198 results about "Feature evaluation" patented technology

There is provided a harmonic structure acoustic signal detection device not depending on the level fluctuation of the input signal, having an excellent real time property and noise resistance. The device includes: an FFT unit (200) which performs FFT on an input signal and calculates a power spectrum component for each frame; a harmonic structure extraction unit (201) which leaves only a harmonic structure from the power spectrum component; a voiced feature evaluation unit (210) which evaluates correlation between the frames of harmonic structures extracted by the harmonic structure extraction unit (201), thereby evaluates whether or not the segment is a vowel segment, and extracts the voiced segment; and a speech segment determination unit (205) which determines a speech segment according to the continuity and durability of the output of the voiced feature evaluation unit (210).

Disclosed is a data analysis and cybersecurity method, which forms a time-based series of behavioral features, and analyzes the series of behavioral features for attack detection, new features derivation, and / or features evaluation. Analyzing the time based series of behavioral features may comprise using a Feed-Forward Neural Networks (FFNN) method, a Convolutional Neural Networks (CNN) method, a Recurrent Neural Networks (RNN) method, a Long Short-Term Memories (LSTMs) method, a principal Component Analysis (PCA) method, a Random Forest pipeline method, and / or an autoencoder method. In one embodiment, the behavioral features of the time-based series of behavioral features comprise human engineered features, and / or machined learned features, wherein the method may be used to learn new features from historic features.

Some embodiments of the present invention provide a system that measures the quality of a software system. During operation, the system performs a series of stress tests on the software system and determines a set of failure rates for the software system from the stress tests. Next, the system obtains a failure distribution from the failure rates. Finally, the system assesses the quality of the software system based on characteristics of the failure distribution.

This disclosure is directed to a system and method for preventing malware, spyware and other undesirable applications from affecting mobile communication devices (e.g., smartphones, netbooks, and tablets). A mobile communication device uses a server to assist in identifying and removing undesirable applications. When scanning an application, a device transmits information about the application to a server for analysis. The server receives the information, produces a characterization assessment and can also provide a characterization re-assessment for the application, or data object, and transmits the assessment to the device. By performing analysis on a server, the invention allows a device to reduce the battery and performance cost of protecting against undesirable applications. The servers transmits notifications to devices that have installed applications that are discovered to be undesirable. The server receives data about applications from many devices, using the combined data to minimize false positives and provide comprehensive protection against known and unknown threats. The server can accumulate this data and then perform a characterization re-assessment of a data object it has previously assessed.

The invention relates to a bearing fault mixing intelligent diagnostic method based on second generation wavelet transform and a BP neural network. The bearing fault diagnostic method based on the second generation wavelet transform and the BP neural network includes steps: firstly, using the second generation wavelet transform to resolve a bearing original vibration signal measured by a sensor; secondly, extracting time domain statistical features and frequency domain statistical features from the resolved signal so as to form a combined feature set, and then performing feature evaluation on the extracted feature set so as to obtain a sensitive feature set; using the sensitive feature set as input of the BP neural network for network training, and building a fault diagnostic model based on the BP neural network so as to achieve classification and diagnosis of faults. The bearing fault diagnostic method based on the second generation wavelet transform and the BP neural network and the fault diagnostic model based on the BP neural network are used in the classification and the diagnosis of the bearing faults, and results indicate that the bearing fault diagnostic method based on the second generation wavelet transform and the BP neural network is high in classification and diagnosis accuracy, high in speed and high in efficiency, effectively improves bearing fault diagnostic effects, and is conveniently used in engineering practice.

The invention discloses a POI (point of interest) Chinese text categorizing method based on a local random word density model. The method includes: judging whether the subject of a text is related to POI or not, and using a modified word concentration, dispersion and frequency method to screen out feature words to build a feature space; performing local area dividing according to the similarity degree of the text and POI category, converting the text into feature vectors in each local area through feature mapping matrixes, and using SVM to perform POI text categorization. The method is good in execution efficiency and categorizing coverage and accuracy. The method is planned to be combined with large knowledge libraries of Hownet so as to capture semantic concepts of low-frequency words and unseen words and further increase the recognition capability of POI text differences, and the problem that good categorization performance cannot be obtained by existing conventional feature evaluation functions and text dimensionality reduction methods is solved.

