235results about How to "Enhanced description ability" patented technology

The invention discloses a pronunciation quality assessment and error detection method based on the fusion of multiple characteristics and multiple systems, which carries out assessment and error detection on pronunciation quality by a method utilizing multiple characteristic parameters to describe pronunciation quality and utilizing multiple inspecting systems to mutually fuse, and comprises the following steps: recognizing voice and automatically segmenting and aligning the voice; extracting the characteristic parameters used for voice quality assessment and error detection; acquiring pronunciation quality assessment and error detection model training data; training a plurality of pronunciation quality assessment and error detection systems; fusing a plurality of pronunciation quality assessment and error detection systems; and assessing pronunciation quality and detecting pronunciation errors. By utilizing the invention, multiple voice characteristics are effectively utilized, and multiple assessment and detection system are fully utilized and perform information fusion, thereby maximally exerting the advantages of various characteristics and systems, and ensuring the accuracy and reliability of pronunciation assessment and error detection.

The invention relates to a face feature extraction method based on a face feature point shape drive depth model. The method comprises the steps that the face feature point shape drive depth model is set up, N depth convolution neural networks are utilized for extracting features of N face regions divided according to the positions of face feature points to obtain the discrimination feature and the attributive feature of each region, and then all the discrimination features and the attributive features are fused to obtain features higher in descriptive ability. According to the face feature extraction method based on the face feature point shape drive depth model, the problem of robustness under change conditions of illumination, angles, expressions, shielding and the like can be well solved, and the recognition rate of face recognition under these conditions is increased.

The invention belongs to the crossing field of computer vision and intelligent robot, and discloses a new multi-feature multi-sensor method for a mobile robot to track a moving body. The method comprises the following steps: 1, coarsely positioning a body carrying a passive tag around a radio frequency identification (RFID) system by using the RFID system; 2, initially positioning the body in an image by using an adaptive template matching algorithm based on head and shoulder features; 3, accurately positioning the body in the image by using a multi-feature-based mean-shift algorithm; 4, predicting the moving state of the body by using a extended Kalman filter algorithm; 5, screening the acquired target position information by using a double-layer collaboration positioning mechanism; and 6, controlling the robot to move along with the body by using a robot following control algorithm. By the method, bodies with different poses can be tracked, the problem that the tracking is influenced when a target suddenly turns and is shielded is solved, and the robot can accurately, stably and continuously track the moving body.

The invention discloses a video abstraction generating method based on sketches, which belongs to the field of man-machine interaction. The method comprises the following steps: (1) carrying out shot segmentation and shot screening on the video to obtain key frames; (2) carrying out character analysis and scene analysis on the key frames, and determining the semantic significance level of each key frame according to the character analysis; (3) calculating an image quality assessed value of each key frame, and screening the key frames according to the image quality assessed value and the semantic significance level value; (4) generating a video object sketch according to each key frame determined in step (3); (5) resetting the size of the corresponding sketch according to the size of the abstraction drawing region and the significance level of the video contents corresponding to each sketch; and (6) determining the position of each sketch by utilizing the scene analysis result, generating connecting lines among the sketches, and connecting the sketches to generate a video abstraction. By using the method of the invention, the sketch abstraction can intensively and effectively reflect the main plot semanteme of the video, thereby conforming to the perception habits of users.

The invention discloses an image retrieval method and an image retrieval device and relates to the field of image search. The method comprises the steps of extracting global characteristics of a picture and then carrying out dimension reduction; quantifying the characteristics into binary bit strings; and retrieving similar pictures from a database according to a distance of the binary bit strings. According to the image retrieval method and the image retrieval device, the description for the picture is accurate by use of the global characteristics, the data size of picture characteristics can be decreased through dimension reduction for the global characteristics, meanwhile as a dimension reduction model can allow the distance of the global characteristics, subjected to dimension reduction, of a picture to be decreased and the distance of the global characteristics subjected to dimension reduction of dissimilar pictures increases, the descriptive power is improved after the global characteristics of the picture are subjected to dimension reduction; furthermore, the picture characteristics are quantified into the the binary bit strings, the data complexity can be reduced, the data storage volume is decreased, the matching complexity during retrieval is reduced, the retrieval efficiency is improved, and the image retrieval processing power is enhanced to hundreds of millions of magnitude orders.

