487 results about "Neighbor algorithm" patented technology

A method of automatically labeling of unlabeled text data can be practiced independent of human intervention, but that does not preclude manual intervention. The method can be used to extract relevant features of unlabeled text data for a keyword search. The method of automated labeling of unlabeled text data uses a document collection as a reference answer set. Members of the answer set are converted to vectors representing centroids of unknown groups of unlabeled text data. Unlabeled text data are clustered relative to the centroids by a nearest neighbor algorithm and the ID of the relevant answer is assigned to all documents in the cluster. At this point in the process, a supervised machine learning algorithm is trained on labeled data, and a classifier for assigning labels to new text data is output. Alternatively, a feature extraction algorithm may be run on classes generated by the step of clustering, and search features output which index the unlabeled text data.

The invention discloses a voice synchronous-drive three-dimensional face mouth shape and face posture animation method. A user can input new voice information, and the new voice information can be preprocessed to combine mouth shape animations and face posture animations which are synchronous with voice on the face head of a virtual man. The method specifically comprises two stages. In a training stage, voice visualization modeling can be achieved through a k-nearest neighbor algorithm (KNN) and hidden Markov model (HMM) mixed model. In a combining stage, the user can input new voice information, characteristics of voice signals are extracted, face posture and mouth shape sequence parameters corresponding to the voice signals can be generated through the KNN and HMM mixed model and are processed in a transition mode, and X face open source software is used to combine delicate and abundant three-dimensional face animations. The method has significant theoretical study value and has wide application prospect in the fields of visual communication, virtual meetings, games, entertainments, teaching assistance and the like.

The invention belongs to the technical field of privacy protection in cloud computing and discloses a privacy protection method in face authentication system based on edge calculation. The method comprises the steps that: a camera collects a face image and uploads the face image to a connected edge computing node, an authority assigning mechanism sends a permission vector Li of a user to the edgecomputing node after collecting user face information, an edge computing device carries out feature extraction on a face image of a user requesting registration to obtain a feature vector by using a method based on a deep convolutional neural network, a safe nearest neighbor algorithm is performed to encrypt the feature vector of a registered user, and a secret sharing homomorphic algorithm is performed and n sub-secrets are generated according to the feature vector and are distributively stored in n edge computing devices. The direct upload of sensitive data to a cloud server is avoided, theprivacy of the face data is protected, and the fault tolerance of a system is improved. Results prove that the accuracy of face recognition under a ciphertext of the present invention is completely equal to the accuracy of face recognition under a plaintext.

The invention discloses a deep feature and transfer learning-based rolling bearing fault diagnosis method under variable working conditions, relates to the technical field of fault diagnosis, and aimsto solve the problem of low state identification accuracy of different fault positions and different performance degradation degrees of a rolling bearing under the variable working conditions. The method comprises the following steps: firstly, carrying out feature extraction on the vibration signal frequency domain amplitude of the rolling bearing by adopting SDAE to obtain vibration signal deepfeatures, and forming a source domain feature sample set and a target domain feature sample set; then, adopting the JGSA to carry out domain adaptation processing on the source domain feature sample and the target domain feature sample, the purpose of reducing distribution offset and subspace transformation difference of feature samples between domains is achieved, and domain offset between different types of feature samples is reduced. And finally, completing rolling bearing multi-state classification under variable working conditions through a K nearest neighbor algorithm. Compared with other methods, the method disclosed by the invention shows better feature extraction capability under the variable working condition of the rolling bearing, the sample feature visualization effect of therolling bearing is optimal, and the fault diagnosis accuracy of the rolling bearing under the variable working condition is high.

