70results about How to "Increase feature dimension" patented technology

The invention discloses a multi-gesture and cross-age oriented face image authetication method. The method comprises the following steps of: rapidly detecting a face, performing key point positioning, performing face alignment, performing non-face area filtration, extracting the face features by blocks, performing feature dimension reduction and performing model prediction. The method provided by the invention can perform the face alignment, realize the automatic remediation for a multi-gesture face image, and improves the accuracy rate of the algorithm, furthermore, the feature extraction and dimension reduction modules provided by the invnetion have robustness for aging changes of the face, thus having high use value.

The invention relates to the field of ground object classification, in particular to a ground object classification method and device based on a multispectral image and an SAR image, and the method comprises the steps: obtaining the multispectral image of a preset area, and carrying out the multispectral image feature extraction of the multispectral image; obtaining a time sequence SAR image of apreset area, and performing time sequence SAR image feature extraction on the time sequence SAR image; and performing feature level fusion on the multispectral image features and the time sequence SARimage features to obtain a ground object classification result. According to the method and the device, the advantages of all-day working, all-weather working and short revisit period of the synthetic aperture radar SAR are utilized to obtain a long-time sequence SAR image, and the input feature dimension is increased; Characteristic level fusion is carried out on multispectral images and SAR images, and while multispectral information is fully utilized, ground feature interpretation is assisted by combining ground feature structures, textures and electromagnetic scattering characteristics reflected by time sequence SAR images.

The embodiment of the invention relates to the field of network security, and in particular relates to an evaluation method and device of a DNS. The accuracy of the DNS evaluation is improved through the method and device disclosed by the invention. The evaluation method comprises the following steps: determining M domain names in a DNS system and multiple feature dimension information of each type of the preset types corresponding to each domain name in M domain names according to a DNS traffic log, wherein M is an integer more than and equal to 1; determining association identification information associated with each domain name in M domain name, wherein the association identification information comprises an IP address and/or identity information; determining a cluster score of each type of the preset types of the DNS system according to the association relation between each domain name in the M domain names and the association identification information, or the attribute information of each domain name in M domain names; and determining the total system score of the DNS system according to the cluster score of each type of the preset types of the DNS system. Therefore, the accuracy of the DNS evaluation is improved.

The invention relates to the technical field of natural language processing, and provides a text semantic recognition method and device, computer equipment and a storage medium. The method comprises the steps of determining text characters contained in a target text and text segmented words to which each text character belongs; calculating character vectors corresponding to the text characters andword vectors corresponding to the text segmented words; splicing the word vector of each text character with the word vector of the text segmented word to which the text character belongs to obtain aspliced vector of the corresponding text character; inputting the word vector into a first neural network layer to obtain a first feature vector, and inputting the splicing vector into the first neural network layer to obtain a second feature vector; and inputting a comprehensive feature vector obtained by splicing the first feature vector and the second feature vector into a second neural network layer to obtain a semantic type of the target text. By adopting the method, the accuracy of text semantic recognition can be improved.

The invention discloses a microblog multimoding sentiment analysis method based on a microblog environment. The method is characterized by comprising the steps of obtaining microblog data, obtaining microblog text content and a first modeling text characteristic of commenting content, extracting a second modeling characteristic of the commenting content, combining the first modeling text characteristic and the second modeling text characteristic so as to obtain the final characteristic vector, conducting emotion mark treatment on microblog data so as to obtain a multimoding sentiment training model, and pre-testing the emotion state of the microblog text content to be tested. The method can be used for being further combined with the microblog commenting content for characteristic extraction of the microblog text content so as to improve the judging accuracy of microblog emotion states.

The invention relates to a radar radiation source individual recognition method and system based on Bayesian curve fitting. The method is characterized by comprising the following steps: firstly, extracting a radar pulse signal UPMOP characteristic, and then fitting the characteristic by a Bayesian curve, and finally, taking the control point of the fitted curve as an individual characteristic tobe input to a neural network classifier. Starting from the actual characteristics of high feature dimension and low recognition efficiency faced by the radar radiation source individual identification, UPMOP with high stability is selected as an individual feature, the fitting of UPMOP is realized by Bayesian curve, the extracted control point is extracted to serve as a new individual recognitionfeature vector, and the feature dimension is greatly reduced. The problems that in the prior art, modeling is difficult, fingerprint features are not easy to measure, and operation complexity is highare solved.

The invention discloses an aerospace engine abnormity intelligent detection method based on hierarchical adversarial training, and the method comprises the steps: employing a plurality of sensors to collect original signals of an aerospace engine in an operation state as multi-source data, intercepting a time sequence at a fixed length to obtain a multi-channel data sample set, and converting a one-dimensional sequence into a two-dimensional image; dividing the two-dimensional image sample into a training set and a test set; constructing a relative generative adversarial network as an anomalydetection model, and performing hierarchical adversarial training by using the training set; using the training model to evaluate the state of the training set sample, modeling the obtained evaluationscore distribution, and calculating the score threshold of the normal sample; using the model for evaluating the state of a test set, aggregating neighborhood information during testing, and conducting anomaly detection according to a score threshold value. According to the method, the model detection capability is improved through hierarchical adversarial training, multi-source information is fused, neighborhood information is aggregated to improve the result reliability, and finally, intelligent detection of abnormal operation of the aerospace engine can be realized.

