39 results about "Minimum description length" patented technology

The present technique provides for the analysis of a data series to identify sequences of interest within the series. The analysis may be used to iteratively update a grammar used to analyze the data series or updated versions of the data series. Furthermore, the technique provides for the calculation of a minimum description length heuristic, such as a symbol compression ratio, for each sub-sequence of the analyzed data sequence. The technique may then compare a selected heuristic value against one or more reference conditions to determine if additional iteration is to be performed. The grammar and the data sequence may be updated between iterations to include a symbol representing a string corresponding to the selected heuristic value based upon a non-termination result of the comparison. Alternatively, the string corresponding to the selected heuristic value may be identified as a sequence of interest based upon a termination result of the comparison.

The invention discloses a moving object destination prediction algorithm. The method comprises the steps that firstly, an improved track segmentation method with the minimum description length (IMDL)is provided, weight parameters are introduced, meanwhile, a Dijkstra algorithm is used for converting the track segmentation problem into the shortest path problem of an undirected graph, the optimalsegmentation of a track is obtained, and therefore the track is simplified; then, a deep learning prediction model EP-LSTM based on embedded processing and long-short-term memory fusion is provided, and a high-dimensional input vector is converted into an embedded vector to serve as input of the model. The method has the advantages that the problems of data sparseness and long-term dependence in moving object destination prediction are effectively solved, and an excellent prediction effect is achieved by performing a large number of experimental verifications on a real data set.

The invention discloses a convolutional neural network-based moving object destination prediction method, which comprises the following steps of: introducing a parameterized minimum description length(PMDL) strategy to carry out optimal segmented representation on an original trajectory, weakening the difference degree between similar trajectories, and enhancing respective important characteristics of the different trajectories; proposing a trajectory pixelated representation method (PRT), and converting the one-dimensional trajectory sequence into a two-dimensional pixel image; and intercepting important feature parts from the track image and inputting the important feature parts into a convolutional neural network (CNN) model, and performing feature extraction and destination prediction. The method has the advantages that a one-dimensional track sequence is converted into a two-dimensional pixel image, and destination coordinates are determined by mining spatial features of the track image; the invention provides an effective solution for the data sparsity problem which is common in destination prediction but cannot be solved in many traditional methods, and high-accuracy destination prediction is realized.

The invention provides a compressed sensing synthetic aperture radar radio frequency interference suppression handling method. The method includes the following steps: (1) building a compressed sensing model of a radio frequency interference (RFI) signal based on an obvious sparse feature of a RFI component in frequency domain, (2) estimating sparseness of the RFI component by using a greedy algorithm to combine with minimum description length, (3) estimating a RFI signal component and filtering directly in time domain to an echo signal of each pulse, and (4) achieving imaging process by using an conventional compressed sensing synthetic aperture radar (SAR) reconstructing algorithm. The compressed sensing synthetic aperture radar radio frequency interference suppression handling method is based on the sparse feature of the RFI component in the frequency domain and builds a RFI sparse model based on a compressed sensing theory, restrains narrow-band and broadband RFI signals effectively by using the iterative greedy algorithm to estimate and filter the RFI component in the echo signal.

The invention claims to protect a pseudo-code sequence blind estimation method of a synchronous DS-CDMA (Direct Sequence-Code Division Multiple Access) signal containing residual frequency offsets based on the combination of eigen decomposition and a DPLL (Digital Phase Lock Loop), and belongs to the technical field of signal processing. The pseudo-code sequence blind estimation method comprises the following steps: on the premise of knowing a pseudo-code sequence period and a chip width, segmenting a received signal according to a single-fold pseudo-code period in order to construct a signal covariance matrix and performing eigenvalue decomposition; estimating the user number of the DS-CDMA signal according to an MDL (Minimum Description Length) principle in order to further extract a feature vector corresponding to a maximum eigenvalue which is equal to the user number, and realize rough extraction of each user pseudo-code sequence; lastly, inputting the pseudo-code sequence undergoing the rough estimation into the DPLL to eliminate the residual frequency offsets, wherein an I-path signal output by the DPLL is taken as a pseudo-code sequence undergoing final blind estimation. By adopting the method, aliasing residual frequency offsets in the pseudo-code sequence are eliminated through the DPLL, and the estimation problem of the pseudo-code sequence of the DS-CDMA signal containing the residual frequency offsets under a low signal-to-noise ratio is solved effectively; moreover, an algorithm is easy to operate and is not limited by a spreading code type, and the method has important engineering practical significance.

The invention provides a Bayesian network structure learning method and system and reliability model construction method. The learning method comprises the steps that relevant data of each module of a system are acquired and are abstracted into corresponding Bayesian network nodes; according to the relevant data of the system, a node mutual information matrix and a node to BIC scoring matrix are calculated and established; a corresponding adjacency matrix is established; and according to relevant data of the system, a family description length value corresponding to each node and the sum of the family description length values of all nodes are calculated. Bordered operation is carried out. Whether a BN structure continues to be updated is judged through a minimum description length principle. Side flipping is carried out. Whether the BN structure continues to be updated is judged through the minimum description length principle. In bordered operation, an MDL standard and a BIC evaluation standard are combined to avoid repeated climbing. The learning method still belongs to a greedy search algorithm. Due to combined specific scoring standards, the search efficiency and the accuracy rate are greatly improved.