A method for performing a market analysis may include collecting survey data, which may include evaluation data for desired outcomes, the desired outcomes consisting of attributes of one or more products or services. The desired outcome evaluation data may include importance and satisfaction data. Further, the survey data may include evaluation data for potential features. The potential features evaluation data may include desirability data indicative of the affinity survey participants have for potential features. The method may further include processing the collected survey data using a computer processor. The processing may include calculating, based on the importance and satisfaction data, opportunity scores for the individual desired outcomes. The processing may also include correlating potential features with desired outcomes. In addition, the processing may include ranking the potential features based on the collected desirability data and the calculated opportunity score.

The embodiment of the invention provides a multi-party joint feature evaluation method and device for protecting privacy security. The device at least comprises a first device storing a first sample set and a second device storing a second sample set, and the method is applied to the first device. The method comprises the following steps: encrypting an initial ID of each sample in a first sample set, and sending an obtained first encryption ID and a label of the first sample set to the second device; receiving a first encryption ID of the second sample set, an identifier of a sub-box where thesecond sample set is located, and a second encryption ID and a label of the first sample set from the second device; encrypting the first encrypted ID of the second sample set to obtain a second encrypted ID of the second sample set; determining a common sample according to the second encryption ID of the second sample set and the second encryption ID of the first sample set; and calculating theinformation value of features according to a label of the common sample and the identifier of the sub-box so as to select the features for a machine learning model.

The invention provides a feature extraction method for text categorization. The feature extraction method is used for solving the problem that the accuracy rate and the recall rate of text categorization need to be increased further. The feature extraction method is a strategic method. In consideration of the concept of entropy in statistical thermodynamics, entropy is used for describing the degree of disorder of a system and is significantly applied to the fields of cybernetics, probability theory, number theory, astrophysics, bioscience, information theory and the like. According to the feature extraction method, entropy can also be used in text categorization, a feature is regarded as an event, a category set of text is a system, and therefore entropy can be used for measuring the degree of disorder of features and categories and converted into the closeness degree of the relation between the features and the categories. According to the feature extraction method, on the basis of improved mutual information, the concept of entropy is combined, a new feature evaluation function is provided, feature extraction is conducted on the basis of the function, a superior feature subset can be selected for showing the text and building a categorizer, and therefore the accuracy rate and the recall rate of text categorization are increased.

The invention discloses a power consumer electricity consumption anomaly detection method based on machine learning. The method comprises the following specific steps: 1) cleaning power load data; 2)extracting power load characteristics; 3) performing power load characteristic evaluation; 4) screening data with high reliability based on an isolated forest algorithm; 5) obtaining an inter-class balance data set based on a clustering layered nearest neighbor undersampling technology; and 6) constructing a collaborative forest anomaly detection model based on semi-supervised learning. A hierarchical nearest neighbor sampling algorithm based on clustering is provided, sampling is carried out according to the proportion and the distance, the generalization ability is high, and the accuracy ofunbalanced data set classification is improved; a data set is subjected to preprocessing and feature extraction and evaluation, a time sequence of repeated information is eliminated, and the influenceof a linear relationship between features on a result is eliminated.

The invention discloses a device of voice detection and evaluation based on a mobile terminal. The device of the voice detection and evaluation based on the mobile terminal comprises a voice detection and evaluation terminal, wherein the voice detection and evaluation terminal comprises a signal collection and data storage unit, a voice signal processing and evaluation unit and a data output unit. The mobile terminal is adopted by the voice detection and evaluation terminal. The signal collection and data storage unit comprises a voice signal collection module, a voice signal data storage module and a detection and evaluation result storage module. The voice signal processing and evaluation unit comprises a voice signal processing module, a voice signal cycle detection module, a feature parameter extraction module and a voice signal feature evaluation and analysis module. The data output unit comprises a voice signal output module, a voice signal waveform drawing module and a detection journal visiting module. The device and the method of the voice detection and evaluation based on the mobile terminal have the functions of the voice signal collection, process, analysis and evaluation. The device of the voice detection and evaluation based on the mobile terminal has the advantages of being convenient to carry, low in cost, insensitive in gender difference and the like.

A computer automated feature selection method based upon the evaluation of hyper-rectangles and the ability of these rectangles to discriminate between classes. The boundaries of the hyper-rectangles are established upon a binary feature space where each bit indicates the relationship of a real feature value to a boundary within the minimum and maximum values for the feature across all samples. Data reduction combines the binary vector spaces so that the number of samples within a single class is within a range which is computationally feasible. Identification of subclasses identifies maximal subsets of S+ which are exclusive against S-. Feature evaluation determines within a single subclass the contribution of each feature towards the ability to discriminate the subclass from S-. The base algorithm examines each feature, dropping any feature which does not contribute towards discrimination. A pair of statistics are generated for each remaining feature. The statistics represent a measure of how many samples from the class are within the subclass and a measure of how important each feature is to discriminating the subclass from S-. The values for each subclass are then combined to generate a set of values for the class. These class feature metrics are further merged into metrics evaluating the features contribution across the entire set of classes. Feature reduction determines which features contribute the least across the entire set of classes.

