108 results about "Supervised clustering" patented technology

The invention discloses a cross-domain pedestrian re-identification method and system based on three stages, and the method comprises the steps: a domain adaptive learning stage: carrying out the processing of a source domain image and a target domain image through a domain adaptive network, calculating each loss, and updating the parameters of the domain adaptive network; a self-supervised training stage: carrying out the supervised training on the domain self-adaptive network through pseudo tags, calculating the loss of the triad difficult to sample, and updating network parameters; a jointloss training stage: constructing a joint loss training network, and defining label smooth regularization loss and difficult-to-sample triple loss; for a target domain image, inputting a joint loss training network, calculating the losses and updating the parameters of the joint loss training network, and carrying out the re-identification of cross-domain pedestrians. According to the cross-domainpedestrian re-identification method, the domain self-adaptation stage, the self-supervision clustering re-training stage and the joint loss learning stage are effectively integrated, and compared with a single training mode, the cross-domain pedestrian re-identification accuracy is further improved. The cross-domain pedestrian re-identification method has the advantages that the cross-domain self-adaptation stage, the self-supervision clustering re-training stage and the joint loss learning stage are effectively integrated.

The invention relates to an automatic extraction method for newly-increased construction land image spots in high-resolution remote sensing images based on the NDVI and the PanTex index. The method includes the first step of inputting a front time phase high-resolution remote sensing image and a back time phase high-resolution remote sensing image, and then conducting geometric fine correction and relative radiation correction, the second step of calculating a front time phase NDVI image, a back time phase NDVI image, a front time phase PanTex image and a back time phase PanTex image, the third step of conducting unsupervised clustering on the two NDVI images and the two PanTex images respectively, the fourth step of extracting a binary change image from vegetation to buildings, bulldozed and filled earth according to the two NDVI clustered images, the fifth step of extracting a binary change image from vegetation, bulldozed and filled earth to the buildings according to the two PanTex clustered images, the sixth step of extracting interfering ground object regions, the seventh step of conducting union operation on the two extracted change images and removing interfering ground object masks to obtain a newly-increased construction land image, the eighth step of segmenting the back time phase images, and the ninth step of calculating the proportion of changed pixels in each segmented image spot and extracting the newly-increased construction land image spots. Through the method, three types of newly-increased construction land image spots can be effectively extracted, and auxiliary information can be provided for land utilization change investigation.

The invention relates to an audio editing system. The audio editing system comprises a plurality of initial segmentation devices, a multi-sound track fusion device, an audio clustering device and a re-segmentation device, wherein the plurality of the initial segmentation devices are respectively used for initially segmenting audio streams from a plurality of sound tracks into a plurality of different paragraphs; the multi-sound track fusion device is used for integrating segmentation points of the plurality of the initial segmentation devices, selecting the audio stream of the optimal sound track between every two adjacent segmentation points, further getting a plurality of initially segmented fragments and fusing the plurality of the obtained initially segmented fragments into an uniform audio data file; the audio clustering device is used for performing clustering on the plurality of the initially segmented fragments under supervision based on a hierarchical clustering algorithm and clustering the initially segmented fragments belonging to the same nature to a category; and the re-segmentation device is used for training according to the clustering result of the audio clustering device to get a hidden Markov model corresponding to each type and performing Viterbi alignment segmentation on the uniform audio file to get the audio stream after re-segmentation. The accuracy in final speaker clustering can be improved through a high-precision speaker segmentation system.

The invention discloses a cloud network end cooperative defense method and system based on end-side edge computing, and relates to information security of an electric power industrial control system. The method comprises the following steps: setting an edge computing center at a terminal side, collecting industrial control system terminal equipment information and communication flow information, defining and identifying attribute characteristics of an electric power industrial control terminal by utilizing equipment fingerprints, automatically collecting the fingerprints of the electric power industrial control terminal equipment by utilizing an Nmap scanning method, establishing a training model by a decision tree algorithm, and achieving the dynamic fingerprint authentication of the terminal equipment; through setting a switch mirror image, intelligent monitoring host flow control and cloud computing center training flow baseline, industrial control terminal equipment flow anomaly detection is realized, and a cloud cooperative defense technology based on edge computing is realized. Through flow data acquisition, information entropy quantification flow characteristic attribute preprocessing and improved semi-supervised clustering K-means algorithm training, abnormal flow detection of the electric power industrial control intranet is realized, and cloud network real-time defense based on abnormal flow detection is realized.

