41 results about "Class discrimination" patented technology

The invention, which belongs to the field of voice signal processing and machine learning, relates to a deep-belief-network-characteristic-vector-based channel-robust voiceprint recognition system comprising a voice acquisition and preprocessing module, an original spectral characteristic extraction module, a deep belief network training module, a speaker voiceprint characteristic vector extraction module, a speaker acoustic model generation module and a speaker identification module. On the basis of voice data from different channels and corresponding speaker identity numbers, a deep belief network is trained in a manner of supervision; and a discrimination ratio is provided to select a deep belief network hidden layer output having an optimal class discrimination property, thereby constructing a speaker voiceprint characteristic vector having channel robustness. Compared with the traditional i-vector-based speaker confirmation system, the provided system has higher voiceprint recognition accuracy on the condition of channel mismatching.

The invention discloses a brain image segmentation method and system based on a saliency learning convolution nerve network; the brain image segmentation method comprises the following steps: firstly proposing a saliency learning method to obtain a MR image saliency map; carrying out saliency enhanced transformation according to the saliency map, thus obtaining a saliency enhanced image; splitting the saliency enhanced image into a plurality of image blocks, and training a convolution nerve network so as to serve as the final segmentation model. The saliency learning model can form the saliency map, and said class information is obtained according to a target space position, and has no relation with image gray scale information; the saliency information can obviously enhance the target saliency, thus improving the target class and background class discrimination, and providing certain robustness for gray scale inhomogeneity. The convolution nerve network trained by the saliency enhanced images can be employed to learn the saliency enhanced image discrimination information, thus more effectively solving the gray scale inhomogeneity problems in the brain MR image.

An inspection apparatus includes a discrimination function determination unit which determines whether or not a discrimination function forms an area including a discrimination sample. The discrimination function is used in non-parametric one-class discrimination. The discrimination sample is discriminated into a class as a single area in an input space where learning samples are plotted.

The invention discloses a cross-domain text sentiment classification method based on domain countermeasure self-adaption. The method comprises the following steps: inputting a word vector matrix, a category label and a domain label of a source domain sample and a target domain sample; Utilizing a feature extraction module based on a convolutional neural network to extract low-level features of thesample; constructing a constraint based on distribution consistency of a source domain and a target domain in a main task module, mapping a low-layer sample to a regeneration kernel Hilbert space, and learning a high-layer feature with transferability; inputting the high-level features of the source domain into a class classifier, and ensuring that the classifier has class discrimination on samples on the basis of reducing domain difference; a domain invariance constraint based on adversarial learning is constructed in an auxiliary task module, and low-level features are input into a domain classifier with adversarial properties, so that the classifier cannot judge the domain to which a sample belongs as much as possible, high-level features with domain invariance are extracted, and the migration problem of a source domain classifier to a target domain is effectively solved.

The invention relates to an action recognition method based on a neural network and an action recognition device based on the neural network. The method comprises the steps that a video to be recognized is inputted to a trained first three-dimensional neural network model to be processed so that the action extraction result of the video to be recognized is obtained; the action instance detection result of the video to be recognized is determined according to the action extraction result of the video to be recognized; the video to be recognized is inputted to a trained second three-dimensionalneural network model to be processed so that the action class discrimination result of the video to be recognized is obtained; and the action class of the video to be recognized is determined according to the action instance detection result of the video to be recognized and the action class discrimination result of the video to be recognized. Different recognition results obtained by using two three-dimensional neural network models are combined so that the recognition efficiency of the three-dimensional neural network models can be enhanced and the computational burden of the single three-dimensional neural network model can be reduced.

Classifiers, which are combinations of a plurality of weak classifiers, for discriminating objects included in detection target images by employing features extracted from the detection target images to perform multi class discrimination including a plurality of classes regarding the objects are generated. When the classifiers are generated, branching positions and branching structures of the weak classifiers of the plurality of classes are determined, according to the learning results of the weak classifiers in each of the plurality of classes.

A hyperspectral image migration classification based on a depth joint distribution adaptive network includes such steps as inputting hyperspectral images in source domain and target domain, normalizing features and unifying dimensions; combining features of hyperspectral images in source domain and target domain; The edge probability distribution adaptation network is constructed to adapt the edgeprobability distribution of hyperspectral images in source domain and target domain. According to the principle of one-to-many classification, the training samples of hyperspectral images in source domain and target domain are selected. A conditional probability distribution adaptation network is constructed to adapt the conditional probability distribution of the hyperspectral images in the source domain and the target domain. One-to-many classification of hyperspectral images in target domain is performed. A depth-based joint distribution adaptation network is proposed, which realizes feature adaptation of a source domain and a target domain hyperspectral image, and reduces that joint probability distribution difference between the source domain and the target domain. At the same time,one-to-many classification model is used to improve the intra-class and inter-class discrimination, and then the accuracy of hyperspectral image migration classification is improved.