Methods for automated uniformity assessment and modification of image non-uniformities using an image measurement device capable of determining image reflectance and / or transmitter as a function of position, such as a spectrophotometer, a calorimeter, and / or a densitometer. One or more of these devices scan an image on a substrate, such as a sheet, thereby generating data representing image characteristics, such as, for example, image non-uniformity. The sheet may contain a reference / test pattern including one column, strip or patch intended to have a uniform density. This data generated by the device after the scan of the substrate is analyzed with signal processing algorithms for image characteristics assessment, including image spatial uniformity and compared to reference image characteristics including image spatial uniformity. An image modification profile may be generated to be applied to a marking system to thereby alter subsequent image data and improve image spatial uniformity of the marking system.

The invention discloses a distributed optical fiber vibration signal feature extraction and identification method, which belongs to the field of optical fiber sensing signal processing, and comprisesthe following steps of: firstly, acquiring a space-time matrix signal of a vibration source, extracting a space column signal, dividing a short-time signal unit, and constructing an optical cable vibration event data set; constructing, training and optimizing an improved mCNN model, and performing feature evaluation on features extracted by the model during optimization until model iteration is optimal; secondly, extracting time structure feature vectors under multiple scales in parallel by utilizing an optimal mCNN model, recombining the time structure feature vectors into a short-time feature sequence according to a time sequence, and constructing a time structure feature sequence set; finally, constructing and training an HMM model, and constructing an offline vibration event HMM modellibrary to serve as a classifier for vibration source recognition. The problems that in the prior art, local structure features and time sequence features of distributed optical fiber vibration signals cannot be extracted at the same time, and the vibration source recognition accuracy and the generalization ability of the model are low are solved.

The present disclosure includes methods of assessing a histologically stained specimen based on a determined color signature of a region of interest of the specimen. Such assessments may be performed for a variety of purposes including but not limited to assessing the quality of the histological stain, as part of identifying one or more biologically relevant features of the image, as part of differentiating one feature of the image from other features of the image, identifying an anomalous area of the stained specimen, classifying cells of the specimen, etc. Also provided are systems configured for performing the disclosed methods and computer readable medium storing instructions for performing steps of the disclosed methods.

The present invention provides a filtering-package combination flow feature selection method based on a support vector machine. The method comprises an initial filtering feature selection method and asecondary package feature selection method embedded with an improved sequence forward search strategy. The initial filtering feature selection method checks contribution of a certain feature on network flow classification and deletes features lower than a set threshold value [Delta] according to the weight of each feature in an original feature set so as to significantly reduce the computation complexity of subsequent feature subset screening; and the secondary package feature selection method embedded with an improved sequence forward search strategy is based on a support vector machine classifier and is embedded with the improved sequence forward search strategy to perform secondary feature selection to select a combination flow feature subset with high distinguishing ability so as to overcome the problems that combination features are deleted by mistake and there is deviation between a feature evaluation result and a final classification algorithm and significantly improve the network flow classification precision. The method is scientific and reasonable, and can be suitable for various flow classification networks.

Quantitative evaluation on registration results is an important content in the field of image registration. Many scholars propose to evaluate the registration results with pixel physical coordinates RMSE (root mean square error) and MSE (mean square error), or pixel gray level CC (correlation coefficient) and NMI (normalized mutual information) and the like, however, those methods are normally used for evaluating registration of single-modal or retrospective multi-modal images, but quantitative evaluation results are difficult to give to real multi-modal image registration due to lack of accurate measurement criteria. Through research on image matching curves, the invention provides a novel registration evaluating method, namely a matching curve feature evaluating method. Peaks, peak deviations and peak values of matching curves and RMSEs among the peak values are taken as quantitative evaluation indexes, and quantitative evaluation results are given on the basis of the peak deviations and the peak values. By the method, registration performance is visually described from features of smoothness, sharpness and the like of the curves, registration effect can be evaluated quantitatively via feature indexes of the curves, and given evaluation results for sub-pixel registration are accurate.