Chat categorization uses semi-supervised clustering to provide Voice of the Customer (VOC) analytics over unstructured data via an historical understanding of topic categories discussed to derive an automated methodology of topic categorization for new data; application of semi-supervised clustering (SSC) for VOC analytics; generation of seed data for SSC; and a voting algorithm for use in the absence of domain knowledge / manual tagged data. Customer service interactions are mined and quality of these interactions is measured by “Customer's Vote” which, in turn, is determined by the customer's experience during the interaction and the quality of customer issue resolution. Key features of the interaction that drive a positive experience and resolution are automatically learned via machine learning driven algorithms based on historical data. This, in turn, is used to coach / teach the system / service representative on future interactions.

The present application provides a risk control method, device, apparatus and medium for user payment behavior, which relates to the technical field of data processing. The method comprises the following steps: acquiring a plurality of historical transaction result sample data corresponding to user payment behavior; Taking the transaction behavior characteristics and transaction behavior attributes in the sample data of historical transaction results as the input and output of semi-supervised clustering model respectively, constructing and training the semi-supervised clustering model to obtain the risk identification results; Inputting the transaction data corresponding to the user's payment behavior into the trained risk identification model to obtain the risk identification result; According to the risk identification result and the service type corresponding to the transaction data to be identified, determining the response operation to the transaction data to be identified. The present application is capable of risk identification in milliseconds with high speed and accuracy. The application can also automatically intercept the high-risk payment behavior, thereby improving thesecurity of the payment behavior of the user and reducing the property loss of the user.

The invention belongs to the image processing field and relates to a multi-target tracking method and a system based on kernel function unsupervised clustering. According to the method, a binocular camera is utilized to acquire left and right sequence images at one same time, and parameters of the binocular camera are utilized for image correction; a parallax error is calculated through extracting image characteristic points and matching characteristics; the acquired parallax error is utilized to calculate the coordinate position of a target characteristic point relative to the camera, namely the coordinate of the camera, ground calibration is accomplished, ground shadow characteristic points can be filtered according to height from the characteristic point to the ground, and ground shadow interference is eliminated; according to the three-dimensional coordinate characteristic point, in combination with the kernel function, unsupervised clustering is carried out for targets with undetermined category quantity, all characteristic points of one target are gathered to form one set, one category corresponds to the position and the direction of one observation value, a present frame of the target can be acquired in combination with the position and the direction of the previous frame target, namely the prediction position value and the prediction direction value, an optimum estimation algorithm is utilized to acquire the position and the direction of the optimum target, and thereby the multi-target fast tracking effect is realized.

The invention discloses a cone-beam computed tomography image correspondence and registration method. On the basis of a hybrid measure forest, denseness correspondence between super voxels between cone-beam CT images is generated by using a non-supervised clustering forest; and then denseness correspondence between super voxels between cone-beam CT images as well as cone-beam CT image registrationis estimated by using a forest-based measure. The step of training a hybrid measure forest is implemented as follows: extracting features of the super voxels; training an initial clustering forest and estimating similarity between the super voxels; estimating a weak mark; carrying out iterative enhancement of a forest-based measure based on the initial clustering forest; estimating flexibility consistency; carrying out on-line testing, inputting the features of the super voxels into the trained hybrid measure forest, estimating the similarity and correspondence between the super voxels, and estimating the flexibility consistency; and obtaining deformation parameters between the cone-beam CT images and realizing registration. Therefore, three-dimensional cone-beam computed tomography imagecorrespondence and registration can be established rapidly.

The invention provides a credit scoring model construction method and a device, equipment and a storage medium. The construction method comprises the steps of dividing an original unbalanced credit data set into a training set and a verification set; wherein a plurality of data samples in the original unbalanced credit data set comprise credit information of a plurality of users, and the pluralityof data samples are in one-to-one correspondence with the plurality of users; dividing the data samples in the training set into a majority class of training samples and a minority class of trainingsamples; clustering the majority class training samples by using an unsupervised clustering algorithm to generate a plurality of sample clusters; obtaining a preset number of balance training subsetsaccording to the plurality of sample clusters and the minority class training samples; and constructing a credit scoring model according to the obtained balance training subset, the verification set and a preset decision tree base classifier. According to the method, the classification performance of the credit scoring model can be improved.

