305 results about "Unsupervised clustering" patented technology

A system and method for performing domain-specific knowledge based metasearches. A metasearch engine is provided for accessing a searching text-based documents using generic search engines while simultaneously being able to access publication based databases and sequence databases as well as in-house proprietary databases and any database capable of being interfaced with a web interface so as to produce search results in text format. A data mining module is also provided for organizing raw data obtained by unsupervised clustering, simple relevance ranking, and categorization, all of which are done independently of one another. The system is capable of storing previous search data for use in query refinement or subsequent searches based upon the stored data. A search results collection browser may be provided for analyzing current browsing patterns of the user for developing weighting factors to be used in ordering the results of future searches.

The present disclosure relates to an early warning and recommendation system for proactive management of a wireless broadband network. Without human intervention, the system processes highly heterogeneous network and non-network data and applies unsupervised machine learning to the data to predict and understand the situations that lead to different network state conditions. More specifically, unsupervised clustering is applied to the data to understand “situations” that can lead to non-normal network state conditions. A deep neural network model of situations is then created to further understand the underlying data relationships between the elements of a situation and network states. The deep neural network model and Reinforcement Learning is used to provide recommendations as to changes in wireless/mobile broadband network configuration parameters that will improve the state of a predicted situation associated with non-normal network conditions. The system displays warnings and corresponding recommendations regarding predicted non-normal network conditions in a user interface for a network operator.

The invention discloses a fast high-resolution SAR (synthetic aperture radar) image ship detection method based on feature fusion and clustering. The fast high-resolution SAR image ship detection method comprises the following steps: on the basis of the back scattering characteristics of each ground object and the prior information of a ship target in an SAR image, positioning a target potential position index map by an Otsu algorithm and range constraint; on the index map, pre-screening to obtain a detection binary segmentation map by a CFAR (constant false alarm rate) algorithm based on a local contrast; carrying out morphological processing to a detection result, and extracting a potential target slice from the SAR image and a detected binary segmentation map according to a processing result; and carrying out K-means clustering to the extracted slice by a designed identification feature to obtain a final identification result. According to the fast high-resolution SAR image ship detection method based on feature fusion and clustering, the data volume of a detection stage is effectively reduced by pre-processing, and point-to-point detection is not needed/the time of point-to-point detection is saved. Meanwhile, a target identification problem under the condition of insufficient training samples at present can be solved by the designed characteristic and a non-supervision clustering method, the target can be effectively positioned, and the size of the target can be estimated.

The invention discloses a voice endpoint detection method based on waveform morphological characteristic clustering. The voice endpoint detection method based on waveform morphological characteristic clustering includes the following steps that first, a pure voice signal is acquired through an original voice signal; second, an envelope signal of the pure voice signal is acquired and divided into a plurality of voice sub-segments; third, according to the waveform morphological characteristics of all the voice sub-segments, the voice sub-segments are clustered and non-voice voice sub-segments are removed; fourth, all the voice sub-segments of reserved parts in the third step are processed, and a voice endpoint is obtained. According to the method, a good result can be obtained fast and accurately with a relative simple non-monitoring clustering method under the condition that a single characteristic is utilized.

The invention discloses a wind power prediction method based on continuous time slice clustering and support vector machine (SVM) modeling. The method comprises the following steps of: (1) performing annual similar day unsupervised clustering according to the wind characteristic; (2) partitioning an entire year into n continuous time slices according to a similar day clustering result obtained in the step (1), and clustering and classifying every time slice according to the frequency of each type of date within every time slice and the wind characteristic in the continuous time slices; and (3) modeling the time slices of the same type in the step (2) by using an SVM for predicting the same time of future years. An annual continuous time slice clustering method is adopted on the basis of day similarity, so that day similarity and time continuity are considered simultaneously, and the similarity of a training sample in a prediction model and the accuracy of wind power prediction are increased greatly. Compared with the conventional method, the wind power prediction method has the advantages that: the relative power prediction error is decreased by 7.2 percent, and the prediction accuracy of the wind power is up to 83.96 percent.

A non-supervision clustering-based distributed cooperative spectrum sensing method for a cognitive self-organizing network relates to the field of cognitive radio in wireless communication technology. Aiming at solving the problems of difficult distributed cooperation of the cognitive self-organizing network and large overhead of whole-network cooperation, the method adopts the latest achievements of a non-supervision clustering theory and a co-recognition theory to achieve fully-distributed, steady and reliable distributed cooperative spectrum sensing under the condition of simplifying the overhead of network sensing; and users with potential optimal detection performance spontaneously gather only via information interaction between neighbors, further the users carry out cooperation spectrum sensing by utilizing an average co-recognition protocol, and a sensing result is broadcast to the whole network users. The method does not require local users to receive apriori information of noise-signal ratio, and does not need any central controllers, thereby greatly lowering sensing overhead and acquiring detection performances similar to optimal soft combination solution.