The application discloses a test question automatic generation system (100) and a test question automatic generation method based on project models. The system can enable each project model to be analyzed into a plurality of different test questions, and uses a main body to mark and inquire the project models and the constraints for forming a test paper; the system can analyze of the discriminability of each project model and provide feedback information from various aspects based on the results measured of the tester.

The invention discloses a video description method based on multi-feature fusion. The method is characterized in that (1) deep spatial-temporal features of a video are extracted by fusing traditional CNN features and SIFT flow features; (2) an S2VT sentence generation model with added average pooling features being video overall features is adopted to generate corresponding sentence description according to the deep spatial-temporal features extracted in the step (1); and (3) a word2vec word vector method is adopted to replace one-hot vector word representation to optimize the sentence generation model in the step (2). The method has the advantages that more robust spatial-temporal features can be better extracted through multi-feature fusion; meanwhile, the average pooling features are added into the sentence generation model, so that more relations are established between visual information and words; finally, the word2vec word vector method is adopted to replace one-hot vector word representation, so that more relations are established between words, and video description performance is effectively improved.

The invention discloses a night pedestrian detection method based on statistical features of infrared pedestrian brightness. According to the method, gray average information of parts of one pedestrian and a negative sample in a sample database is firstly subjected to statistical processing, a mapping interval boundary is determined by the aid of the information, and a brightness histogram feature for distinguishing vote interval division is constructed; then, a histogram feature in the gradient direction is calculated, and the two features are combined to form a final feature descriptor; secondarily, model training is performed by the aid of Adaboost in combination of a decision tree, and pedestrian determination and positioning are performed with a sliding window scanning method; finally, when a classifier obtains lower degree of confidence through classification judgment on a certain detection frame, a brightness interval template is adopted for secondary judgment, so that night pedestrian detection is realized. Pedestrians in the night environment are effectively detected, and the method has the characteristics of high detection rate and adaptability.

The invention provides a face recognition method and face recognition equipment. The method comprises the following steps of: performing down-sampling on an original face image to obtain down-sampled images with different sizes; blocking the down-sampled images and the original face image; performing characteristic extraction on each image block obtained by blocking; performing similarity matching on characteristics of each extracted image block and characteristics of the corresponding image block of a pre-registered face image to obtain the similarity of each image block; and obtaining a recognition result of the face image according to the obtained similarities of all the image blocks. According to the face recognition method and the face recognition equipment, the original face image is subjected to down-sampling to obtain the down-sampled images with different sizes for multi-size face image characteristic extraction, so that the capability of the characteristics of the face image in the description of the image face is improved; and the down-sampled images and the original face image are blocked, and the recognition result of the face image is obtained according to the obtained similarities of all the image blocks, so that the recognition accuracy of the face image is improved.

The invention discloses a method and a system for detecting network security. The method comprises the following steps: 1) according to the initial state of equipment in a network, connection relationships among different pieces of equipment and the vulnerability information of the equipment, generating an attack graph comprising an attack node and a state node; 2) by aiming at a set target node, converting the attack graph into a stochastic Petri net model; 3) introducing the strategy and utility information of an attack behavior on the stochastic Petri net model, generating the stochastic game net model of an attack visual angle, and introducing the strategy and utility information of a defensive behavior on the stochastic Petri net model to generate the stochastic game net model of a defensive visual angle; 4) combining the stochastic game net model of the attack visual angle with the stochastic game net model of the defensive visual angle to generate an attack-defense game strategy model; and 5) detecting network security by the attack-defense game strategy model. According to the method and the system, the accuracy for detecting the network security can be improved.

