11 results about "Euclidean vector" patented technology

A plurality of motion vector candidates is extracted, which includes motion vectors of a first set of blocks and a second set of blocks each adjacent to an object block. The first set of blocks has the same size as the object block. The second set of blocks has a smaller size than the object block. A first predicted motion vector of the object block is selected from the plurality of motion vector candidates. A residual cost of each neighboring motion vector is calculated. Each neighboring motion vector points a point within a predetermined range from a point pointed by the first predicted motion vector. A second predicted motion vector having the minimum residual cost is selected from the plurality of neighboring motion vectors. If the first predicted motion vector is equal to the second predicted motion vector, the first predicted motion vector is set as the motion vector of the object block. If the first predicted motion vector is not equal to the second predicted motion vector, a search region pointed by the first predicted motion vector is searched by block matching.

The invention discloses an electroencephalogram signal feature extraction method combining with a public space mode algorithm and an EMD. Firstly, electroencephalogram signals of a subject are selected as a training set and a testing set, and the signals of the single subject in two channels C3 and C4 are pre-processed; then, experience mode decomposition is conducted on the prepressed EEG signals to obtain a series of intrinsic mode functions IMFi, and energy spectrum diagrams of all the intrinsic mode functions are drawn; then, previous three order IMF components of the channels C3 and C4 undergoing a single pass test are merged to form a N * T matrix Xi, wherein N represents IMF number, T represents the number of sampling points undergoing one test, the whole experiment process includes G groups of tests, and G groups of vector matrixes are obtained and are divided into G1 groups of test vector matrixes and G2 groups of training vector matrixes which are respectively subjected to public space mode decomposition. CSP filtering is conducted on the intrinsic mode functions decomposed by utilizing the three channels and the EMD, frequency domain information of the EMD is added on the basis of the CSP, and the problem that the CSP is lack of the frequency domain information is solved.

The invention discloses a data enhancement method applied to neural machine translation. The method comprises the steps that word alignment processing is conducted on bilingual training linguistic data to obtain word alignment information; minimum translation units contained in the bilingual training linguistic data are determined according to the word alignment information; all the minimum translation units are calculated according to a preset vector representation calculation rule, and corresponding vector values are obtained; the similarity value between all the minimum translation units iscalculated through a cosine formula; whether the similarity value exceeds a threshold or not is judged, and if yes, the position of the minimum translation unit, corresponding to the similarity valueexceeding the threshold, in the bilingual training linguistic data is exchanged, and new bilingual training linguistic data is obtained. By means of the method, the new bilingual linguistic data canbe formed, and the training linguistic data can be effectively increased; the structure information of the original linguistic data is enriched, and the improvement of the low-resource language translation quality of neural machine translation is achieved. The invention further discloses a data enhancement system and device and a computer readable storage medium which have the above advantages.

The invention discloses a dynamic multi-picture typesetting method and apparatus. The method comprises the following steps: determining an N-dimensional initial space vector, wherein an element of the N-dimensional initial space vector is a length-width ratio of each picture to be typesetted, and N is the number of the pictures to be typesetted; according to the pictures to be typesetted and the size of an area to be typesetted, obtaining all typesetting combinations, obtaining a new N-dimensional space vector corresponding to each typesetting combination, wherein each element of the N-dimensional space vectors is the length-width ratio of each picture in the corresponding typesetting combinations; establishing a distance model between the N-dimensional initial space vector and the new N-dimensional space vectors, wherein the distance model is used for calculating a distance between the N-dimensional initial space vector and the new N-dimensional space vectors; solving the distance model, and obtaining a new N-dimensional space vector with the smallest distance; and taking a typesetting combination corresponding to the new N-dimensional space vector with the smallest distance as a final typesetting combination.

The invention discloses a protein structure class prediction method based on weighted composition position vectors and a support vector machine. Firstly, protein structure class normal data sets are selected or established, weighting factors are introduced based on composition position vectors, and to-be-predicted protein sequences are represented through the weighted composition position vectors; the weighted composition position vectors are combined with the support vector machine, and the protein structure class prediction method is established through a direct multiclass classification strategy. According to the method, information of amino acid compositions is contained, position information of each amino acid residue in the protein sequences is also contained, the one-to-one correspondence function relation is formed between the position information and the protein sequences, the weighting factors are introduced to the composition position vector representation method, and prediction accuracy can be obviously improved by means of adjusting the weighting factors. The method is simple, rapid and sensitive, and is expected to be applied to other protein prediction fields.

The invention discloses a method for accurately estimating a spatial channel system function based on a sparse theory, which comprises the following steps of: establishing a sparse regular least square model based on a mixed l0 norm when channel estimation is carried out by using received data in consideration of the fact that the number of main channels in a scene is limited; in consideration ofthe non-convexity of the problem, using a convex optimization theory to carry out convex relaxation on the non-convexity, and giving a sparse regular least square model of a mixed l1 norm; and then reconstructing the received data according to a solving result, and adjusting a weight factor between a fitting error and channel sparsity in the optimization model according to an error between the received data and the original data until the weight factor meets a convergence condition. According to the method, the priori information of spatial channel sparsity can be fully utilized, so that the receiving data fitting error and the sparsity of the channel coefficient vector are optimized, the method better conforms to the actual situation, and the deviation of channel coefficient estimation caused by only considering the fitting error when a traditional method is used for estimating the spatial channel is effectively avoided.

The invention discloses an indoor positioning method based on four-dimensional code mapping, and the method comprises the steps: (S100), carrying out the Kronecker product of a horizontal-dimensionalfeature vector matrix and a vertical-dimensional feature vector matrix, generating a feature vector matrix of an antenna group, mapping the feature vector matrix into a four-dimensional code, carryingout the matching, obtaining an approximate code word four-dimensional code, and obtaining a pre-coding matrix; (S200), performing beam forming on the base station transmitting signal according to theprecoding matrix of the mobile terminal; and (S300), forming a virtual cell taking the mobile terminal by the to-be-positioned mobile terminal and the antenna group as the center; and forming a van virtual cell taking the mobile middle end as the center with surrounding mobile terminals, constructing an estimation function to obtain a positioning target function, solving the positioning target function by adopting a nonlinear least square algorithm, and taking the coordinate when the positioning target function takes the minimum value as the coordinate of the to-be-positioned mobile terminal.According to the method, beam matching of four-dimensional code mapping is combined with a virtual cell cooperation mode, and indoor real-time space positioning is achieved.

The invention relates to the field of deep learning, and discloses a method and device for training a word vector model, and the method for training the word vector model comprises the steps: obtaining first information which is used for reflecting the correlation degree between a target text and a context text; obtaining second information, wherein the second information is used for reflecting the association degree between the target texts; and training a word vector model according to the first information and the second information to obtain a word vector of the target text. According to the method provided by the embodiment of the invention, missing statistical information of a large number of unobserved text information pairs is made up, the problem that a co-occurrence matrix is extremely sparse is alleviated, and the accuracy of the word vector determined by the word vector model is effectively improved.