The invention discloses a method for detecting the number of signals under the condition of mixing of uncorrelated and correlated signals in a uniform linear array. The method for detecting the number of the signals is characterized by including steps of constructing a difference matrix, dividing the difference matrix into superimposition submatrixes, constructing a connecting matrix, and acquiring the number of the correlated signals in incident signals by solving a rank of the connecting matrix; and acquiring a rank of an array covariance of the incident signals by a ratio criterion based on singular value decomposition, and acquiring the estimated number of the uncorrelated signals. Large quantities of simulation experiments prove that the method for detecting the number of the signals is superior to an MDL (minimum description length) method/ an AIC (akaike information criterion) method, an MENSE method and an SRP (smoothing rank profile) method which are pre-processed by an FBSS (forward-backward spatial smoothing) under the conditions of fewer snapshots and low signal noise ratio.

The invention discloses a greedy algorithm-based self-adapting compression perception signal restoring method. On the basis of an OMP (Orthogonal Matching Pursuit) algorithm, a detector used for judging whether a signal component exists in residual vector or not is designed, the detector is obtained through a hypothesis test model in the signal detection theory, the threshold of the detector is given through preset false alarm probability, when the threshold of the detector is smaller than the given threshold, the state that the signal component does not exist in residual vector can be judged, at the moment, iteration is stopped in the algorithm and restoring value of the signal is obtained, otherwise, iteration is continued. According to the method, the greedy algorithm serves as a channel estimation method without depending on the multipath number of a channel; compared with the MDL (Minimum Description Length), the method has the advantage that communication resources are saved due to the fact that observation for multiple times is not required; the method has the advantage of smaller calculation amount of the greedy algorithm, the problem that the requirement of the greedy algorithm on the real-time capability of sparse signal restoration is relatively high is solved, and the method can be applied to the condition that the signal sparsity is unknown.

The invention discloses a vehicle trajectory clustering method based on a bag-of-word model and metric learning. The method comprises the following steps: extracting vehicle trajectory point motion parameters; dividing the vehicle track into a plurality of optimal homogeneous vehicle track segments based on a minimum description length principle, and obtaining statistical features of each segment to obtain a vehicle track segment feature set; coding each optimal homogeneous vehicle track segment into a fixed-length vector based on a bag-of-word model and a non-convex low-rank constraint metric learning method so as to obtain a feature descriptor of the optimal homogeneous vehicle track segment; and gathering the feature descriptors through a K-means algorithm and the non-convex low-rank constraint metric learning method to realize vehicle trajectory clustering. According to the method, speed, acceleration and steering angle information is introduced, feature coding is carried out on vehicle tracks, and then the vehicle tracks in different time periods and areas are gathered; in addition, a non-convex low-rank constraint-based metric learning method is provided to improve the similarity measurement performance.

The invention discloses a graph digest algorithm based on node importance, which comprises the following steps of: establishing a node set, and carrying out importance index calculation and sorting on all points in the set; according to the importance sequence, nodes which are low in importance and are spaced by two hops are selected to form node pairs, new graph abstract vertexes are generated through combination, and meanwhile, a graph abstract vertex set is updated until all abstract nodes reach an algorithm threshold value; and traversing the summary vertex set, selecting summary point pairs to generate edges according to an MDL (minimum description length) principle, generating a correction edge set according to the original graph and the summary graph, and finally outputting the summary graph and the correction edge set. According to the method, abstracting can be carried out according to the importance of the nodes in the large graph, the defect that important nodes are merged in advance in a traditional graph abstracting algorithm is overcome, and the method has wide practical value and application prospects in the field of graph processing.

The invention provides a model optimization method, device and equipment and a readable storage medium, and relates to the technical field of deep learning, and the model optimization method comprises the steps: determining M network models; determining description lengths of the M network models based on a to-be-analyzed data set; and determining a target network model with the minimum description length from the M network models as an optimized network model, and carrying out data processing based on the optimized network model. In this way, the problems that the optimization degree of an existing deep learning network model is weak, and the complexity of the deep learning network model cannot be effectively reduced are solved.

The invention provides a target number and target angle estimation method based on an MDL criterion, and belongs to the field of array signal processing. The method has an advantage that under the condition of a non-uniform noise, correct estimation of a target number and a target angle can be realized. The method mainly comprises the following steps: firstly, taking the target number and a whitening vector as unknown parameters, and establishing a target function taking a minimum description length (MDL) as the criterion; solving a minimum value of the MDL target function by using a genetic algorithm so as to obtain an estimated value of the target number and an estimated value of a whitening vector; whitening a received signal covariance matrix by using the estimated value of the whitening vector; and finally, according to the whitened covariance matrix, the estimated value of the whitening vector and the estimated value of the target number, realizing accurate estimation of the target angle.

A method for generating a graphical depiction of summarized event sequences includes receiving a plurality of event sequences, each event sequence in the plurality of event sequences including a plurality of events, and generating a plurality of clusters using a minimum description length (MDL) optimization process. Each cluster in the plurality of clusters including a set of at least two event sequences in the plurality of event sequences that maps to a pattern in each cluster. The pattern in each cluster further includes a plurality of events included in at least one event sequence in the set of at least two event sequences in the cluster. The method includes generating a graphical depiction of a first cluster in the plurality of clusters, the graphical depiction including a graphical depiction of a first plurality of events in the pattern of the first cluster.