The invention discloses a method for quickly identifying two-dimension code system type in images, comprising a learning training process and a classification identifying process. The learning training process is as follows: collecting and building a sample image set of various two-dimension code images; converting each sample image into a grey image, performing Gaussian smoothing filtering and binaryzation to obtain binaryzation images; scanning prospect boundaries of the binaryzation images in the horizontal and vertical directions, obtaining an outer boundary point set of the two-dimension code; enabling the two-dimension code to be horizontal by rotating images, achieving horizontal correction of the two-dimension code; performing partitioning, combining and normalizing for the two-dimension code; performing fast wavelet transform for the normalized sample image to obtain a wavelet characteristic sample set. The classification identifying process is as follows: extracting wavelet characteristic of the to-be-identified image to build a distance measurement model; using the K nearest neighbor algorithm to identify code system type. The method is convenient and quick, has real-time performance, accuracy and high identification rate.

The invention discloses a method and system for locating a mobile terminal, and belongs to the field of wireless service supporting in mobile communication. The method integrates the advantages of a support vector machine and a K-nearest neighbor algorithm, and through the stages of off-line drive test sampling and model training, two layers of SVM locating models and KNN locating models are constructed. In an on-line locating stage, the two layer of the SVM locating models and the KNN locating models are combined, so that the locating precision of the mobile terminal is greatly improved. The method achieves compromise and balance between the algorithm complexity and the locating precision, and compared with a traditional locating method, the method effectively improves the locating precision, greatly reduces calculated amount and improves the locating speed.

The present invention discloses a method for providing prediction of content usage in a website. The method comprising the steps of: real time monitoring traffic of visitors in a website, tracking visitors that are visiting the monitored website to identify one or more parameters relating to user profile, navigation path or content usage, applying at least one statistical algorithm on identified parameters, said algorithm is at least one of: clustering algorithm, nearest neighbor algorithm or probability algorithm and defining content replacement and recommendation for visitors when visiting the monitored website according analysis results of the at least one statistical algorithm.

The invention discloses an indoor positioning method based on distribution of channel state information (D-CSI), which comprises a training stage and a positioning stage. In the training stage, a fingerprint database is established by using fingerprint information of each reference point location. In the positioning stage, firstly fingerprint information of the location of a user to be positioned is calculated, then a distance value between the fingerprint information of the location of the user to be positioned and the fingerprint information of each reference point location is calculated by using KL distance, and the location of the user to be positioned is acquired by using a K-nearest neighbor algorithm. The indoor positioning method makes full use of the frequency diversity between different sub-bands and the diversity between different antennas, the time stability and the space difference of a location characteristic fingerprint can be enhanced, and thus the positioning accuracy is improved.

The invention discloses an intelligent urban construction examining and approving method based on a case-based reasoning technology. The intelligent urban construction examining and approving method based on the case-based reasoning technology comprises the following steps of constructing an examining and approving case library; inputting new examining and approving case and model parameter information; submitting jogs to a Hadoop cluster to search KNN (k-nearest neighbor algorithm) Map Reduce cases; statistically analyzing a searching result on the basis of a 'weighted integral model'; evaluating and correcting the cases; and performing distributed full-text searching on examining and approving data. The intelligent urban construction examining and approving method has the advantages that by the method, the circumstance of manual examination and approval application at present can be changed, the work efficiency is improved, the basis on examining and approving is increased, and an examining and approving process is intelligent. Distributed searching can be carried out by using a Hadoop frame and a MapReduce frame through a cloud computing center, and a distributed case searching model based on the case-based reasoning technology is established. The 'weighted integral model' is creatively raised to statistically analyzing searched similar cases, and a guidance which is beneficial to new examining and approving cases is obtained.

The present invention facilitates identifying applications based on communicated packets between applications. Characteristics of communicated packets are used to identify the packet as being part of a communication between applications. Identification can be accomplished through the use of packet fingerprints or through a K nearest neighbor algorithm.