A web service modeling and the test method based on an interface semantic contract model relate to the fields of software test modeling and automatic test case generation. The web service test method comprises (1) building a service interface semantic contract model to expand descriptive capability of an existing service interface; (2) generating test input data and a test case according to description of the interface data ontology to provide input data for the test case; and (3) providing an expected output result for the test case according to regular test data. The web service modeling and the test method based on the interface semantic contract model have the advantages of being applicable to a web service test, capable of expanding the descriptive capability of the existing service interface, capable of being used for generating the test data and the test case and improving the degree of automation and intelligentization of test generation through knowledge descriptions and reasoning technologies.

The invention discloses a multilayer bitmap color feature-based image retrieval method. In the method, fast clustering is performed on an image with rich color information to obtain rational statistical distribution centers of each color cluster, and based on the rational statistical distribution centers, features capable of reflecting color differences among different distribution layers of the image are extracted to perform image retrieval. The method comprises the following steps of: first performing meshing on a color space of the queried image, counting the numbers of pixel points in each mesh and selecting the mesh with a number local maximum; then quickly generating each color cluster and the rational statistical distribution centers thereof by adopting a novel distance optimization algorithm and an equal-average nearest neighbor algorithm search (ENNS) algorithm in a K-average clustering algorithm, and on the other hand, performing space sub-block division on the queried image and calculating a Gaussian-weighted color average of sub-blocks; next comparing the color average of the image sub-blocks with the rational statistical distribution centers of the color clusters to extract the features of a K-layer bitmap; and finally performing the matched searching of the image features by combining the similarity measurements of the rational statistical distribution centers of the color clusters and the bitmap.

The invention discloses a texture image classification method based on a BoF (bag of feature) and multi-feature fusion. The method comprises: performing local image selection on a texture image to form a fragment set; extracting GGCM (gray level-gradient co-occurrence matrix) features and SIFT (scale invariant feature transform) local features of all fragments, and performing importance weighted fusion on different features; generating a feature word for fusion feature clusters and performing preference and weighting on the words by using DWDPA (dynamic weighted discrimination power analysis), and assigning a fusion feature vector by using the preference and weighted word to form a training set fusion feature word bag model; computing the fusion feature vector of the texture image to be tested and acquiring a corresponding fusion feature word bag; and training the feature word bag model by using a SVM (support vector machine) as a classifier. The method effectively overcomes a defect that the GGCM is low in accuracy for large texture classification, compensates a weakness of information loss of the BoF feature space, and is more accurate and good in robustness.

The invention discloses a robust learning model and an image classification system. The robust learning model comprises the steps that a training set is initialized so that an initial category tag matrix is obtained, and training samples in the training set include the samples of which the categories are known with calibration of category tags corresponding to the categories and the samples of which the categories are unknown without calibration of the category tags; the training samples are processed by a construction method based on neighboring definition and reconstruction weights, a reconstruction coefficient matrix is constructed according to similarity between the samples, and symmetrization and normalization processing is performed; soft tags without calibration samples are determined by utilizing the reconstruction coefficient matrix and the initial category tag matrix, and l2,1 normal regularization is performed on the soft tags of the training samples by adopting an iteration method so that a projection matrix and a soft tag matrix are obtained; mapping is performed on samples under test by utilizing the projection matrix so that the soft tags of the samples are obtained; and the samples under test are the samples of which the categories are unknown without calibration of the categories. Influence of mixed signals in an original space can be effectively reduced by the model so that classification accuracy can be enhanced.

The invention discloses a seismic data static correction method and system. The seismic data static correction method includes the steps that the first arrival time is obtained according to seismic data of an experiment area, refraction layering is carried out in a CMP gather of the experiment area, and the refraction speed and the delay time of each shot point or each receiver are obtained through calculation; an initial model of the experiment area is built, and an underground speed space model is obtained according to first arrival time tomography inversion; certain part of shot points or receivers are selected for micro-logging investigation, high-speed layer burial depth data of the certain part of shot points or receivers are obtained, depth calibration is carried out on the underground speed space model, and the corresponding tomography inversion stratum speed is obtained; the space interpolation is carried out on other shot points or receivers, the corresponding tomography inversion stratum speeds of the shot points or receivers at the high-speed layer burial depth positions are obtained, and high-speed layer burial depth data of the shot points or receivers are extracted point by point; the static correction amount of the shot points or receivers are calculated, and static correction is carried out on seismic data through the static correction amount.