The present invention discloses a speech emotion recognition method through fusion of feature assessment and a multi-layer perceptron. The method comprises the steps: S1, extracting multi-dimensional emotion feature parameters of a training speech set corresponding to various emotion states, and obtaining an original feature set; S2, performing rating ordering of various emotion feature parameters in an original feature set, and obtaining a feature set after ordering; S3, obtaining a plurality of feature subsets with different quantities from the feature set after ordering, using a multi-layer perceptron to perform classification of each feature subset, and selecting an optimal feature subset according to a classification result; and S4, using the multi-layer perceptron to train an emotion classification model for the optimal feature subset, and performing emotion recognition of the speech to be recognized through the classification model obtained through training. The realization method is simple, the speech emotion recognition method through fusion of feature assessment and the multi-layer perceptron can fuse the feature assessment and multi-layer perceptron to realize emotion recognition, and the emotion recognition precision and the efficiency are high.

The invention relates to a sparse reinforcement type low-rank constraint face image clustering method, which aims to effectively cluster shielded, illuminated and expression-changed face images and obtain a relatively high recognition rate, and comprises the following steps of: (1) carrying out weighted feature evaluation of sparse representation on the images to ensure that reconstruction representation is implemented through effective features; (2) using a regularized non-convex penalty function to constrain the representation coefficient matrix, and giving a parameter norm to keep the convexity of the overall objective function; (3) establishing an FERP clustering model by taking the image matrix X as input and taking the feature weight vector p and the reconstruction representation coefficient matrix Z as variables to be optimized; (4) solving the problem of Z and p double-variable optimization by using an augmented Lagrangian function and an alternating direction multiplier method; (5) calculating an incidence matrix S by using a formula (|Z<*>|+| Z<*T>|) / 2; and (6) segmenting the incidence matrix S by using an Ncut segmentation algorithm to obtain a clustering image. The method has the advantages of high operation efficiency, high data adaptability, high accuracy and high expansibility, and is suitable for face image clustering and recognition.

The invention discloses a case reasoning classifier case retrieval method. The method comprises the steps of S1, case library protocol feature selection; S2, clustering decision tree training on a C4.5 algorithm and a case library; and S3, a weighted voting KNN method. According to the method, improved research is carried out on a case retrieval method of a case reasoning classifier: a method of integrating and using a inductive index strategy in neighbor case retrieval is provided, and a feature evaluation method of an information gain ratio is adopted to select a feature training case library clustering decision tree with a relatively remarkable information gain ratio; during case retrieval, a cluster corresponding to a target case is retrieved according to the clustering decision tree, and then neighbor case retrieval is performed in the case cluster. Experimental data of the open source data set show that compared with a traditional CBR model, the improved model can fully utilize the classification capability of case characteristics to carry out case retrieval, the calculation time complexity of the CBR reasoning classifier is effectively reduced, and the reasoning accuracy is improved.

A system includes a learning data input unit for receiving initial and long term characteristic learning data of a battery to be a learning object; a measurement data input unit for receiving initial characteristic measurement data of a battery to be an object for long term characteristic estimation; an artificial neural network operation unit for converting the learning data into first and second data structures, allowing an artificial neural network to learn the learning data based on each data structure, converting the measurement data into first and second data structures, and individually applying the learned artificial neural network corresponding to each data structure to calculate and output long term characteristic estimation data based on each data structure; and a long term characteristic evaluation unit for calculating an error of the estimation data of each data structure and determining reliability of the estimation data depending on error.

The invention provides a click-through rate estimation method, a click-through rate estimation device, a computing apparatus, and a storage medium. The method includes the following steps that: a feature evaluation network with user features and item features adopted as nodes is established; a click-through rate estimation model is established based on the feature evaluation network; and the click-through rate of users for an item can be estimated based on the click-through rate estimation model. Thus, the click-through rate estimation model can reflect a relationship between the user featuresand the item features in the form of a network, so that the click-through rate of an advertisement can be easily estimated.

The invention discloses a face photo quality evaluation method, system and device and a storage medium, and the method comprises the steps: evaluating image quality of a face image to acquire a plurality of preliminary evaluation coefficients; according to the weight of each preliminary evaluation coefficient, weighting each corresponding preliminary evaluation coefficient to obtain a weighted evaluation value of the face image; and performing feature extraction on the face image to obtain a feature evaluation value, and performing final quality evaluation on the image quality of the face image according to the feature evaluation value. The problem that a face recognition system cannot recognize a face image or is wrong in recognition due to the fact that the quality of the face image becomes poor in various actual complex environments is solved, and the recognition rate of face recognition is greatly increased.