The invention relates to a pattern recognition method capable of achieving cluster, classification and metric learning simultaneously. According to the method, the Bayesian theory is used to construct probability relation matrix P between a cluster and a category, and final cluster and category results are enabled to only rely on a cluster center through the matrix; accordingly, cluster learning and classification learning can be achieved simultaneously under a frame. As the statistical relation between the cluster and the category can be reflected through the matrix P, meaningful information can be mined from the P, and design of classifier is transparent. From the cluster point, cluster learning monitoring results are provided, and potential structures of data can be obtained reliably; from the classification point, an effective classification learning mechanism is constructed, and good classification results can be obtained; and from the metric learning point, effective characteristic weight is provided, and significant degree of characteristics can be reflected.

The invention discloses an image scene classification method and system combined with semi-supervised clustering, and the method comprises the steps of redefining an objective function of semi-supervised Kmeans through employing a labeled sample, and supplementing and defining an objective function of SVM, and obtaining semi-supervised Kmeans clustering and a base learning device based on SVM classification; enabling the two base learners to carry out cooperative training, and forming a selection and iterative training scheme of a pseudo label sample; and finally, according to the confidence coefficient, fusing results of the two learners to obtain a scene image category to which the sample belongs. According to the invention, different types of methods in the image scene classification field are used to construct a base classifier and carry out cooperative training. Meanwhile, a pseudo label sample is introduced to expand a training set, so that the problem of insufficient label samples is effectively solved. Furthermore, clustering is carried out on the label-free samples to obtain the distribution characteristics of the label-free samples, and the concept drift problem is solved. Finally, the labeling cost of the scene image is reduced, concept drift is solved, and the image scene classification accuracy is improved.

The invention relates to the technical field of high-resolution SAR images, in particular to a ship target detection method on the basis of superpixel and a random forest. Superpixel segmentation andrandom forest non-supervision clustering are blended, the image is partitioned and then clustered, a rare ship target is accurately detected, and meanwhile contour shape information of the ship is better reserved. Compared with a traditional CFAR chip target detection method, time-consuming and labor-consuming operations such as clutter modeling, parameter estimation and window sliding are avoided.

A system, method and computer program product provides a solution to a class of categorization problems using a semi-supervised clustering approach, the method employing performing a Soft Seeded k-means algorithm, which makes effective use of the side information provided by seeds with a wide range of confidence levels, even when they do not provide complete coverage of the pre-defined categories. The semi-supervised clustering is achieved through the introductions of a seed re-assignment penalty measure and model selection measure.

The invention relates to a license plate Chinese character recognition method. The license plate Chinese character recognition method includes steps of manually constructing a standard training sample gray image; preprocessing the training sample gray image to generate a preprocessed gray image; performing band-pass filtering for the preprocessed gray image by the aid of P Gaussian band-pass filters with interconnected pass bands to obtain P filtered gray images; performing dimensionality reduction for the P filtered gray images by a linear or nonlinear process to form images with M-dimensional training sample characteristic vectors; and performing license plate Chinese character recognition for the images with the M-dimensional training sample characteristic vectors by the aid of an RBF (radial basis function) neural network after the images are subjected to dimensionality reduction, and outputting Chinese characters of license plates. The method has the advantages that the RBF neural network is adopted, network parameters are solved by an unsupervised clustering and least square process for solving linear equations, repeated iteration is avoided in a solving procedure, the time efficiency is high, algorithm convergence is good, and the method is high in generalization ability.

The invention discloses a short text classification method based on multiple weak supervision integration, and the method comprises the steps: obtaining an original data set and a knowledge base, andcarrying out the data preprocessing; carrying out knowledge extraction on the preprocessed data; representing the extracted knowledge as an annotation function, and using the annotation function for data annotation; carrying out label integration through a conditional independent model; training a classification model based on a full-connection neural network; evaluating and optimizing the classification model to obtain an optimal model; and performing short text classification by utilizing the optimal model. According to the short text classification method based on multiple weak supervisionintegration, explicit knowledge and implicit knowledge are completely expressed in a mode of combining keyword matching, regular expression and remote supervision clustering; by means of probability labels generated by a label integration mechanism, automatic labeling of label-free data is achieved, the problem of data sparsity of short texts is relieved, and the problem of unbalanced classification of the short texts is effectively solved.