The invention discloses a self-adaptive weight selected incremental characteristic background modeling method. The invention can conduct incremental real-time updating of a background model according to the movement contained in each video frame in the movement detection, and appoint a weight for every frame in the updating process to improve the expression and description abilities of the background model. The method comprises the following steps of: roughly detecting the movement area of the present frame by an un-updated background model, constructing a weight for the movement area based on the reconstruction error of the model, updating the background model by right of the incremental principal component analysis based on the weighted present frame and generating background images. With well expressing the dynamic changes of the complex scenes and being sensitive to the objects with obvious movement prospect, the inventive technique has great application value in the fields of video surveillance, etc.

The invention discloses a nearest neighbor subspace SAR target identification method based on multiple sparse descriptions and aims at mainly solving the problem that the accuracy is low during target identification and the identification effect on locally-changing targets are poor in the prior art. The nearest neighbor subspace SAR target identification method comprises the steps of 1 performing preprocessing to obtain training samples and normalized sub-images of test sample images; 2 establishing dictionary matrixes to obtain multiple dictionary matrixes identical to categories in number; 3 calculating sparse vectors; 4 calculating reconstruction errors; 5 confirming identification results, utilizing a nearest neighbor subspace formula to use a target category corresponding to a reconstruction error minimum value as an identification result. Compared with the prior art, the identification accuracy of the locally-changing targets is improved. The description capacity and the identification rate of test sample detail features are improved.

The invention is suitable for the technical field of pattern recognition, and provides a human movement recognition method and device. The method comprises the steps: obtaining a depth image sequence, carrying out the conversion of the depth image sequence, and obtaining a corresponding depth movement sequence; carrying out the dividing of the depth movement sequence in the time dimension and space dimension, and obtaining a plurality of movement historical cubes and a plurality of corresponding space cube subblocks; calculating corresponding characteristic vectors corresponding to the movement historical cubes according to the space cube subblocks, and obtaining the characteristic vector of the depth movement sequence through combining the characteristic vectors corresponding to the plurality of movement historical cubes; and carrying out the model training and testing through employing an SVM (support virtual machine) according to the characteristic vector of the depth movement sequence, so as to obtain a recognition result of human movement. The method solves a problem of self-sheltering in a conventional human movement recognition method, improves the description capability for human movement, and improves the recognition accuracy of human movement.

The invention discloses a software run-time property monitoring method based on property specification mode, aiming at providing a software run-time property monitoring method beneficial to fault discovery and diagnosis. The technical scheme includes that software run-time property monitoring requirements are classified into multiple property specification modes, a property specification template is constructed for each property specification mode; meta-information data in software to be monitored is extracted, so as to generate a meta-information file; the property specification template is configured, so as to generate a monitoring script file of the software to be monitored; the monitoring script file is analyzed, so as to generate a monitoring code; the software to be monitored and a comprehensive monitoring file are woven, so as to generate new software with run-time property monitoring capability; and the new software generated by weaving is run, software property is monitored and monitoring result is displayed. By adopting the invention, support can be provided for software fault discovery, diagnosis and defect location, software failure and maintenance cost is reduced, monitoring workload and monitoring difficulty are reduced, and timeliness is improved.

The invention relates to a natural interactive method of a virtual learning environment based on speech emotion recognition, belonging to the field of depth learning. The method comprises the following steps: 1, collecting speech signals of students and users through kinect, resampling, adding windows by frames, and mute processing to obtain short-time single frame signals; 2, carrying out fast Fourier transform on that signal to obtain the frequency domain data, obtaining the pow spectrum thereof, and adopting a Mel filter bank to obtain a Mel spectrum diagram; 3, inputting the features of the Mel spectrum map into a convolution neural network, performing convolution operation and pooling operation, and inputting the matrix vectors of the last desample layer to the whole connecting layerto form a vector output feature; 4, compressing and inputting the output characteristic into a bi-directional long-short time memory neural network; 5, inputting the output features into a support vector machine to classify and output a classification result; 6, feeding back the classification result to the virtual learning system for virtual learning environment interaction. The invention driveslearners to adjust the learning state and enhances the practicability of the virtual learning environment.