The invention provides a digit recognition method for identification cards. The digit recognition method comprises the following steps: acquiring a classifier training file; transforming a color image to be identified into a gray level image; accurately positioning digit element positions of each identification card; accurately segmenting digit characters of each identification card; verifying whether a candidate region is a real digit region of each identification card; carrying out digit recognition on the basis of a nearest neighbor algorithm; judging whether the current digit element is the last group of digit element. According to the digit recognition method provided by the invention, positioning, segmentation and identification of digits on the identification cards are carried out by adopting an image processing technology; the recognition precision is relatively high, no special equipment is needed and the cost is low; the digit recognition method has strong applicability to various environments and can be used for reading information of multiple identification cards simultaneously.

The invention discloses a locating method in a wireless LAN and a device thereof. The locating method comprises the following steps: acquiring a sample set composed of a plurality of sample points closest to the location to be determined according to a nearest neighbor algorithm; acquiring central positions of the sample points in the sample set, and the distance between the sample points and the central positions; judging whether the sample points with the distance from the central positions which is more than a preset standard exist, if yes, deleting the sample points with the distance from the central positions which is more than the preset standard from the sample set, and returning to the previous step; otherwise continuing to the next step; and determining the location to be determined according to the various sample points included in the sample set. The locating method in the wireless LAN and the device thereof help improve precision and credibility of locating results in the wireless LAN.

The present invention discloses a polarized SAR image classification method based on the semi-supervised depth distance metric network, and the technical problems that the traditional depth learning only considers the non-linear relationship between the sample characteristics and the classification accuracy is not high when the number of marked samples is relatively small are solved. The method comprises the following steps: inputting to-be-classified polarized SAR image data; solving a neighboring sample of the marked sample; constructing the loss function of the semi-supervised large boundary neighbor algorithm; initializing parameters of the network; pre-training the network; carrying out fine tuning on the network; carrying out classification prediction on the unmarked samples; and outputting a classification result image and classification accuracy of the to-be-classified polarized SAR image. According to the method disclosed by the present invention, by constructing a depth distance metric network, a popular learning regular term is added to the large boundary neighbor algorithm, so that problems of the influence of insufficient marked samples on the classification accuracy and the waste of information of a large number of unmarked samples are overcome; and the characteristics learned in the method of the present invention fully depicts intrinsic attributes of the samples, and the method can be applied to the earth resources survey, military systems and other technical fields.

The invention discloses an indoor passive positioning method based on geographic fingerprints. The method is applied to the technical field of indoor positioning and comprises: in a situation that a mobile terminal is in an off-line state, collecting a received signal strength (RSS) value of each reference point where the mobile terminal is by utilizing an access point device; calculating a prior probability that the access point device collects the RSS values in a period; establishing a geographic fingerprint database according to a coordinate position of each reference point, the RSS values and the prior probability; when the mobile terminal is in an on-line state, testing an RSS value of each test point where the mobile terminal is; matching the test points with reference points in the geographic fingerprint database based on a prior probability nearest neighbor algorithm according to the RSS value at each test point where the mobile terminal is; and determining the positioning position of the mobile terminal. According to the invention, the prior probability is applied to the K-nearest neighbor algorithm for positioning, and the positioning accuracy is improved at the same time and space complexity.

The invention provides a point cloud feature point detection method and a cloud point feature extraction method. The checking method comprises the following steps that: taking one data point in pointcloud data as a sphere center to establish a local spherical coordinate frame, and finding k pieces of neighbourhoods through a k nearest neighbor algorithm; independently connecting the sphere centerwith each nearest neighbor point to obtain k pieces of line segment groups; according to the horizontal projection included angles of the line segment groups, sorting k pieces of line segment groups,checking each line segment group through a Laplace operator, and determining whether the data point is a feature point or not; and according to the above steps, processing each data point in the cloud point data, and obtaining all feature points of the cloud point data. The point cloud feature extraction method comprises the following steps that: according to the above detection method, determining feature points, and recording the regional connection information of each feature point; and determining a sequential connection relationship between the feature points, connecting the feature points, and forming a segmented feature polygon to realize region segmentation. The amount of labeled feature points is small, feature points are extracted orderly, and feature lines are conveniently reconstructed.