The invention relates to an image segmentation method adopting semi-supervised RFLICM (Robust Fuzzy Local Information C-Means) clustering on the basis of a seed set, which is characterized by at least comprising the following steps of: S101, starting the image segmentation method on the basis of semi-supervised RFLICM clustering of the seed set; S102, leading in an image to be segmented and marking the image as A; S103, carrying out noise adding processing; S104, carrying out clustering on the image added with noise by using a semi-supervised RFLICM clustering method on the basis of the seed set, wherein a clustering result is a final segmentation result of the image; and S105, ending the image segmentation method on the basis of semi-supervised RFLICM clustering of the seed set. The method not only shows the advantages of semi-supervised clustering in the clustering process, but also utilizes an RFLICM algorithm to add local space information and gray information, and thus, the algorithm can utilize more local texture information. Therefore, the image segmentation method has good robustness for noise and a profile and can well kep details of the image, so that accuracy is greatly improved.

The invention discloses a CPM-CGCS2000 building method for a CGCS2000 plate model. The method comprises the following steps: S1, picking a global core framework point as the initial standard, picking a CGCS2000 framework point according to an ITRF (International Terrestrial Reference Frame) framework point picking principle and a 7 parameter supervised cluster and station homogenization picking principle, and processing a CGCS2000 framework net observation data through a calculation method to obtain an accurate station coordinate under ITRF and framework point long-period trend term information; S2, determining a primary plate boundary and a secondary plate boundary of CPM-CGCS2000 of geologic structure motion of China; S3, building a primary mass movement model and a secondary mass movement model of CPM-CGCS2000; and S4, assessing and checking the plate models. The method can realize high-precision maintenance of the national CGCS2000 framework point and can realize centimeter-sized reduction from GNSS actual measurement point to CGCS2000.

The invention provides a coal slime flotation reagent adding amount evaluation method based on semi-supervised clustering, belongs to the field of coal slime flotation, and aims at solving the problems that an existing method has subjective randomness and lagging performance, and fluctuation of flotation working conditions is large, the resource recovery rate is low and reagent consumption is large easily. The method comprises the steps of obtaining a first number of known sample images, and determining a foam type of each known sample image; obtaining a second number of unknown sample images; extracting foam feature vectors of each known sample image and each unknown sample image to obtain a third number of mixed foam feature vectors; marking the foam feature vector of each known sample image according to the foam type of each known sample image; clustering the third number of mixed foam feature vectors by using an EM algorithm of GMM to obtain a clustering result; determining a foam type of each unknown sample image according to a clustering result; and determining whether the agent adding amount is appropriate when each unknown sample image is shot according to the foam type.

The invention provides an image segmentation method converting an unsupervised cluster into a self-supervised classification. The method comprises the flowing sequentially executed steps that firstly, specific samples are extracted from an image and used as training samples through a fuzzy C-means method according to the characteristics of the samples, and an unsupervised clustering problem is converted into a self-supervised classification problem; secondly, subspace partition is carried out on the samples on the basis that the unsupervised clustering problem is converted into the self-supervised classification problem, and therefore an independent subspace is generated for each class of the self-supervised classification; thirdly, category classification is carried out on other samples except the specific samples in an iteration process with the help of KL conversion of multiple subspaces on the basis of the partitioned subspaces.

The invention relates to a method for diagnosing the operation state of a transformer area. The method comprises the following steps: S100, collecting marketing data and distribution network data fromeach service system of each transformer area; S200, preprocessing the marketing data and the distribution network data; S300, unsupervised clustering is carried out on the preprocessed marketing dataand distribution network data, and the type of each transformer area is determined; Step S400, specifying different index weights by aiming at different types of districts in combination with expertexperience, and performing comprehensive scoring on the operation of each district by adopting a classical quantitative scoring model in which an analytic hierarchy process, an entropy weight method and TOPSIS are combined; and S500, determining the operation state of each transformer area according to the operation comprehensive score of each transformer area. In addition, the invention also discloses a system for diagnosing the operation state of the transformer area.