The invention discloses a text multi-granularity similarity comparison method based on semantic aggregation fingerprints. The method comprises the following steps: training word vector representation;extracting semantic features; performing multi-feature aggregation; constructing a hierarchical index; calculating similarity. According to the method, word vector representation modeling is carriedout in combination with multi-dimensional semantic correlation; semantic information among words is fully mined; characteristics are extracted by taking sentences as units, semantic features are represented by adopting multiple weights, text library statistics and distribution information are mined by utilizing a statistical learning method, finer division of a feature space is realized, a compacttext fingerprint with high identification degree is generated on the basis of multi-feature aggregation, and the description capability and the discrimination degree of the text fingerprint are effectively improved. According to the method, text similarity comparison is carried out by adopting a top-down thought and using semantic aggregation fingerprint and local semantic features, and global-to-local multi-granularity similarity comparison of texts can be quickly and efficiently realized by constructing hierarchical indexes; the method has good expandability.

The invention discloses a voice emotion recognition model and method based on joint feature representation, and relates to voice emotion recognition technology. A convolutional recurrent neural network model is improved, a hidden layer in the neural network is configured to learn the joint feature representation of a spectral depth feature and a manual feature, and the joint feature extraction andsentiment classification are integrated into an end-to-end network model. The joint feature utilizes the complementarity between the spectral depth feature and the manual feature, makes full use of the emotional information carried in the voice, and more perfectly models the voice emotion. In addition, the end-to-end network model reduces parameter redundancy due to an intermediate output layer.The voice emotion recognition method based on joint feature representation improves the recognition accuracy of the voice emotions compared with an original voice emotion recognition method based on apure convolutional recurrent neural network.

The invention provides a micro expression automatic identification method based on multiple-dimensioned sampling. The method comprises micro expression image sequence preprocessing, micro expression feature extraction and micro expression identification. For the purpose of reducing influences exerted by face natural displacement and an ineffective area on micro expression identification, a method for automatically aligning a face and effectively partitioning a face area, and thus the robustness of an identification result is improved; and for the purpose of solving the defect of an existing feature descriptor, the invention brings forward a novel micro expression feature description operator CPTOP, the CPTOP operator the same sampling point number as a LBP-TOP operator yet employs a multiple-dimensioned sampling strategy, and under the same time complexity and space complexity, better description information is obtained.

The invention discloses an automatic identification method and system for a vehicle logo. The automatic identification system for the vehicle logo comprises an offline training subsystem and an online identification subsystem. Features of the vehicle logo are extracted by adopting dense scale invariant feature transform (dense-SIFT), and abundant and stable features which are identifiable and discriminative can be extracted from a low-resolution vehicle logo picture; clustering analysis is performed on the dense-SIFT features, and a visual word bank and a K-dimension (KD) tree index are established so as to accelerate computation speed; according to the relevance between the dense-SIFT and visual words, the dense-SIFT is mapped into all visual words for representation to enhance feature descriptiveness; when a visual word histogram is established, a rectangular-ambulatory-plane spatial pyramid is proposed to increase the spatial position information of the features ,and thus, discriminating effect is promoted; and by adopting a support vector machine (SVM) training vehicle logo classifier, vehicle logo identification is realized. The automatic identification system for the vehicle logo disclosed by the invention still has high identification rate on the low-resolution picture, has high robustness on blocking, illumination, weather and shooting angle changes and the like, is short in computation time and has real-time performance.

The invention discloses a multi-characteristic matching multi-target tracking method based on Hough forest. A conserved and reliable track fragment is obtained through double-threshold correlation. A positive and negative sample set is generated in an online manner according to a sample selecting principle. The Hough forest is constructed. Through Hough forest learning, training samples with color, shape, class and motion information are divided to different leaf nodes. Statistics information of the leaf node is used for forecasting an association probability of two track fragments. When a reliable long-track fragment is obtained, the reliable long-track fragment is converted to a re-matching problem between the tracks. Two manners of similarity measuring and characteristic point matching are used. The reliable long-track fragment is associated to a real track through the association probability, thereby finishing matching. The multi-characteristic matching multi-target tracking method has advantages of settling problems of error accumulation and low tracking precision, improving capability for processing target shielding and deformation, and realizing multi-target tracking in a complicated scene.

