36 results about "Feature vector" patented technology

The invention discloses a face recognition method and system. According to the method, the method comprises the following steps of carrying out face detection on a video stream image; after alignment,extracting a face feature vector; calculating the similarity between each frame of image and the face feature vector of the registered face feature library; establishing a tracker for each face; determining a tracker to which the human face belongs through the cross-parallel ratio of the human face position coordinate area and the weighted value of the human face similarity; and when the trackercollects face images with a preset frame number, counting a recognition result of the face tracked by the tracker, and judging a final result of face recognition according to the recognition result, the voting proportion of each group and a similarity average value. The technical scheme has the advantages of high recognition efficiency, accurate face recognition and high feature expression richness.

The invention provides an abnormality detection method and device based on log graph modeling. The method and the device are applied to a nonsocial network. The method particularly includes: buildinga bipartite graph according to a key field of abnormal data annotated in advance in the nonsocial network, wherein left-side nodes of the bipartite graph correspond to multiple user accounts while right-side nodes of the same correspond to parameter combinations during business interface requesting; extracting features from the bipartite graph, and splicing the extracted features to form a featurevector; performing k-means cluster processing on the basis of the abnormal data and the feature vector to acquire optimal cluster number; according to the optimal cluster number, using a Gaussian mixing model to fit dark industry feature probability distribution; when receiving incoming data, calculating dark industry probability of the data according to feature vector of the incoming data and the Gaussian mixing model, and judging whether the data are abnormal or not according to the dark industry probability and the dark industry feature probability distribution. When the data are judged abnormal, access behaviors of users can be intervened timely, so that network attacks by hackers can be avoided.

The invention discloses a real-time gesture recognition method which comprises the steps of (1) decomposing an obtained gesture video into image sequences sorted in a chronological order and preprocessing obtained images and then carrying out hand-region segmentation, (2) extracting a hand shape feature of a hand region in each image and using an SVM support vector machine to identify the hand shape feature as a corresponding gesture value, (3) combining the gesture value of each image and a direction feature of a motion trajectory obtained by an iterating LK pyramid optical flow algorithm asa feature vector of each dynamic gesture image, (4) carrying out loop execution of steps (2) and (3) with a loop end condition that all images of a current video are processed so as to obtain a complete set of feature vector sequences, (5) creating a gesture template library, (6) carrying out optimization DTW match on the obtained feature vector sequences and all templates in a template library, calculating the degree of distortion of the match, wherein recognition is failed if the degree of distortion is larger than a distortion threshold, and a recognition result is outputted if the degree of distortion is smaller than the distortion threshold.

The invention relates to a fraud detection method and device for video data, computer equipment and a storage medium. The method comprises the following steps: acquiring to-be-detected video data; extracting image data of each video frame from the to-be-detected video data, and dividing the image data into a plurality of image data sets according to the time sequence of each video frame, the imagedata sets which comprises image data corresponding to continuous video frames; inputting each image data set into a pre-trained image feature extraction model to obtain an image feature vector; extracting voice data from the to-be-detected video data, and obtaining a voice feature vector of the voice data; performing cascade splicing on the image feature vector and the voice feature vector to obtain a multi-modal feature vector; and inputting the multi-modal feature vector into a pre-trained fraud detection model to obtain a fraud detection result corresponding to the to-be-detected video data output by the fraud detection model. By adopting the method, the characteristic information amount can be increased, the comprehensiveness and diversity of the characteristic information are improved, and the accuracy of video data fraud detection is effectively improved.

A method for segmenting mesh models in digital images includes providing (11) a mesh surface model (V, F, E) of a shape embedded in a 3-dimensional space, providing (12) a ground truth segmentation of the mesh surface model, constructing (13) a feature vector v_i composed of underlying geometric properties of the mesh at each vertex point, using (14) the ground truth segmentation and the feature vectors to determine weights w_ij for edges e_ij where the weights w_ij are the probabilities of a random walker crossing edges e_ij between vertices v_i and v_j in a random walker segmentation initialized with at least two seeds of different labels.