Embodiments of the present invention describe a collaborative framework for mining of surveillance videos to detect abnormal events, which introduces a two-stage training process to alleviate the high false alarm problem. In the first stage, unsupervised clustering is performed on the segments of the video streams and a set of abnormal events are combined with user feedback to generate a clean training set. In the second stage, the clean training set is used to train a more precise model for the analysis of normal events and the motion detection results from multiple cameras can be cross validated and combined. This description is not intended to be a complete description of, or limit the scope of, the invention. Other features, aspects, and objects of the invention can be obtained from a review of the specification, the figures, and the claims.

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.

A method mines unknown content of a video by first selecting one or more low-level features of the video. For each selected feature, or combination of features, time series data is generated. The time series data is then self-correlated to identify similar segments of the video according to the low-level features. The similar segments are grouped into clusters to discover high-level patterns in the unknown content of video.

The invention discloses a human face beautifying method based on non-supervision optimal beauty features and a depth evaluation model. Apparent features extracted with the CDBN are used for establishing the human face beauty depth evaluation module, the optimal human face beauty features are studied through a non-supervision cluster and a beauty approximation function, and therefore human face sample images are beautified under the guidance of the human face beauty depth evaluation model by utilizing the features and a geometric deformation mode. After different samples to be beautified are beautified by the human face beautifying method, human faces of different beautiful styles can be obtained, so that the aesthetics of the human faces cannot be excessively single, diversity of beauty is embodied, and different requirements for aesthetics of different people are met. Meanwhile, based on the reasonable beauty attraction approximation function, it can be ensured that the human face images cannot be beautified into human faces of other types, and the basic features of the human face images can be reserved.

The disclosed embodiments provide a system that detects an anomaly in a computer system based on log messages. During operation, the system receives log messages generated by the computer system during operation of the computer system. Next, the system maps each received log message to a cluster in a set of clusters of log messages, wherein each cluster is associated with a specific event. The system then forms events for consecutive log messages into sequences of events. Finally, the system performs anomaly detection based on the sequences of events, wherein if an anomaly is detected, the system triggers an alert.

The invention discloses a labeling method and device for web page topics. The method includes the steps that based on titles and main bodies of web pages, topic feature vectors of the web pages are acquired; classification processing is performed on the topic feature vectors through a classifier which is obtained through training in advance; whether types which the topic feature vectors belong to exist is judged; if yes, the web pages are labeled as the types which the topic feature vectors belong to; otherwise, the web pages are labeled as web pages to be labeled; furthermore, clustering processing is performed on the multiple web pages to be labeled; the type of each cluster is obtained through analysis; the web pages to be labeled are labeled as the types of the clusters which the web pages belong to. By the adoption of a supervised classification method and unsupervised clustering method cascading mode, the topics are automatically acquired from the web pages, the web pages are labeled, and the labeling efficiency and accuracy of the web page topics are effectively improved.

The invention provides a method for automatically generating a main character abstract in a television program. According to the method, image characteristics and an unsupervised clustering method in a video are combined, so that main characters in a program video are extracted, an interviewer and an interviewee are marked, and a preview of the main character abstract is generated; and therefore, the user experience is enhanced. The method comprises the following steps of: extracting a key frame at a uniform time interval based on a video paragraph with marks, performing face detection and characteristic extraction, performing linear clustering on face images based on characteristics, obtaining rough character classifications according to time information and space information of the face images, obtaining a precise clustering result through graph-theory-based clustering, filtering the main character classifications of independent paragraphs by an adaptive method, integrating filter results of all the paragraphs, secondarily performing the graph-theory-based clustering, judging the interviewer and the interviewee according to a rule, and marking and generating the main character abstract of each paragraph. By the unsupervised clustering method, the structure is simple, and the method is easy to implement; and the method is relatively high in universality and robustness.

The invention discloses a semantic similarity analysis method based on text clustering. The method comprises the following steps: taking unprocessed text data as input; performing word frequency statistics on texts subjected to data preprocessing, adding word frequency statistics information serving as priori knowledge into text clustering, proposing a posteriori judgment criterion, and performingan unsupervised clustering method on the basis of taking the word frequency statistics as a classifier to improve the accuracy and timeliness of a text clustering result; carrying out synonym ambiguity elimination on the processed text, and carrying out semantic role labeling; and generating a semantic vector fused with the context features, processing the text sequence by adopting two LSTMs withcompletely same structures and parameters, adding the product and variance of the results, amplifying the same points and differences of the texts, and calculating to obtain a final result of similarity analysis. The method can be applied to actual scenes of text similarity analysis in various different fields, and text data of different types can be well processed.