The invention provides a method for detecting an abnormal time sequence without a class label, and aims at solving the problems that ideal effect of segmenting fixed points based on satellite remote detecting data cannot be achieved, the clustering number is manually set during layer-based clustering, and offline and online abnormality detection methods for the label time sequence without the class label are currently not developed. According to the technical scheme, the method comprises the steps of 1, segmenting the satellite remote detecting historical data according to the cycle property of the satellite remote detecting data to obtain the time sequence without class label, namely, X={x1, x2..., xn}; 2, performing adaptive layer-based clustering for the X={x1, x2..., xn} obtained in step 1, and determining and deleting the abnormal sequence in the time sequence without the class label to obtain the formulas as shown in the specification; 3, adopting the formulas as shown in the specification as samples, performing mode matching for the formula shown in the specification by the nearest neighbor algorithm according to the matching threshold, so as to finish the abnormal satellite remote detecting data detection. The method is applied to the field of satellite data detection.

The invention discloses a multilayer bitmap color feature-based image retrieval method. In the method, fast clustering is performed on an image with rich color information to obtain rational statistical distribution centers of each color cluster, and based on the rational statistical distribution centers, features capable of reflecting color differences among different distribution layers of the image are extracted to perform image retrieval. The method comprises the following steps of: first performing meshing on a color space of the queried image, counting the numbers of pixel points in each mesh and selecting the mesh with a number local maximum; then quickly generating each color cluster and the rational statistical distribution centers thereof by adopting a novel distance optimization algorithm and an equal-average nearest neighbor algorithm search (ENNS) algorithm in a K-average clustering algorithm, and on the other hand, performing space sub-block division on the queried image and calculating a Gaussian-weighted color average of sub-blocks; next comparing the color average of the image sub-blocks with the rational statistical distribution centers of the color clusters to extract the features of a K-layer bitmap; and finally performing the matched searching of the image features by combining the similarity measurements of the rational statistical distribution centers of the color clusters and the bitmap.

The invention discloses a signal modulation type identification method and a signal modulation type identification system. The method comprises the following steps: preprocessing a signal to be identified, extracting a predetermined number of feature parameters, and using a feature vector formed by the predetermined number of feature parameters to represent the signal to be identified; using an optimal projection matrix to extract the features of the signal to be identified, and projecting the signal to be identified into a low-dimensional feature subspace, wherein the optimal projection matrix is obtained through a locality preserving projection algorithm; and calculating the Euclidean distance between the signal to be identified in the low-dimensional feature subspace and a training signal of which the signal modulation type is known, and determining the signal modulation type of the signal to be identified based on the nearest neighbor algorithm of Euclidean distance. According to the technical scheme, the signal to be identified is projected into the low-dimensional feature subspace by using the optimal projection matrix, which reduces the amount of calculation; and the optimal projection matrix obtained through the locality preserving projection algorithm reduces the deviation of the signal in the process of projection, and has better robustness and higher identification rate.

The invention discloses a dental jaw movement locus recording device and a dental jaw relationship transferring method. The dental jaw relationship transferring method comprises the following steps of carrying out digital dental jaw recurrence; extracting an optimal occlusion locus from a plurality of obtained dental jaw movement loci; driving an integral three-dimensional dental jaw by an optimal occlusion locus; and simulating the real lower jaw movement relative to an upper jaw so as to finish the transfer of the dental jaw relationship. According to the dental jaw relationship transferring method disclosed by the invention, movement loci are analyzed by a machine self-learning system based on a distance-weighted nearest neighbor algorithm so as to bring good robustness for noise in the trained movement locus data, k neighbor weighted averages are taken, the influence of an isolated noise movement locus is eliminated so as to obtain a real lower jaw movement locus, the one-sidedness for obtaining a single lower jaw movement locus is avoided, and the real lower jaw movement locus is utilized to drive the integral three-dimensional dental jaw to simulate the real lower jaw movement relative to the upper jaw so as to realize the transfer of the dental jaw relationship.