The invention discloses a software defect prediction method and device and electronic equipment, and the method comprises the steps: obtaining a feature vector of to-be-predicted software; and determining a software defect prediction result of the to-be-tested software based on the feature vector and a pre-trained prediction model used for software defect prediction with the prediction model beingobtained based on a gradient lifting algorithm and random forest learning machine training. By adopting the method provided by the invention, the gradient lifting algorithm and the random forest learning machine are used for training to obtain the training model for software defect prediction, so that the software defect prediction result output by the prediction model obtained by training is relatively high in accuracy, and meanwhile, relatively large influence on the calculation complexity is avoided.

The invention belongs to the technical field of digital forensics, and particularly relates to a malicious code fragment intelligent forensics method and system, and the method comprises the steps: constructing a code fragment training set and a code fragment test set for training and testing through extracting the underlying data features of a storage medium; training the set full-connection neural network model by using the data in the code fragment training set, the input being a feature vector, and the output being a normal or malicious prediction result; for the code fragment test set, performing test output by utilizing the trained full-connection neural network model to judge whether model input is a malicious code fragment; and performing feature extraction on the target code snippets, and inputting the target code snippets into a fully-connected neural network model generated through training and testing to obtain an intelligent malicious code recognition result of the targetcode snippets. According to the method, malicious code fragments in storage media such as computers, mobile phones and tablets and evidence containers such as RAW, E01 and AFF can be recognized, and the method has a good application prospect in the field of digital evidence collection such as crime event evidence underlying data automatic analysis.

An embodiment of the invention provides a human voice segmentation method and a human voice segmentation device. The human voice segmentation method comprises the steps of: extracting feature vectorsfrom audio data; performing voice activation monitoring on the audio data, and labeling muted segments and voice segments separately; extracting the voice segments according to labels, segmenting thevoice segments according to a predetermined time length, performing clustering operation on the feature vectors in the segmented voice segments by adopting a probability distribution clustering method, and outputting corresponding clustering labels; and arranging the voice segments corresponding to the different clustering labels according to a time sequence, and outputting the voice segments withdifferent clustering labels after arrangement and merging. The human voice segmentation method adopts the probability distribution clustering method for performing clustering operation, can perform rapid clustering on the feature vectors of voice without modeling the voice segments, adds the voice activation monitoring, only processes the voice segments, improves the working efficiency, and solves the technical problem of low efficiency and long time consumption of the traditional human voice segmentation system.

The invention provides an intelligent vehicle assisted driving method based on driver visual information. The intelligent vehicle assisted driving method combines a plurality of different informationacquisition methods to obtain external visual information, and organically divides various types of information from the perspective of driver information load, so that the obtainable information hasbetter hierarchy and interpretability. The collected and classified information is used, through a machine learning method, feature vectors and optimal time windows are selected, a feature parameter library based on levels and a driver information load amount classifier are separately established, an effective supporting method for selecting the timing and method of an assisted driving strategy isprovided, and human-computer interaction in intelligent vehicle assisted driving is improved.

An embodiment of the application discloses a data processing method for a demand side platform and a demand side platform. The data processing method for the demand side platform comprises the steps of receiving an information data set sent by a data system, and establishing a feature vector based on the information data set; determining at least an object data set corresponding to the feature vector based on a preset analysis model; generating at least one intermediate datum based on a first predetermined output model corresponding to each determined object data set; and outputting one of the generated intermediate data to the data system according to a second predetermined output model. The data processing method for the demand side platform and the demand side platform can be used for reducing the number of quotations received by an advertising trading system, and reducing the workload of the advertising trading system.

The invention discloses a financial abnormity detection method and device based on multi-label learning. The method comprises the steps: generating a feature vector sample of each enterprise accordingto the financial information of an existing enterprise, carrying out the sample balance according to a preset sampling parameter, and obtaining a training sample set; labeling the training sample setaccording to a plurality of labels obtained according to historical responsibility asking data, and constructing a financial abnormity detection model based on a multi-label learning algorithm; and finally, obtaining financial information of a to-be-detected enterprise, constructing a sample input vector, and inputting the sample input vector into the financial abnormity detection model to obtaina detection result. By adopting the technical scheme of the invention, the problem of low accuracy caused by few data samples and incapability of detecting financial whitewash means can be solved, and the detection accuracy is improved.

