic items to generate a customized characteristic-document space, and combining the document characteristics generated by the two methods to generate a new mail vector space; thirdly, constructing a reverse neural network model, generating characteristic vectors corresponding to network neurons according to the characteristic items of a mail training sample space, and training the network model by using the mail training sample vector space to obtain a trained mail classifier; and finally, generating a test sample vector space by the mail test sample according to the generated characteristic vectors corresponding to the network neurons, and testing the mail type judgment accuracy of the trained mail classifier. The embodiment of the invention can judge the junk mails so as to filter the junk mails.

A method of detecting interference in a received sample vector using hidden Markov modelling by first estimating noise variance, where estimating noise variance comprises the steps of receiving a sample vector of noise and interference, sorting the sample vector in the frequency domain by order of increasing magnitude to produce an ordered vector, finding a sub-vector of the ordered vector that minimises the distance from a noise measure, and estimating the noise variance.

Each vehicle in a fleet comprises a climate controller having an auto mode. The auto mode controls climate actuators in response to a model relating sensed climate conditions to respective settings for the actuators. The vehicle has a buffer memory periodically storing sample vectors comprised of actuator/operation settings and sensed climate conditions. A user interface in the vehicle is responsive to a user override commands to modify respective operation settings while in auto mode. Each vehicle has a wireless communication system for sending data packages to a remote server when the user generates the override. Each data package is comprised of a plurality of stored sample vectors and an identification of the override command A central database associated with the remote server receives the data packages from the fleet vehicles in order to identify patterns within the received vectors that are associated with any given override command

A method for recognizing a transformer partial discharge pattern based on a singular value decomposition (SVD) algorithm includes a training model and a classification recognizing process, comprising: firstly setting up an experimental environment having artificial defects, collecting at least one datum sample, and calculating statistical feature parameters of each datum sample to form a datum sample matrix; performing singular value decomposition on the datum sample matrix and determining an order of an optimal retention matrix by judging whether a feature of a retention matrix is clear, so as to obtain a type feature description matrix and a class-center description vector group after dimensionality reduction; preprocessing samples to be recognized to obtain a sample vector, and performing linear transformation on the sample vector utilizing a type space description matrix.

A level gauge using microwaves to determine a distance to a surface of a product in a tank, wherein a measurement signal comprises a first frequency sweep, and a second frequency sweep, and a mixer is arranged to mix the measurement signal with an echo signal to form a first IF signal based on the first frequency sweep, and a second IF signal based on the second frequency sweep. Processing circuitry is adapted to sample the first IF signal and the second IF signal, to form a combined sample vector including samples from each tank signal, and to determine the distance based on the combined sample vector.

By combining the samples from two (or more) different sweeps, the number of samples and the bandwidth can both be increased, thus maintaining the range L. However, as the samples are obtained from two separate sweeps, the sweep time for each individual sweep does not need to be increased, and the average power consumption can be maintained.

The invention discloses an overlapped-between-clusters-oriented method for classifying two types of texts. The method comprises the following steps of: forming training sample vectors, identifying training samples in an overlapped area and judging whether the training samples are in the overlapped area between clusters; re-dividing training sample vector sets, constructing a first layer classifier on the newly divided training sample vector sets; in various kinds of training sample sets in the overlapped area between clusters, extracting binary word strings formed by the words, of which adjacent words are verbs or nouns, as characteristics to construct a second layer classifier; and finally, performing the first layer classification on test samples, if the conditions are met, identifying the test samples by the second layer classifier, and merging the results of the two layers of classifiers to obtain a final classification result. The method is applied in the fields of classification of the texts with a higher overlapped-between-clusters degree, information filtration and information monitoring, and can ensure the accurate classification of the texts with a higher overlapped-between-clusters degree.

The invention relates to an RFID positioning method based on an adaptive deep belief network. The method comprises a step of laying the positions of a reader and reference tags, and calculating distances from the reference tags to the reader, a step of allowing the reader to send electromagnetic wave signals to the reference tags and receive signal intensity RSSI values, and constructing a training sample vector matrix, a step of constructing the adaptive deep belief network, taking the RSSI value of each reference tag as an input value, and taking the distance d of each reference tag as an output value, a step of using a contrastive divergence algorithm to complete the pretraining of a network parameter, a step of using an adaptive moment estimation method to adjust a weight of each layer of a deep learning network, and a step of allowing the reader to send a signal to a tag to be measured and receive an RSSI value, and using the deep belief network to predict the position of the tag to be measured. According to the method, the adaptive deep belief network is used, a nonlinear relationship between a signal intensity value and the distance is constructed, a cross entropy cost function is used, and a problem of a slow learning rate is alleviated.

The invention discloses an analog circuit fault diagnosis method based on depth learning and complex characteristics. A fault-free state and all fault states are simulated by using simulation software; different representative working frequency points are set successively; an amplitude and phase of a fault-free signal are measured at each measuring point and a real value and an imaginary value of the signal are obtained by calculation; the real values and the imaginary values are processed to form sample vectors; and tag marking is carried out according to a fault state. A classification network is constructed by using a self-coding network and a classifier; training is carried out by using the sample vectors and the corresponding tags; when a fault diagnosis needs to be carried out on the analog circuit, different representative working frequency points are set successively; current amplitudes and phase are measured at all measuring points; sample vectors are constructed according to a pattern; the sample vectors are inputted into the trained classification network to obtain a classification result, thereby obtaining a fault diagnosis result. According to the analog circuit provided by the invention, on the basis of combination of the self-coding network with complex characteristics of signals, the accuracy of analog circuit fault diagnosis is improved.