The invention relates to a feature selection method and application based on feature identification degree and independence. The method comprises following steps: calculating the importance degree of each feature by measuring inter-class distinguished ability with feature identification degree and measuring correlational relationship between features with feature independence and sequencing in a descending order; and selecting top k features with the importance higher than those of others to form a feature subset with high class-discrimination performance. Differently-expressed gene subsets selected in application of oncogene expression profile data obtain fine time and class discrimination performance. The feature selection method and application based on feature identification degree and independence have following beneficial effects: easy calculations can be made; time complexity is reduced; selection efficiency runs high; and a good reference is provided for clinical diagnoses and judgments of tumors and other diseases.

The invention particularly relates to an IG TF-IDF text feature vector generation and text classification method, and belongs to the field of text mining and machine learning. The method comprises thefollowing steps: 1) generating a text feature vector; 2) train that classifier; 3) evaluate that classification performance; 4) classify that target text set; The weight calculated by the invention can more truly reflect the importance of different terms to the text classification, so that the term with strong class discrimination ability is allocated with larger weight, the weight calculation ismore reasonable, and the accuracy of the text classification is improved. Moreover, the calculated term weights do not need to know the specific categories, thus overcoming the shortcomings of supervised methods such as TFADF in multi-category text classification.

The invention provides an inspection apparatus and an inspection method. The object of the invention is to determine criterion function and in order to make the criterion function used for nonparametric 1-class discrimination form a single on-spec product region. In a discriminant function deciding section 20 in an inspection device, nondefective region discriminating section 26 determines, in an input space in which the discriminant function used for the nonparametric 1-class discrimination plots a sample, whether a region containing the sample discriminated into the class is formed as a single region. Then, when the region containing the sample discriminated into the class is not single, a parameter setting section 24 sets region parameters for defining the size of the region of the basis function of the base of a density function so that the discriminant function constitutes a single region in an input space in which the sample is plotted. Wherein, the region parameters prescribes criterion function and the region size of basis function as basic of criterion function.

The invention discloses an early warning method for giving precedence to pedestrians under new traffic rules and an early warning system for giving precedence to the pedestrians under the new trafficrules. The method comprises the steps that vehicle driving information, pedestrian information and traffic sign line information are acquired; the pedestrian information is detected by using a pedestrian detection algorithm based on deep learning so as to complete pedestrian position detection and class discrimination; the traffic sign line information is detected by using a traffic sign line detection algorithm, and the traffic sign line is extracted; and whether the pedestrians have the intention of crossing the road is judged according to the result, the relative position of the pedestriansand the vehicle is analyzed if the judgment result is yes, and an early warning signal is emitted to prompt the vehicle driver to stop to wait for the pedestrians to pass when the vehicle meets the stop rules of giving precedence to the pedestrians. The system comprises a sensor, an embedded platform and an early warning device and is used for performing early warning in case of giving precedenceto the pedestrians under the new traffic rules. According to the early warning method for giving precedence to the pedestrians under the new traffic rules and the early warning system for giving precedence to the pedestrians under the new traffic rules, the driver can be assisted to be clear about the rules of giving precedence to the pedestrians, and the safety of the pedestrians can be guaranteed.

A learning sample collector is configured to collect learning samples which have been classified into respective classes through N-classes discrimination (N is a natural number of 3 or more). A classifier is configured to reclassify the learning samples collected by the learning sample collector into classes applied to M-classes discrimination, where M is smaller than N (M is a natural number of 2 or more and is less than N). A learner is configured to learn a discriminator for performing the M-classes discrimination on a basis of the learning samples reclassified by the classifier.

The invention relates to a method for optimizing the oil-paper insulation time domain dielectric response feature quantity of a two-stage transformer. The advantages of a filter type feature selectionstatistical index and random forest bag external data feature quantity importance estimation are combined, and an optimal feature space with the lowest redundancy, the highest class discrimination degree and the highest classification importance can be finally determined through two-stage feature selection. The method is used for realizing evaluation selection of the time domain dielectric spectrum feature quantity; under the condition that a traditional insulation diagnosis method adopts a similar dimensional feature space, the method can be used for obtaining more effective information, theaccuracy of insulation diagnosis is greatly improved, a new idea of selecting the feature space is provided for subsequent transformer insulation evaluation by using the multivariate time domain feature quantity, and the method has an important application value in practical engineering.

The invention relates to a content-based novel recommendation method, belonging to the technical field of recommendation method. Firstly, the novel text is initialized and the corresponding SinHash fingerprint is extracted to establish the dynamic novel database. Then the reference novel is input and the novel to be recommended is determined by the publishing time and the correlation between the novel texts based on the SinHash fingerprint. Finally, the recommended novels are sorted according to their relevance, and the recommended novels with certain items are outputted. Compared with the prior art, the method mainly solves the problems of low recommendation accuracy, weak class discrimination ability, poor efficiency and the like existing in the prior art when recommending novels, and increases the accuracy and flexibility of relying on computers to recommend novels at present.