The invention discloses a welding spot quality identification method fusing a knowledge graph and a graph convolutional neural network, and the method comprises the steps: photographing a welding spot, and obtaining a welding spot appearance image; the welding spot appearance image comprises a welding spot and a position visual feature of the welding spot; cutting the welding spot appearance image to obtain a welding spot cutting image; the sizes of all the welding spot cutting images are the same, and each welding spot cutting image only comprises one welding spot and the position feature of the welding spot; importing the cutting image of the welding spots into a fine-grained network for feature mining to obtain a visual feature matrix of the welding spots; establishing a knowledge graph according to the quality of the welding spots and the position relationship between the welding spots, and performing feature mining on the knowledge graph by using a graph convolutional neural network to obtain a high-dimensional spot type spatial feature matrix of the welding spots; and carrying out vector inner product on the visual feature matrix and the high-dimensional spot type spatial feature matrix to obtain a classification detection result of the welding spot quality.

The invention discloses a sleep state detecting method based on single-channel EEG signals. The method comprises the steps of collecting single-channel EEG signals in real time, performing downsampling processing on the single-channel EEG signals, and inputting the single-channel EEG signals obtained after downsampling processing to a classifier model to obtain the corresponding sleep state. The technical problems that according to an existing sleep state detecting method, by arranging multiple collecting electrodes, manpower is wasted, the extraction feature dimension is low, and the generalization is low due to over-fitting can be solved.

The invention discloses a non-reference high-dynamic range image objective quality evaluation method. The method comprises the following steps: representing an image as third-order tensor, performingtensor decomposition on a distorted high-dynamic range image by using a Tucker decomposition algorithm in the tensor decomposition since the chromaticity information has important effect in the high-dynamic range image quality evaluation, thereby obtaining a first channel integrated with the brightness distortion and the chromaticity distortion as a first feature image; extracting distortion information on the first feature image, wherein the first feature image further comprises the distortion of the chromaticity channel in comparison with the way of extracting the distortion information onlyon the brightness channel, the data size is same as that of the brightness channel, and the additional data size cannot be increased; and combining a tensor domain sensing feature vector extracted from the first feature image with a support vector regressive training model to obtain an objective quality evaluation value of the distorted high-dynamic range image. Therefore, the objective quality evaluation of the non-reference high-dynamic range image is realized, the evaluation effect is obviously improved without using the reference image.

The invention relates to the technical field of natural language processing, and provides a text semantic recognition method and device, computer equipment and a storage medium. The text semantic recognition method comprises the steps: calculating a character vector of a text character in a target text and a word vector of each text segmented word; splicing the word vector of each text character with the word vector of the text word segmentation to obtain a spliced vector of the text character; sequentially inputting the word vectors and the splicing vectors of the text characters into a firstneural network according to a forward appearance sequence of the text characters in the target text to obtain a first text feature; according to the reverse appearance sequence of the text charactersin the target text, sequentially inputting the word vectors and the splicing vectors corresponding to the text characters into a second neural network to obtain second text features; and inputting acomprehensive text feature obtained by splicing the first text feature and the second text feature into a third neural network to obtain a semantic type of the target text. By adopting the text semantic recognition method, the accuracy of text semantic recognition is improved.

The invention relates to a convolutional-neural-network learning method of multi-scale progressive accumulation. The method can be widely applied to the fields of machine vision and artificial intelligence such as target detection, target classification and target recognition. Firstly, average pooling operations are employed by the method to construct a multi-scale image pyramid on input images; and then images of different scales are progressively sent into a convolutional neural network, the convolutional neural network is enabled to carry out learning on the multiple images of the differentscales with progressive increasing of network depth and carry out progressive feature accumulation, and thus feature learning ability of the convolutional neural network is improved.

The invention provides a micro-expression feature extraction and recognition method based on deep learning, and the method comprises the following steps: constructing a four-layer pyramid optical flow model based on a key frame, inputting a micro-expression video key frame, and obtaining an optical flow feature; extracting LBP (Local Binary Pattern) features from three orthogonal planes of the micro-expression image sequence, and cascading normalized feature gradient histograms obtained in three dimensions into an LBP-TOP (Local Binary Pattern-TOP) histogram vector; the optical flow features are converted into histograms, cascade fusion is carried out on the histograms and LBP-TOP histogram vectors of the image sequences, and histogram representation of fusion features is obtained; constructing a shallow CNN model based on a residual module, introducing a macro expression data set and a micro expression data set, and constructing a cross-data-set micro expression recognition model based on a CNN-GCN transfer learning network; and inputting the fusion features into a trained cross-dataset micro-expression recognition model based on a CNN-GCN transfer learning network for classification to obtain a micro-expression type. According to the method, the micro-expression recognition precision is effectively improved.