The invention provides an unsupervised clustering anomaly detection method, relates to a spacecraft anomaly detection method, and can solve the problems that an accurate physical analysis model of spacecraft operation at present depends too much on priori knowledge of a spacecraft system, a model is difficult to establish in practical application, knowledge is difficult to obtain and the like. According to the specific technical scheme, a large amount of accumulated spacecraft normal state data are utilized, time scale alignment and equal-interval sampling are conducted on sample data, the sample data are generated, and a spacecraft normal state data model is established through unsupervised clustering analysis based on the thought of inductive learning. The minimum distance of the sampledata is calculated by utilizing a clustering result, a minimum distance set of the sample data is counted and analyzed, and a threshold value of telemetry data anomaly detection is established by utilizing Gaussian distribution. On the above basis, abnormal data detection is realized by judging the deviation degree between the real-time observation data of the spacecraft and the normal state datamodel. The method is used for processing and analyzing spacecraft telemetry data.

The invention discloses an unsupervised clustering method used for large data volume spectral remote sensing image classification, comprising the following steps: dividing the original data into a plurality of data blocks, and obtaining a cluster center of each data subblock by virtue of a peak density searching method; dividing each cluster center into a plurality of data blocks again, and clustering again by virtue of the peak density searching method, so that number of the cluster centers is reduced; and repeating a partitioning-clustering process until similarity of any two cluster centerscan be represented by using one two-dimensional matrix, and then obtaining a final classification result. The unsupervised clustering method disclosed by the invention has the advantages that applicability is good, so that the method not only can be used for hyperspectral remote sensing image classification with more spectrum bands but also can be used for hyperspectral remote sensing image classification with fewer spectrum bands after multispectral remote sensing image or spectrum band selection; and operation efficiency is relatively high, blocked processing reduces computation redundancyof a similarity matrix, and clustering processing of all the data blocks is mutually independent, so that parallel processing can be adopted, and classification rate is increased.

The invention discloses a user clustering method and device. The method comprises the steps of: distributing a user to a closest user group; according to goods browsed by the user and a preset objective function, and calculating the value of an objective function of the user group, wherein the objective function is used for indicating a similar degree between pointed goods among user groups; when the value of the objective function is smaller than or equal to a preset threshold, using all user groups as a clustering result; and when the value of the objective function is larger than the preset threshold, adjusting the weight of each identity attribute characteristic vector of the user according to a decreasing direction of the objective function, and re-distributing the user to the closest user group until the value of the objective function is smaller than or equal to the preset threshold. According to the technical scheme, unsupervised clustering is turned into supervised clustering, and the clustering result is better.

The invention discloses an open source community developer recommendation method based on deep learning and unsupervised clustering. A deep learning neural network is combined with unsupervised clustering; the method is used for developer recommendation in an open source community. The method mainly comprises three steps, firstly, according to general feature information of developers, clusteringthe developers through unsupervised clustering; obtaining categories and ratios of different developers participating in each project; and then performing developer category prediction by using the project information and the developer category information based on the deep neural network, and finally performing training by using the deep neural network to obtain feature vectors corresponding to developers, thereby performing similarity comparison with different categories of developers to recommend corresponding developers. Good recommendation precision and efficiency can be obtained in a large-scale open source software community, the defects of existing research in the aspect of open source software community research can be overcome, and a new open source software developer recommendation method is provided for guaranteeing open source software development quality from a new perspective.

The invention discloses a multi-speaker clustering system and method based on an attention mechanism, and the system comprises a noise removal module which is used for removing noise in an audio; a voice activity detection module which is used for detecting the starting and ending positions of the sound and separating a voice part from a non-voice part; a deep feature vector generation network based on the self-attention mechanism is used for extracting deep feature vectors of the audio clips; and a full-supervised clustering network based on the bidirectional long-short-term memory network Bi-LSTM and the self-attention mechanism is used for clustering the deep feature vectors and outputting a clustering result. According to the multi-speaker clustering method based on the attention mechanism, the influence of noise on the clustering result is removed, and the feature vector generation module based on the self-attention mechanism can learn the global structure features of the audio and generate the feature vectors with discrimination features. The full-supervised clustering network based on Bi-LSTM and the self-attention mechanism can better learn the time sequence and discriminate the features, and the clustering effect is better.