The invention provides a steel coil end face defect distinguishing method and system and electronic device. The method comprises: obtaining an adversarial network; training the adversarial network byusing a real steel coil end face image without defects; obtaining a steel coil end surface counter-resistance network; equally dividing the end face image of the steel coil to be measured into a plurality of first image blocks, extracting a first feature vector corresponding to each first image block, outputting a plurality of corresponding reconstructed images of each first image block based on the to-be-detected steel coil end face reactance network; performing feature extraction on each reconstructed image; obtaining a second feature vector, sequentially judging whether a first loss value between each first feature vector and the corresponding second feature vector exceeds a first threshold value or not. Therefore, the confrontation network only needs to train a defect-free steel coil end face image, feature extraction is carried out on the reconstructed image, whether the defect exists or not is judged through the loss value of the feature vector, the calculation amount is reduced,and the recognition efficiency is improved.

The invention discloses a face detection and recognition method. The method comprises the following steps: S1, preprocessing an image marked with a face frame to generate a training sample; S2, constructing a face detection and recognition network, wherein the face detection and recognition network adopts a deep learning network, and high-level features and low-level features of the network are fused; S3, inputting a training sample into the constructed face detection and recognition network for training until a training loss value is smaller than a preset threshold value, and completing training to obtain a deep learning network capable of outputting face detection and face recognition results. The face detection and recognition network is designed, face detection is regarded as a face center point problem, face center point detection and face feature vector extraction are combined for learning, a face feature vector corresponding to a face frame can be obtained while the face frame is obtained, the face feature vectors are compared to obtain a face recognition result, and a face detection result and a face recognition result are output through a network.

The invention discloses a GIS mechanical fault diagnosis method based on mechanical state signal analysis, and relates to the technical field of power equipment monitoring, and the method comprises the steps: collecting a vibration signal of GIS equipment in a running state; performing noise reduction processing on the acquired vibration signal by using a pole symmetry mode decomposition (ESMD) method and a relative entropy method; extracting a feature vector contained in the signal by using Hilbert transformation; performing attribute reduction on the feature vectors by using a rough set theory and establishing a diagnosis decision table; and establishing a three-layer forward neural network fault diagnosis model according to the diagnosis characteristic decision table. Compared with a traditional method, the method can be applied to intelligent diagnosis of different types of mechanical faults of the GIS device, and has the advantages of being good in real-time performance and high in diagnosis accuracy.

The invention provides a DenseNet-based CT image classification method and a DenseNet-based CT image classification device for COVID-19 patients, which are used for classifying according to computed tomography images of suspected COVID19 patients to obtain a classification result. The method includes the following steps of: storing medical image information; preprocessing by using a preprocessing method to obtain preprocessed data; obtaining deep information and shallow information from the preprocessed data by using the trained dense connection neural network model, and fusing the deep information and the shallow information to obtain a fused feature vector; mapping the fusion feature vector to a low-dimensional space by using an activation function to obtain a classification probability prediction value; obtaining a CAM activation graph based on the internal parameters of the densely connected neural network model and the computed tomography image; displaying a computed tomography image, a classification probability prediction value, and a CAM activation map to assist a doctor in diagnosis. The method and the device are suitable for an early screening stage of an epidemic situation area, the problem of too high false negative of accounting detection can be solved, and the diagnosis efficiency can be improved.

The embodiment of the invention provides a translation model training method and device and a translation method and device. The model training method comprises the following steps: respectively inputting a source language statement and a noisy source language statement in a parallel bilingual sentence pair into a translation model to obtain a first prediction target language statement and a second prediction target language statement, first prediction probability distribution and second prediction probability distribution of the translation model and/or first feature vectors and second feature vectors output by the hidden layers are obtained respectively; based on the first prediction target language statement and the target language statement in the parallel bilingual sentence pair, the second prediction target language statement and the target language statement corresponding to the noise-added source language statement, the first feature vector and the second feature vector and/or the first prediction probability distribution and the second prediction probability distribution; determining the current training loss of the translation model, and adjusting the parameters of the translation model. According to the embodiment of the invention, the robustness of the translation model can be improved, the training method is simple, and model training is stable.