The invention discloses a deep learning clustering method for noise images. The deep learning clustering method comprises the following steps: S1, constructing a deep learning clustering model; S2, adopting an AMsoftmax layer as a clustering device, and generating a clustering result according to the feature vector output by the encoder in the step S1; S3, measuring the similarity between the output of the encoder and the output of the twin network by adopting an L2 norm; S4, adopting KL divergence to measure the distribution difference between the clustering result and the auxiliary target distribution; S5, training a deep learning clustering model; And S6, obtaining a clustering result of the data through the AMsoftmax layer. According to the method, unsupervised clustering can be carried out on image data containing noise, and the problems that most image clustering algorithms do not model noise data and an existing deep clustering algorithm is poor in clustering effect on images with high non-linear characteristics are solved.

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 three gene groups for transferring and typing prognostic numerator to breast cancer, a gene chip, an unsupervised clustering map from a gene expression profile and a method for transferring and typing prognostic numerator to breast cancer patients. The invention is characterized in that: the gene chip is point-sampled and prepared by specific cDNA or oligonucleotide of the gene groups with an lattice form on the surface of the supports, and is used for transferring and typing prognostic numerator to breast cancer; the cDNA or amplified products of primary cancer tissue specimens based on breast cancer patients are fluorescence-marked and then crossbred with the gene chip for transferring and typing prognostic numerator to breast cancer, and then the gene expression profile is obtained, and the unsupervised clustering map for transferring and typing prognostic numerator to patients comes from the gene expression profile; the method makes use of the gene chip and the gene expression profile for transferring and typing prognostic numerator to breast cancer.

The present invention provides an image classification method based on a concise unsupervised convolutional network, and belongs to the image processing and deep learning technology field. The method of the present invention utilizes a classic unsupervised clustering algorithm K-means to cluster the image blocks of a training image set, the obtained each clustering center is a convolution kernel in a network model, and the time-consuming process in the conventional convolutional network of utilizing a stochastic gradient descent algorithm repeatedly to obtain the convolution kernels is abandoned. In addition, the present invention provides a probability pooling method to enhance the robustness of the network to the image deformation. By a concise unsupervised depth convolutional network classification model provided by the present invention, the model training time can be shortened effectively, at the same time, the identification capability of the model to the changeable scene pictures is improved.

The invention discloses an operating system scheduling method and a device based on machine learning. The device comprises the following steps: classifying tasks into one or more task groups based onresource requirement characteristics by using an unsupervised clustering algorithm; the pre-established training model being used to adjust the resource allocation parameters of the task group, so that the hardware resource utilization efficiency and the degree to which the task meets the user expectations are optimized; the resource allocation parameter obtained by the training being used for scheduling and allocating resources for the process of the next cycle of the task group.

Described is a system for classifying objects and scenes in images. The system identifies salient regions of an image based on activation patterns of a convolutional neural network (CNN). Multi-scale features for the salient regions are generated by probing the activation patterns of the CNN at different layers. Using an unsupervised clustering technique, the multi-scale features are clustered to identify key attributes captured by the CNN. The system maps from a histogram of the key attributes onto probabilities for a set of object categories. Using the probabilities, an object or scene in the image is classified as belonging to an object category, and a vehicle component is controlled based on the object category causing the vehicle component to perform an automated action.

A platform for transcriptome deconvolution of gene expression data is provided and may be used in assessing metastatic cancer samples. The deconvolution is performed using an unsupervised clustering technique, such as grade of membership, that allows for samples to be assigned to multiple clusters during a training process. A deconvolution gene expression model is generated as a result and is used for accurate assess of metastases in subsequent samples.

The invention discloses a handwritten character recognition method based on an expanded nonlinear kernel residual network. A novel deep learning method, namely an algorithm based on the expanded nonlinear kernel residual network is provided by utilizing structural characteristics of a deep network, and the handwritten character recognition method based on the expanded nonlinear kernel residual network is proposed by applying the deep learning algorithm to handwritten character recognition. Through the method, the correlation between sample data and expected data can be described deeply, and digital image features can be automatically learned from original data efficiently; next, an appropriate intra-class unsupervised clustering algorithm is introduced into the method, and existing technical defects of the deep learning network in the handwritten character recognition field are overcome. The method is simple and easy to implement, and network training efficiency is improved while handwritten character recognition performance is improved.