The invention discloses a gradient binaryzation based rotation-invariant and scale-invariant scene matching method, and relates to the field of scene recognition. According to the method, on the basis of a classical binary description BRIEF algorithm in which only gray scale intensity is compared, horizontal and vertical gradient comparison is added, texture information of a sampled area is saved, and accordingly, matching error rate is reduced. Moreover, an image scale pyramid is created, image feature point detection and feature description are performed within different scales, gravity center vector directions are added during descriptor calculation, and direction and scale invariance of binary descriptors is achieved. Experiments show that binaryzation based rotation-invariant gradient sampling descriptors have high robustness, and the matching accuracy rate is 73.06% higher than that of the BRIEF algorithm in average when a scene image is rotated greatly and the scale is varied.

The invention discloses a method for detecting changes of a remarkable target of a remote sensing image. The method comprises the steps that a remarkable target area is extracted from a reference image; the reference image and an input image are sampled, and sampling points are used for approximately representing the remarkable target area; DAISY characters of the sampling points are extracted; a plurality of candidate matching points are searched in a sampling point set of the input image for the sampling points of the reference image; an optimal matching point is searched in the corresponding candidate matching point set for a sampling point set of the remarkable target area, and the distance between optimal matching point sets serves as the change feature of the corresponding remarkable target area; whether an area, corresponding to the remarkable target area, in the input image changes or not is determined. By means of the method, a large amount of redundant information is filtered, practicability of detection on the changes of the remote sensing image is improved, description capacity and robustness for view angle conversion and registering errors of the area are improved, and inter-class divisibility of the change class and the non-change class is improved. The method can be widely applied to various fields such as disaster monitoring and target reconnaissance.

The invention discloses a text clustering method based on weak supervised deep learning. The method comprises the following steps: (1) by means of an image data set with text click information, imagevisual information and image category labels are utilized, and adopting image amplification and clustering to construct an image category click characteristic matrix of each text; And (2) obtaining asmooth image click feature map on the initial class click matrix by using a sorting and propagation method. Performing text clustering on the feature map to obtain an initial text category, and initializing text weight by utilizing click priori; (3) under the condition of minimizing an intra-class mean square error, building a deep text clustering model to learn deep text characteristics; (4) performing joint optimization on the depth model and the text weight by using a weak supervised learning method, and iteratively updating the depth model and the text weight; (5) deep text features are extracted through the deep text model, and K-based text feature extraction is achieved. And clustering the means method. The method has very high universality, and the semantic gap in image recognitionis effectively solved.

The invention discloses a context feature fused aspect-level sentiment classification method and device. The method comprises the steps of carrying out weak correlation aspect word static shielding and word segmentation processing on the aspect-level sentiment analysis text to be predicted; obtaining a first global context, the first BERT embedded layer and the second BERT embedded layer respectively extract a first global context feature and a first local context feature, extracting a mixed local context feature by the local feature learning layer, and extracting a second global context feature by the MHSA layer; processing a fusion feature obtained after the second global context feature and the mixed local context feature are fused by the interactive learning layer; and finally, outputting an emotion polarity result by the output layer, so that the technical problem of low accuracy of emotion polarity prediction due to neglect of local contexts of aspect words, strong semantic association characteristics of the aspect words and interference of weakly related aspect words on emotion classification in the prior art is solved.

The invention discloses a sparse-representation-LBP-and-HOG-integration-based pedestrian detection method. An LBP characteristic is extracted and sparse representation is carried out; and a sparse coefficient and an HOG feature are integrated. An experiment result demonstrates that the identification rate is effectively improved and the robustness is high on the complicated illumination condition when the method is used. Compared with the existing method using characteristic integration for improving an identification rate, the provided method has advantages of low characteristic dimension and fast identification speed and the like.