The invention discloses a multi-target identification tracking method for a road traffic scene. According to the method, the road traffic field environment and obstacle characteristics are combined. Dynamic obstacles around a road site are detected through a laser radar. The method comprises the following steps of: firstly, preprocessing laser radar data and converting the data into a binary image, secondly, extracting a dynamic obstacle by using a background difference method, carrying out clustering analysis, further carrying out classification recognition by using a Point CNN algorithm, obtaining pose transformation of the obstacle by using an iterative nearest neighbor algorithm, and finally, carrying out tracking by using a UKF for obstacle classification. The multi-target identification tracking method provided by the invention has good real-time performance, identification precision and tracking precision, and can effectively realize safety monitoring of the road traffic scene.

The invention relates to a SIFT and Otsu matching based colored eyeground image splicing method. The method is realized by that (1) a green channel of a reference eyeground image and a target eyeground image is extracted; (2) the green channel is preprocessed via contract limited adaptive histogram equalization and two-dimensional Gaussian filtering to enhance the contract between vessels and tissue background; (3) SIFT characteristics of the image are extracted; (4) characteristic points of the two images are matched initially via a nearest neighbor and a next nearest neighbor algorithm; (5)an Otsu algorithm is used to screen the initial matching points to obtain correct matching characteristic point pairs; (6) a random sampling consistency algorithm is used to calculate an affine transformation matrix to carry out space transformation o the images and complete registration; and (7) the registering images are fused in a maximum value algorithm, and splicing is completed. According tothe invention, SIFT is combined with the Otsu method, the disadvantage of non-ideal registering effect caused by mismatching points is overcome, and the method has significance in the field of diagnosing diabetes and pathology degree thereof.

The invention discloses an indoor WLAN fingerprint quick positioning method based on support vector regression. The method comprises the steps of 1 an off-line stage, wherein a feature fingerprint database of reference points is built, and a relation model of feature fingerprints and reference point locations is trained; and 2 an on-line stage, wherein coarse positioning is conducted through the relation model, and then precise positioning is conducted through a weighted K nearest neighbor algorithm. By adopting the indoor fingerprint quick positioning method based on support vector regression, the on-line fingerprint matching range can be narrowed, noise jamming is reduced by extracting a characteristic value of signal intensities, therefore, the positioning precision is improved, and the fingerprint matching speed is increased.

The invention provides and establishes a multi-agent frame structure based tour multimedia information personalized service system. The system adopts the Agent technology, finds and mines interests of users through constant independent study, and realizes automatic user interest modeling by adopting a vector included angle based K-nearest neighbor algorithm. As for each phase of a system working process, modules with relatively independent functions are divided, and each module is packaged into an Agent agency to form a multi-agent framework; and under the framework, the agencies are separated and cooperated so as to realize the characteristics of intelligence, personalization and the like of the multimedia information service. The system mainly comprises five functional modules, namely a user interface Agent, a filter sequencing Agent, an interest study Agent, a monitoring Agent and an information searching Agent. The method proposed by the invention has higher theoretical significance and application value for establishing high-quality intelligent tour information service.

The invention relates to a potato image acquisition device based on RGB-D cameras and a method for identifying and locating a bud eye. The image acquisition device consists of three RGB-D cameras thatacquire the color maps and the depth maps of a potato sample from three different angles. The image processing part comprises obtaining a potato target image by preprocessing the color maps and performing a mask-based target extraction method; then training a classifier by using an Adaboost algorithm and a Haar-like feature, identifying a bud eye area, and obtaining the two-dimensional coordinates of the bud eye area; calibrating the RGB-D cameras to obtain the internal parameters and the external parameters of respective RGB-D cameras, and generating point cloud in combination with the depthmaps and the color maps acquired by the corresponding cameras; successively registering the three groups of point cloud by using an iterative nearest neighbor algorithm to obtain the three-dimensional model of the potato sample; then converting the obtained two-dimensional coordinates of the bud eye into coordinates in a three-dimensional space where the three-dimensional model of the potato sample is located, so as to achieve the three-dimensional positioning of the potato bud eye and lay a foundation for realizing the automation of the potato seed dicing.