The invention discloses an ADS-B weak signal detection device and method based on multi-feature fusion. According to the ADS-B weak signal detection method and device based on multi-feature fusion, the conventional mode that an existing signal detection system mainly takes energy features of signals as the basis and adopts threshold detection is broken through, multi-feature information containedin received signals are utilized to solve the problem that the detection performance is reduced due to weakening of the target signal state information under a strong noise interference condition.

The invention relates to a target tracking method based on an improved particle filter algorithm. The target tracking method includes the steps: reading a first frame image after acquiring a video stream; selecting tracked targets, and performing multi-dimensional feature extraction, establishing a target template, reading a next frame of image, performing target search by using the target template, performing weighted resampling on a searched target, performing interference detection, if interference exists, performing correction, otherwise, directly determining whether a target template updating condition is triggered; and if so, updating the target template and continuing target tracking. According to the target tracking method, the traditional particle filtering algorithm is improved from four aspects of increasing the feature dimension, reducing the particle attenuation, improving the algorithm immunity and introducing the target template updating strategy, and the target can be effectively tracked under the condition that the tracking environment is migrated and the interference of similar targets and shielding objects occurs, and the requirement of the real-time performanceof the system is met.

The invention discloses a credit score value determination method and device, electronic equipment and a storage medium, and relates to the field of machine learning, and the determination method comprises the steps: obtaining a plurality of account features of a target account based on an account identifier of the target account, obtaining a feature set, inputting the feature set to a pre-trained federal learning model, and obtaining a credit score value of the target account; the method comprises the steps of obtaining a parameter value of each account feature, performing weight coding on each account feature to obtain an evidence weight value of a feature box to which the account feature belongs, and determining a credit score value of a target account in a current scoring mechanism based on the parameter value of each account feature and the evidence weight value of the feature box to which the account feature belongs. According to the method and the device, the technical problem that the security of customer data privacy cannot be ensured in a mode of improving scoring accuracy by acquiring data of a plurality of institutions to jointly train a customer credit scoring model in related technologies is solved.

The invention discloses a power equipment fault detection model based on combination of an attention mechanism and a GRU, the power equipment fault detection model comprises a classification neural network model, and training data of the classification neural network model comes from a preprocessing model. The preprocessing model converts input unbalanced power equipment data into balanced data and performs embedded representation, and outputs intermediate data: a historical state sequence based on power equipment representation, label data embedded representation and embedded representation of power equipment portrait features; time and space features of the power equipment are extracted from the historical state sequence through a GRU module; state sequence features are extracted from the output of the GRU module through an attention mechanism module; environment information of the power equipment is extracted from the embedded representation of the portrait features of the power equipment through a graph attention mechanism module; and the state sequence features, the label data embedding representation and the environment information are aligned and fused to serve as training data input of the classification neural network.

The invention provides a packet level wireless device authentication method based on channel state information, and belongs to the technical field of wireless device security. The method comprises two stages of authentication during access and packet-by-packet authentication during communication: when wireless equipment requests access, authentication equipment enters an 'access' authentication mode; firstly collecting a CSI fingerprint of to-be-authenticated equipment, and then authenticating the equipment according to a standard deviation of a fingerprint set; if the authentication fails, refusing to access, and if the authentication succeeds, entering a communication authentication mode, and training a local authenticator by using a CSI fingerprint set; whenever a new data packet is received, enabling the authentication equipment to extract the CSI from the new data packet and input the CSI to the local authenticator for fingerprint matching; and if matching succeeds, accepting the data packet, otherwise, discarding the data packet. The invention provides a non-cryptographic authentication method for an access stage and a communication stage, the non-cryptographic authentication method is generally used for a wireless network based on an OFDM (Orthogonal Frequency Division Multiplexing) technology, and meanwhile, an improved integrated authenticator has high authentication time efficiency and accuracy and can be used for packet-level authentication.

The invention relates to the technical field of photovoltaic system ultra-short-term power prediction. In order to solve the problem of low accuracy of ultra-short-term prediction of a photovoltaic system in case of changeable weather, the invention provides a double-layer collaborative real-time correction photovoltaic prediction method, which comprises the following steps of: obtaining a reference layer photovoltaic prediction set in the future N hours by utilizing a reference layer photovoltaic prediction model F1; in combination with the photovoltaic prediction error of the reference layer, utilizing a real-time layer photovoltaic prediction model F2 to correct prediction values in the reference layer photovoltaic prediction set one by one, so that a final real-time layer photovoltaicprediction set can be obtained; according to a photovoltaic fluctuation rule, considering an optimal time sequence translation characteristic, and obtaining a final photovoltaic power prediction valueafter a photovoltaic prediction result of a reference layer is corrected, so that the influence of a process weather factor on photovoltaic prediction can be weakened, and the photovoltaic predictionprecision is improved.