The invention provides a feature selection algorithm based on Relief and mutual information, and belongs to the field of computer algorithms. The algorithm comprises the following steps of (1) settingan optimal feature subset as an empty set, and setting a weight of the optimal feature subset; (2) selecting features not belonging to the optimal feature subset in all features in one piece of data,putting the features into a candidate optimal feature subset, and calculating a weight of the current candidate optimal feature subset through a composite feature evaluation criterion; (3) evaluatingand replacing the weight of the candidate optimal feature subset at the moment; (4) removing to-be-selected features which do not meet the requirements; and (5) if the to-be-selected features still exist, returning to the step (2) for continuous calculation, otherwise, ending the algorithm. A method provided by the invention is improved for the problem that a Relief feature selection algorithm only can handle the binary classification problem and cannot process redundant features, and an improved Relief weight-based feature selection algorithm is provided, so that the feature selection algorithm has higher calculation accuracy while having high calculation efficiency.

The invention provides a rolling bearing intelligent fault diagnosis method based on multi-classifier integration and parallel feature learning, and aims to improve the classification precision of a model, and the method comprises the following implementation steps: obtaining a training sample set and a test sample set; establishing a plurality of stacked auto-encoder models, carrying out paralleltraining on the stacked auto-encoder models by taking the training sample set as an input, and extracting a plurality of characteristics of the training sample set; performing feature evaluation on the extracted features based on a softmax model, and screening the features according to a corresponding threshold value and an evaluation index value to form a feature subset; establishing a pluralityof classifiers based on a softmax model according to the feature subset, obtaining the classification precision of each classifier by taking the feature subset as input, reselecting a plurality of classifiers according to a threshold value to construct an integrated multi-classifier model, obtaining an integrated multi-classifier model prediction label through a majority voting method, and mapping the prediction label and the fault type of the rolling bearing to realize the intelligent fault diagnosis of the rolling bearing.

Disclosed is an improved feature evaluation function based Bayesian spam filtering method. The method includes the steps of 1), preprocessing a training mail set into a mail head part and a text part; 2), respectively deleting prepositions, pronouns, adverbs, auxiliary words, conjunctions and words with the work frequency lower than the given threshold P in two feature sets T1 and T2; 3), respectively calculating a mutual information value MI (tk)' in the feature sets T1 and T2 by adopting the improved feature evaluation function; 4), in the training set, sorting the MI (tk)' according to the order from big to small, and selecting feature items corresponding to first n values as representation of the training set; 5), performing spam filtering on to-be-tested samples by adopting a Bayes classifier at the sorting phase. With the method, mails can be classified highly accurately, and spasm can be filtered out.

The invention provides a recognition technology-based electronic transaction authenticating system comprising an electronic transaction authenticating system and an object recognition device connected with the electronic transaction authenticating system. The object recognition device is used for recognizing and tracking an object based on visual features. The object recognition device comprises a suspected object obtaining module, a color information processing module, a contour information processing module and a feature evaluation module that are all orderly connected; the color information processing module is used for subjecting an original frame image to RGB color space-to-HSV color space conversion operation and building a tone and color model of the suspected object in the HSV color space; the contour information processing module is used for dividing an actual contour of the original frame image into two types of zones: feature zones and non-feature zones, neighboring zones of the same type are combined, and filters of different parameters are chosen to subject the combined feature zones and combined non-feature zones to smoothing operation. The recognition technology-based electronic transaction authenticating system is advantaged by high recognizing precision and high recognizing speed.

The invention discloses a road network state determination method and device, equipment and a storage medium. The method comprises the steps of determining an association relationship between an intersection and a road section in a road network; constructing a graph neural network model of the road network based on the incidence relation; training the vertexes representing the road sections in the graph neural network model, and obtaining feature vectors of the vertexes; under the constraint of the attribute information of the road section, calculating the similarity between the vertexes based on the feature vectors; and determining intersections and road sections with correlated traffic modes and traffic states in the road network according to the similarity. According to the method, the feature vectors of the road sections are extracted from the graph neural network model, the attribute information of the road sections is combined with the feature vectors, the similarity between the road sections in the road network is calculated, and the state of the road network is evaluated according to the similarity, so that the state of the road network is evaluated from two dimensions of space and time; and the attribute information of the road section is integrated into the feature evaluation of the whole road network, so that the road network state can be analyzed more comprehensively.