The invention discloses a video image stabilization method and system based on feature matching and motion compensation. The method comprises the following steps: selecting a local feature matching area of a video frame image; extracting video frame image feature points in the local feature matching area of the video frame image by adopting an SURF algorithm, and calculating a corresponding SURF feature point descriptor; carrying out local feature point matching on the video frame image by adopting an improved fast approximate nearest neighbor matching algorithm; and calculating affine transformation parameters of the video frame image through a least square method according to the matching result, and then, carrying out global motion compensation on the video frame image through a bilinear interpolation method according to the calculated affine transformation parameters. The method and system combine local feature matching and motion compensation, thereby accelerating algorithm processing speed, increasing the process of adopting a weight value screening method, a bidirectional matching method and a K-nearest neighbor algorithm to screen out the final matching point, and improving feature matching accuracy. The method and system can be widely applied to the field of image processing.

The invention provides a disease detection method and device based on a wearable device and the wearable device. The method comprises the steps that original electrocardio signals of a user are collected through an electrocardio sensor; feature vectors corresponding to the original electrocardio signals are determined, wherein the feature vectors comprise time-domain feature data of the original electrocardio signals and frequency-domain feature data of the original electrocardio signals; a disease type corresponding to the original electrocardio signals is determined according to the feature vectors and a trained large-distance nearest neighbor algorithm corresponding to the original electrocardio signals. According to the technical scheme, the accuracy on disease detection of the user can be greatly improved, and the user can recognize and predict health risk as early as possible through the wearable device.

The invention relates to a robot SLAM object state detection method in a dynamic sparse environment. Firstly, image collecting of the environment is carried out through a vision sensor, and a feature vector set of images is obtained through a SURF describor; then matching of an image of the current moment and an image of a historical moment is carried out according to a nearest neighbor algorithm, whether matching is successfully carried out is detected by means of an RANSAC algorithm and whether an object of the current moment is consistent with the object of the historical moment is judged according to the detection result; depth information of the object is obtained in a parallax error method, and a world coordinate and a relative position difference of the object at the two moments are obtained according to a plane geometrical relationship; ultimately, an acceptance region is obtained by combination with hypothesis testing, and the state of the object is judged by detecting whether the relative position difference of the object is within the acceptance area. When the state of the object in the environment is detected, influences of positioning and measurement errors of a moving robot on an object position observation value are taken into consideration, and object state judgment accuracy is improved.

The invention discloses an odometer method and device for sports equipment; the odometer method comprises: acquiring current pose of sports equipment, current second laser frame and a local point cloud map; calculating according to the current pose, the second laser frame and the local point cloud map to obtain a current optimal pose; and other steps. In the odometer method, initial relative exercise amount and pose are acquired through the number of turns of a coded disc, laser inter-frame matching is performed by means of iterative nearest neighbor algorithm and the like, and accurate relative exercise amount and pose are acquired accordingly; the pose acquired in the above step is used as an initial value to perform matching on the laser frame and the local point cloud map so as to obtain more accurate pose and exercise amount that are then output. The odometer device provided herein enables the poor precision and accumulative error defects of an existing odometer to be overcome, and is applicable to general or special computing system environments or configurations.

The invention relates to text classification of natural language processing, and specifically relates to a class center vector text classification method based on dependency, word class and a semanticdictionary. To overcome the semantic defect of a feature selection algorithm based on statistics, the invention introduces the dependency, the semantic dictionary and the word class to optimize and cluster text features, provides an improved weight calculation formula, and further provides an improved class center vector text classification method. The text classification method of the inventionhas advantages of both high classification efficiency of a traditional class center vector method and high classification precision of a K-nearest neighbor algorithm, and can be widely used in variousclassification systems.