The invention provides a particle swarm algorithm based on an artificial intelligence semi-supervised clustering target, and the algorithm comprises the steps: S1, inputting a data set, and randomly selecting K elements as a clustering center; S2, updating the clustering center, calculating the self-adaptive quantity of the current K value, comparing the self-adaptive quantity with the previous self-adaptive quantity, and retaining the K value with higher self-adaptive quantity; S3, repeatedly executing the step S1S2 until the optimal K clustering centers are obtained; S4, encoding and initializing the particles according to the optimal K clustering centers to obtain individual optimal and global optimal solutions; S5, performing dynamic clustering on the particles, obtaining new positionsof the particles and judging whether updating is needed or not; S6, performing immune disturbance and chaotic disturbance processing on the particles; S7, calculating an individual optimal solution and a global optimal solution of the current particle, comparing with the last time, and judging whether to update the individual optimal solution and the global optimal solution or not; S8, repeatingthe step S5 and the step S7, and if the current number of iterations reaches a preset value, exiting the algorithm.

Identifying insect species integrates image processing, feature selection, unsupervised clustering, and a support vector machine (SVM) learning algorithm for classification. Results with a total of 101 mosquito specimens spread across nine different vector carrying species demonstrate high accuracy in species identification. When implemented as a smart-phone application, the latency and energy consumption were minimal. The currently manual process of species identification and recording can be sped up, while also minimizing the ensuing cognitive workload of personnel. Citizens at large can use the system in their own homes for self-awareness and share insect identification data with public health agencies.

The invention relates to the technical field of intelligent power grids, in particular to a power distribution network fault diagnosis method based on deep feature clustering and LSTM. The method comprises the following steps: extracting high-level features of time series data by utilizing a feature extractor built by a convolutional neural network, and performing semi-supervised clustering on allthe features extracted from the data, so as to obtain corresponding labels for label-free samples. Therefore, the fault type of the label-free sample can be determined and utilized. After samples ofdifferent types of faults are subjected to an oversampling algorithm, a classifier built by a recurrent neural network is used for classification and recognition, and then fault diagnosis is achieved.According to the invention, the label-free data can be utilized through the feature extractor and the semi-supervised clustering built through the convolutional network, the time sequence signals ofvarious monitoring quantities in the sample are successfully utilized through combination with the recurrent neural network, loss of time sequence information contained in original data is avoided, and the accuracy of fault diagnosis is effectively improved.

The invention discloses an intelligent classification method for distinguishing electroencephalogram blink artifacts and frontal electrode epilepsy discharge. The method comprises the following steps: firstly, carrying out filtering processing and signal cutting on electroencephalogram EEG signals; then performing smooth nonlinear energy operator (SNEO) signal conversion and variational modal extraction (VME) signal conversion on the processed EEG signal to obtain an SNEO data set and a VME data set; performing 20-dimensional feature extraction on signals corresponding to the three data sets; then performing binary unsupervised clustering through a K-means algorithm, and constructing an unsupervised classification model; and finally, realizing the classification of blink artifacts and epilepsy discharge signals through the established unsupervised classification model. According to the method, the difficulty of low blink artifact detection precision under the background containing epilepsy discharge signals is overcome, the problem that an existing model neglects epilepsy discharge can be solved, and accurate and automatic classification of the blink artifacts and the frontal electrode epilepsy discharge can be realized.

The invention discloses a method for judging affiliation of an Internet website through a clustering algorithm, belongs to the technical field of network space security, and aims to solve the problemthat a traditional website record information judgment method and a webpage information judgment method cannot judge the affiliation of the website or cause an affiliation judgment error. The method comprises the following steps: a, inputting a website set of a to-be-judged affiliated unit, wherein basic data is a website URL; b, extracting the basic information of the website; c, quantifying allthe information extracted in the step 2; d, mapping the various characteristic values to a [0, 1] interval under the same dimension; normalizing the feature vector FN < websize >; and e, clustering the data set by using an unsupervised clustering algorithm DBSCAN. According to the method for judging the Internet website affiliation through the clustering algorithm, the clustering analysis algorithm is used, automatic judgment of the website affiliation is achieved, and the affiliation judgment accuracy is effectively improved.