Embodiments of the invention provide a fuzzy classification technology-based slope reliability parameter obtaining method and apparatus, and belong to the field of data processing. The method comprises the steps of generating k training sample vectors through an orthogonal design method according to mean values and standard deviations corresponding to m uncertainty parameters respectively; according to the k training sample vectors and one or more deterministic parameter values, obtaining slope stability coefficients corresponding to the k training sample vectors through a slope stability analysis method; by taking the k training sample vectors as independent variables and taking the slope stability coefficients corresponding to the k training sample vectors as dependent variables, forming a mapping relationship, and obtaining a mapping relationship expression through a support vector machine algorithm; and according to randomly generated N to-be-tested sample vectors obeying joint probability distribution, the mapping relationship expression and a preset instability state fuzzy judgment function, obtaining slope reliability parameters. The slope stability is quantized through the instability state fuzzy judgment function, so that the accuracy of the slope reliability parameters is improved.

A system for detection of a target sound in an environment of a vehicle, includes an audio sensor, a computer processor, and a memory storing digital target sound templates produced by converting samples of the target sound in accordance with conversion parameters. The computer processor receives the sample sound signal from the audio sensor, converts the sample sound signal to a digital sample vector in accordance with the conversion parameters, compares the vector to each of the templates to determine if a degree of similarity of the vector to any one of the templates exceeds a predetermined threshold value, and notifies a user or a vehicle control system if the threshold value is exceeded.

The invention provides a method and device for detecting a junk mail. The method includes: generating a sample vector according to a sample library and a feature word lexicon which includes normal mail type feature words and junk mail type feature words which are extracted from sample mails of the sample library; selecting a linear kernel function of a support vector machine, using the sample vector as input and training to obtain a classification function; determining weights of feature words in the feature word lexicon according to a coefficient of the classification function, picking up feature words whose weights are nonzero values to generate a feature word set, and determining a judging threshold value according to an offset of the classification function; and making statistics of a sum of the weights of the feature words contained in a mail to be detected according to the feature word set, and judging the mail to be a junk mail when the sum of the weights exceeds the judging threshold value. The method for detecting a junk mail saves the calculation amount of a detection process, and improves detection efficiency under the condition of guaranteeing detection precision.

The invention discloses a generative confrontation network-based method for writing calligraphy by a robot and relates to the robot. The method comprises the steps of collecting standard calligraphy stroke data, collating and sorting according to stroke types and making annotations; training two deep neural networks: a generative network G and a confrontation network D, on the basis of a generative confrontation network model; inputting a randomly sampled vector into the generative network G to obtain the probability distribution of track points of strokes; obtaining position information of the strokes from the probability distribution by adopting a sampling method and writing the strokes to a drawing board by a calligraphic robot and shooting and recording images of the strokes by a camera after the writing; and preprocessing the to-be-processed images and inputting into the confrontation network D to train and adjusting parameters to achieve convergence. The generation mechanism hasa good learning ability to enable the calligraphic robot to have a generation mechanism that can automatically generate various styles of strokes, so that the difficulty that a large number of labor power is consumed to manually input, of a current calligraphic robot, is solved.

The invention relates to a technique for processing petroleum geophysical prospecting seismic data, which is a method for improving the velocity spectrum resolution by using phase information. The method comprises the following steps of: stimulating and recording seismic waves to obtain a common midpoint gather; performing a first-order derivative operation along a time direction, and performing Hilbert transform to obtain a complex seismic trace; superimposing P(t) traces to obtain the optimal velocity of normal moveout correction (NMO); calculating a velocity spectrum by using a conventional velocity analysis method; and performing side lobe suppression on the selected common midpoint (CMP) for velocity analysis and each time sample vector in SPS to finally obtain the velocity spectrum. By the method, influences on the velocity analysis due to amplitude changes can be avoided, the distribution of noises is converted into the distribution with gauss distribution characteristics, so that influences on the velocity analysis due to abnormal noises can be effectively inhibited, and the resolution of the velocity spectrum and the accuracy of the velocity analysis are improved.

Methods, systems, and devices are described for wireless communication at a UE. A user equipment (UE) may utilize orthogonal frequency division multiple access (OFDMA) for demodulating downlink messages and a combination of Gaussian minimum shift keying (GMSK) and single carrier frequency division multiple access (SC-FDMA) for uplink modulation. The uplink modulation process may include generating a symbol vector with an M-point discrete Fourier transform (DFT), filtering the symbol vector with a frequency domain Gaussian filter, generating a sample vector from the filtered symbol vector utilizing an inverse DFT, and modulating the sample vector utilizing GMSK. In some cases, the uplink modulation may be based on a narrowband resource allocation received from a base station.

The invention discloses a target detection method based on structural type Haar and Adaboost. The method includes following steps: creating a sample, and obtaining a positive example sample vector description file and a counter example sample description file; then creating a structural type Haar characteristic, and performing training according to the structural type Haar characteristic to obtain a weak classifier and a strong classifier; and obtaining a cascaded classifier, and finally performing target detection by employing the obtained cascaded classifier to obtain a final detection result. According to the target detection result obtained by the method, good detection precision can be guaranteed, the false detection rate is reduced, the training time is effectively reduced, and the method can be applied to the fields of intelligent traffic detection, video monitoring, and image identification and search etc.