The invention discloses a point cloud data segmentation method based on spectral clustering . The method comprises the following steps of reading an input point cloud data set and a clustering number;normalizing the coordinates of the point cloud data set to obtain normalized coordinates; constructing a similar matrix by normalizing the coordinates, calculating a Laplacian matrix Lrw, and calculating minimum m + 2 eigenvalues and eigenvectors f corresponding to the minimum m + 2 eigenvalues; standardizing a matrix formed by the feature vectors f according to rows to obtain a feature matrix F;taking each row in the feature matrix F as an m + 2-dimensional sample, obtaining m clusters from the samples by using a K-means clustering method, and outputting the m clusters; the method solves aproblem that a conventional point cloud segmentation result is not fine enough, maintains the invariance of the segmentation result for the translation, rotation and zooming of the point cloud data, and is also suitable for point cloud data or sparse data with non-uniform density.

The invention discloses a face feature extraction method based on a convolutional neural network, and the method is based on the construction of a convolutional neural network model. Comprising the following steps: step 1, proposing a convolutional neural network model; step 2, performing feature extraction on the processed face image through different convolutional layer faces in a convolutional neural network; and step 3, performing normalization processing on the feature vector output in the step 2, projecting the face sample from the gray scale space to the feature space, and performing classification in the feature space by using a classification method based on sparse representation. The face detection and classification method based on the convolutional neural network has the advantages of being high in face image feature extraction speed, high in accuracy, good in expressivity, simple in network structure and the like.

The invention belongs to the field of computer technology, in particular to a fatigue-driving detection method based on face recognition, a computer readable storage medium and a terminal device. Themethod comprises the steps of: collecting the face image of a driver; extracting a first sub-image from the face image of the driver, the first sub-image being an image of an area where the eyes ofthe driver are located; calculating a feature vector of the first sub-image; calculating a vector similarity between a feature vector of the first sub-image and a preset reference vector, the reference vector being a feature vector of an eye image in a fatigue state; if the vector similarity between the feature vector of the first sub-image and the reference vector is greater than a preset similarity threshold, determining that the driver is in a fatigue driving state. Through the embodiment of the invention, reliable detection standards are provided for fatigue driving detection of drivers, and the occurrence of traffic accidents can be greatly reduced.

The invention discloses an indoor positioning method based on four-dimensional code mapping, and the method comprises the steps: (S100), carrying out the Kronecker product of a horizontal-dimensionalfeature vector matrix and a vertical-dimensional feature vector matrix, generating a feature vector matrix of an antenna group, mapping the feature vector matrix into a four-dimensional code, carryingout the matching, obtaining an approximate code word four-dimensional code, and obtaining a pre-coding matrix; (S200), performing beam forming on the base station transmitting signal according to theprecoding matrix of the mobile terminal; and (S300), forming a virtual cell taking the mobile terminal by the to-be-positioned mobile terminal and the antenna group as the center; and forming a van virtual cell taking the mobile middle end as the center with surrounding mobile terminals, constructing an estimation function to obtain a positioning target function, solving the positioning target function by adopting a nonlinear least square algorithm, and taking the coordinate when the positioning target function takes the minimum value as the coordinate of the to-be-positioned mobile terminal.According to the method, beam matching of four-dimensional code mapping is combined with a virtual cell cooperation mode, and indoor real-time space positioning is achieved.

The invention relates to a face recognition method based on GCSLBP and DBN. According to the method, firstly, original Gabor transformation is improved, optimization is carried out by introducing a central symmetry local binary pattern method, and then a final feature vector is represented by using a histogram method, so that rich local features of the image are extracted, and the dimension of the feature vector can be reduced. And finally, the classification robustness is improved by using a deep belief network method, and face classification and recognition are completed. The method has a good effect in the aspects of face classification and recognition.

The invention discloses a multidimensional-similarity-based personalized news recommendation method. The method comprises the following steps of: extracting a set time record from a news log, capturing news and extracting a title and a text according to a news source address, performing word segmentation and noun extraction on the title and the text, and analyzing a noun sequence by using a subject model to acquire a subject feature character of the news; constructing a user model and a news model respectively according to the subject feature vector of the news and user behavior data; computing the content similarity and behavior similarity of users and the news respectively according to the user model, the news model and a time feature, computing final user similarity and final news similarity according to the content similarity and behavior similarity of the users and the news, and extracting a plurality of users and news which are most similar respectively; and generating user-based personalized recommendation results according to a latest news log record and a plurality of similar users which are most similar to a set user; or generating news-based personalized recommendation results according to the news on which the set user behaves and the news which is most similar to the news.